AMD and TSMC have enjoyed a very close working relationship ever since the former spun off its chip-making facilities into a separate business in 2008. Its latest Ryzen desktop processors, which comprise two or three chiplets, are made entirely by TSMC, but if one report is to be believed, future CPUs could well be made by the Taiwanese firm's biggest rival: Samsung.

That's according to Nikkei Asia (via Jukan on X), which claims that "AMD…is in talks to build some future CPUs with Samsung from 2028." Although the report says this snippet comes from "one source with knowledge of the situation," the publication explains that the discussions are due to capacity constraints.

As things currently stand, AMD employs various TSMC node options for its products. The Ryzen 7 9800X3D, for example, has two chiplets under its heatspreader: a Core Complex Die (CCD) that's made on N4P and an Input/Output Die (IOD), which is fabricated on the cheaper N6 node.

The former node is heavily used by AMD, with all of its Ryzen AI 300 and 400-series laptop APUs also being fabricated on N4P, as well as its latest Radeon GPUs, but it also uses TSMC's N3 process to make some of its Zen 5c chips.

What we do know for sure is that AMD will be using TSMC's cutting-edge N2 process node for its next generation of Epyc server-grade processors, and while this is no guarantee that all Zen 6 chips will spring from these wafers, Nikkei Asia's remark suggests that this could well be the case.

Given its higher cost and relative newness, TSMC's N2 will probably be more restricted in capacity than N4, for example, and rather than limit its supply of consumer-grade processors to meet the demand of the AI market, it would seem that AMD is exploring the possibility of using the next best thing, i.e. one of Samsung's process nodes.

The question to consider here is just what processors the South Korean chip giant would make. To my mind, there are two main contenders: low-end APUs and the IOD for Zen 6 processors.

Budget-level CPUs are often just rebadged older products, occasionally with a mild architectural tweak or two, so they don't need to be made on the very smallest and most expensive of process nodes. AMD's IODs house a lot of analogue circuitry (to handle the interchiplet Infinity Fabric, DDR5 memory, PCIe, USB, etc), which doesn't scale particularly well with process node shrinks.

In other words, both are suitable to be used on a 'bigger', and thus cheaper, process node, and Samsung's 4LPP system would seem to be an ideal candidate. But why not just use TSMC's N4? After all, once Zen 6 launches in force, surely the demand for that process node will reduce in favour of N2?

Well, possibly not, because Nvidia uses it for all its Blackwell AI chips and while it wants everyone and their grandparents to be using its Vera Rubin products, the current generation of megachips will still be in high demand by the time AMD releases its next series of processors.

However, the talks might not be about older process nodes at all, as rumours were floating around last year that AMD might use Samsung's 2 nm system, in conjunction with TSMC's similarly named node, for its future chips. We could possibly see two 'types' of Zen 6 CCDs: cutting-edge TSMC ones for Epyc processors, cheaper and slightly slower Samsung ones for Ryzen CPUs.

Then again, it might not actually be about wafer capacity and instead just be a simple matter of cost. With the global memory crisis causing DRAM and NAND flash prices to rocket through the stratosphere, the PC industry is somewhat in decline at the moment, with a reduction in shipments across every sector bar data centers (which is very much doing the opposite).

AMD and all its rivals will be considering every possible option to reduce the cost of making processors, and if Samsung is broadcasting that it can significantly undercut TSMC, you can certainly see why chip firms would be lining up to discuss using their services.

For PC gamers, none of this really matters, of course. Regardless of how potent or poor the next generations of CPUs or GPUs turn out to be, they're not going to be cheap enough to offset the fact that DDR5 memory kits and SSDs are now between three and five times more expensive than they were this time last year.

Processors seem to be one product category, along with displays, where there are glimmers of hope for us gamers amidst the RAMpocalypse. From an Nvidia chip with good gaming performance to new handheld chips from Intel and even the promise of some Intel-Nvidia collab chips that could also be great for handhelds, there are definitely things to be excited about. And now we can add chips made on Intel 18A-P to that list.

Going by the Intel presentation screenshots shared by Hardwareluxx, Intel's next version of its 18A process, 18A-P, could deliver a nice chunk of extra performance—or power saving, depending on which way you go—essentially for free. Free, that is, in the sense that it's "backwards compatible to Intel 18A", meaning no need for new fabs, machinery, or even chip designs.

The process is now in "risk production", which is essentially the limited production stage that occurs before mass production. It should be at least a few more months before mass production begins in earnest.

Intel's slides refer to 18A-P as "the first performance enhancement in the Intel 18A family" and describe the following highlights:

• "18% lower power at iso-performance
• 9% fully routed block level performance at iso-power
• 20–40% improved thermal resistance
• 10–30% improved Via resistance at perf critical layers"

In other words, compared to the same chips on 18A, chips on 18A-P should deliver 9% improved CPU performance for the same power, or 18% less power consumption for the same performance. Plus improved thermal resistance and "Via resistance", meaning less performance impact from backside power delivery.

Intel's 18A process is used to make Panther Lake chips for laptops, and we've already been pretty impressed with those—thus why we're excited for new Arc G3-series chips for handhelds, which are Panther Lake-based.

It's unlikely 18A-P will be used for these mobile chips given they're already in full swing, but they could be used for future product lines. Perhaps a Panther Lake refresh—the same chips but with a slight bump to performance sounds nice. Or more reasonably, perhaps Intel's upcoming Nova Lake desktop chips, as it would make sense for the more power-hungry CPUs to use the new and more efficient process.

There's a chance it could all be timed well for that upcoming Intel-Nvidia collab, too. But let's not get ahead of ourselves.

We might end up having to wait until 2028 for new AMD graphics cards. But the company has said its next-gen Zen 6 CPUs are coming later this year. Even better, the latest rumours claim the chips will clock up beyond 6.5 GHz, setting a record for factory-spec PC processors.

The source here is the Moore's Law is Dead (MLID) YouTube channel, which doesn't have a perfect track record, but it certainly has been first with leaks that have subsequently proven accurate.

The channel is claiming the clock speed part of its information on Zen 6 is "very high confidence" and "100%" confirmed. And the specific claim is that Zen 6 will "clock above 6.5 GHz."

If true, that will be well above the 6.2 GHz boost clock achieved by the Intel Core i9-14900KS, thus far the fastest PC processor in terms of raw operating frequency. For the record, AMD's current fastest clocking chips are the Ryzen 9 9950X and Ryzen 9 9950X3D, which both hit 5.7 GHz.

MLID says this is possible in part thanks to the use of cutting-edge TSMC N2 silicon. This bit is certainly very plausible. AMD is on record saying that its upcoming next-gen EPYC server CPUs based on the Zen 6 architecture will be built on TSMC N2.

It doesn't automatically follow that AMD will also manufacture CPU dies for consumer desktop chips. But it does make the idea very credible.

If it turns out Zen 6 desktop CPUs are indeed built on N2, it will mean AMD has effectively jumped a full node, leapfrogging N3 and going straight from an N5-class node to N2.

I say "N5-class" because the N4 node AMD uses for Zen 4 CPUs, including the AMD Ryzen 7 9800X3D, the current ultimate weapon of choice for PC gaming, has CPU dies built on TSMC N4, which is really just a minor variation of N5, not a true node class of its own.

However you slice it, the jump to N2, thus skipping N3, is both unusual and pretty exciting. Apart from clock speed, N2 ought to allow for a much larger transistor budget and hence on the one hand larger, more complex and more performant CPUs and, on the other, more of those new cores.

With that in mind, MLID says the Zen 6 desktop CPU die will contain 12 cores. With the top Ryzen CPUs traditionally containing two CPU dies, that adds up to 24 cores, a big step up from the 16 cores at which desktop Ryzen CPUs have maxed out for several generations now.

It's not currently known what kind of IPC or instructions per clock boost Zen 6 will offer. But there's certainly the potential for a substantial per-core performance boost given the rumoured boosted clock speed. Add in a big jump in core count, and the overall multi-threaded performance uplift could be something very special indeed.

And as I said, with Zen 6, the one thing we know for sure is that there's not long to wait. It's coming later this year.

For the launch of its Core Ultra 200S Plus processors, Intel released a software package called Binary Optimization Tool (IBOT for short), which promised to give certain games a substantial boost in performance. The chip giant also promised that it would be rolling out updates when possible, and true to its word, the first one has just landed.

As explained in a short article about the update, IBOT "analyzes how a game’s machine code behaves on Intel hardware and intelligently redirects certain workloads through optimized execution paths." It's only available for Core Ultra 200S Plus desktop and mobile chips, plus the Core Ultra Series 3 range (Panther Lake), so only a relatively small number of Intel CPU owners can use it.

But for those who can, you might be happy to know that Hollow Knight: Silksong, Homeworld 3, Little Nightmares III, Metro Exodus Enhanced Edition, Ori and the Will of the Wisps, The Callisto Protocol, and Warframe have been given the IBOT treatment.

Intel claims that with a Core Ultra 7 270K Plus, paired with a GeForce RTX 5090 and 32 GB of DDR5-7200, you can expect to see up to 27% more performance in Silksong at 1080p High, down to a mere 2% frame rate bump in Metro Exodus.

The sequel to Hollow Knight runs generally fine as is, though I have seen some reports that the frame rate can unexpectedly plummet in some boss battles. Whether that's a sign of a CPU limitation or game optimisation issues isn't overly clear, but hey, getting 27% more oomph for simply updating some software and then clicking a few buttons is fine, yes?

Well, as with all things CPU-related, it depends very much on what resolution, graphics settings, and graphics card you have. Playing a game at 1080p High with an RTX 5090 is a pretty unusual scenario, and if you had that level of hardware, you'd be playing the likes of Metro and The Callisto Protocol at 4K with ray tracing enabled.

In such circumstances, the GPU becomes the limiting factor to performance, unless you use upscaling to drop the rendering resolution down to 1080p or lower. Although that will bring the CPU back into the mix, the game's frame rate will still be heavily controlled by the GPU because of the ray tracing.

For example, I've tried the new BOT update in Metro on my 270K Plus, RTX 4080 Super main PC and didn't see any difference at all when using 4K High. But with DLSS Performance enabled, the average frame rate increased by an average of 1% across three runs. So in line with Intel's figures, but also well within a margin of error.

Anyway, if you do have a compatible processor and you want to check out IBOT for yourself, then all you need to do is download the latest Intel Platform Performance Package (IPPP).

Thermal Grizzly has some products in the works that will make any custom loop aficionado salivate. Though there's one that stood out to me while visiting its Computex 2026 booth: an experimental direct-die block with diamonds under the cover.

"So this block includes a diamond, industrial diamond, which is like exactly like what you see here," Sasha Robey, Senior Marketing Manager at Thermal Grizzly, says as she gestures to the two incredibly thin diamond sheets on display "To put it on the block, we have to put a special metal coating on it, because otherwise liquid metal won't stick to it."

The block sits directly atop of the CPU silicon—hence the name, direct-die. The model on display at Thermal Grizzly's booth is designed for an AM5 processor, and one with two CCDs, so ideally a Ryzen 9 9950X. Though you do have to delid a processor to get down to the silicon, which can be a precarious business.

You can see on the inside of the block where each chip will sit: the IO chip is cooled by lapped metal, but above the two CCDs sit the two razor-thin CVD diamond inserts.

CVD stands for Chemical Vapor Deposition. The end product is what, in regular parlance, would be called a lab-grown diamond. Though it's just as real as any other diamond. Thermal Grizzly is then using a very thin sheet of this diamond for its thermal benefits.

"The thermal conductivity is three or four times better than normal blocks the industry is using," Robey says.

Thermal Grizzly notes that the thermal conductivity of copper, which is often used for these applications, including the rest of Thermal Grizzly's direct-die block, is approximately 390 W/mK. This diamond is up to 2200 W/mK.

Though Robey is quick to point out this is a personal project, experimental as heck, and unlikely to make it to market because Thermal Grizzly understands the price tag may be a bit of a shock. Each diamond insert costs around €500 and I'm given a very rough estimate for the total cost of the block: €1500 or more.

"I think this will never make it to the market, but yeah, an interesting example of what is possible."

What is eventually coming to market are these gorgeous new coatings for Thermal Grizzly's non-diamond CPU and GPU blocks. It's joined forces with Swiss company Platit to test out a range of new coatings, including a carbon-based option (light gray), a nitride-based option (dark grey), and a oxynitride-based coating (rainbow).

"We're exploring different types of coatings, so that we could potentially move away from nickel-plating, because it's not great because of corrosion," Robey says. "Plus for us it means when we have to ship blocks out for nickel plating, we wait four weeks or more sometimes to get products back. So we would really like to internalise this process to also be able to do it."

TG O is by far the best-looking up close, with a rainbow patina that sorta stole the spotlight from the rest for me. It looks superb on the CPU block too.

"Everybody likes the rainbow one," Robey says, "and it looks like rainbow because it's so thin that the light reflects in a way that we see rainbow."

Thermal Grizzly hasn't yet decided which options it's going to go for or if it'll just offer a few as customisation options.

Intel is struggling to supply laptops CPUs built using its latest 18A silicon production node, according to a new report. That basically means the Panther Lake CPU, which forms the basis of its new handheld chip the Arc G3, and the budget-priced Wild Cat Lake chip. But it may not be the 18A node itself that is the problem.

Instead, a combination of factors, some out of Intel's control, are said to be in play. The report comes from Tim Culpan, a technology journalist based in Taipei, Taiwan.

He has been speaking to numerous industry contacts at Computex 2026, including representatives of three of the six largest laptop PC manufacturers. And a common narrative has emerged.

"I spoke with people at three of the world’s top-six laptop brands, and they all gave the same feedback: supply is tight," Culpan says.

"The chipmaker is struggling with delivery of the laptop chips and has been unable to provide clarity on when the shortage would ease, according to people I’ve spoken to at major PC brands, smaller laptop makers, and system assemblers," Culpan added.

Culpan says that Alex Katouzian, Intel’s general manager of Client Computing and Physical AI, told him that there are "some" shortages of the chips but that Intel is "overcoming" them. However, the reasons are not necessarily that Intel has any problems with its 18A node.

Instead, Culpan says the supply chain and most notably TSMC may be at least part of the explanation. As he points out, Intel's Panther Lake CPUs have CPU tiles built on the in-house 18A node. But the I/O tile is made by TSMC.

Culpan claims Intel's relationship with TSMC may be "under strain" and that "TSMC’s capacity is notoriously tight right now, and Intel is unlikely to be high on the priority list."

Another part of the puzzle is Intel's new server chip Clearwater Forest, branded as Xeon 6+. It, too, is built on Intel's 18A node. Culpan says PC makers suspect that Intel is prioritising 18A capacity for Clearwater Forest over lower-margin laptop CPUs.

At the Computex press conference, Culpan asked Intel's data centre boss Kevork Kechichian how 18A capacity is allocated. “It’s complicated,” Kechichian said enigmatically, "it’s not an easy thing.”

Of course, Intel's Nova Lake desktop CPUs are due out towards the end of this year and will add further competition for 18A capacity. So, if 18A really is being hogged by those server chips, that's yet more bad news for we poor PC gamers.

AI is everywhere at Computex. Obviously. It's been like this for the past three years, where those two characters have become so ubiquitous around any tech show or presentation by any hardware or software company that I've become almost inured to it. But it's a big reason about why we can't have nice things.

It's why the DIY PC building market is down across the board, to the tune of around 30 - 40%, according to all the manufacturers I've spoken to at the show. It's why the cost of the Steam Deck has blown up, why the Steam Machine still isn't here, why the MSI Claw is going to cost $1,500, and why possibly the most tantalising Intel Panther Lake processors will probably never appear in actual laptops.

The top Panther Lake chips come with the 12 Xe-core B390 iGPU, and it's a powerful thing. It's the graphics component going into the Intel Arc G3 Extreme processors—a chip with some much graphical grunt Intel's Tom Petersen is calling it a GPU with an integrated CPU. But there is another.

The B370 is the integrated graphics chip which appears in the Core Ultra 5 338H, comes with 10 Xe cores and ought to deliver a similarly high level of gaming performance but in a cheaper package. When Intel first announced Panther Lake that was the chip I was desperate to see in an affordable Core Ultra Series 3 laptop.

But it's looking unlikely that I ever will. And it's also potentially unlikely that we're going to see any handhelds using the Intel Arc G3 chip which also comes with the same B370 iGPU.

At a demo of the new Intel Arc G3 Extreme in a range of different handhelds—including the MSI Claw, Acer Predator Atlas 8, and OneXPlayer 3—I asked Petersen whether any manufacturers were going to use the straight Arc G3 chip.

"I don't know the answer to that one," he says. "It's a great question. I think we're in a weird time with the whole DRAM thing going on right now, where the pricing difference between the two chips is small, and small relative to what's going on with DRAM.

"So, if you're going to spend the money on DRAM, why not just get the best one? I think there's some of that going on. As DRAM prices come down and the market starts normalising a little bit, then the differentiation of price that you can get by picking different SKUs will be a more important differentiator than the end user. That's my guess."

Which makes complete sense, though is all rather depressing in explaining why cheaper Panther Lake laptops with the B370 haven't arrived. It's all AI's fault and the fact it's hoovering up all the spare memory capacity and driving up prices across the board. When the difference in cost of a chip with a B370 or B390 is so comparatively small compared to the speedy LPDDR5x memory (at least 7467 MT/s or faster) that Panther Lake requires, what's the point picking the slower graphics core?

When I spoke to Framework CEO, Nirav Patel, this week at Computex, he essentially said the same thing about why it wasn't offering that version of Panther Lake as an affordable option with the new Framework 13 Pro machine.

"We basically decided that if we go all the way to the X versions, let's just get the full GPU capability," Patel explains.

I asked if that was specifically because of the small difference in comparative cost of the CPU versus the memory and he says, "Yeah, we always look at the total SKU stack that Intel or AMD offer, and we pick what we think the right sweet spots are. So it's basically for the [Core Ultra] X7 we want to make sure someone's getting the fully capable solution, and if they want to go to X9 it's really that they want that last few percent of performance for the people that can pay for it. We're going to make sure the X7 is not a compromise."

So yes, AI is still why we can't have nice things.

AMD has brought back the Ryzen 7 5800X3D specifically to cater to a DIY market where there is still a ton of AM4 motherboards and DDR4-supporting machines in the market. "We see a lot of users still gaming on like Ryzen 5 2600X or 3700X," says David McAfee, VP of Ryzen and Radeon at AMD, "where as a drop in replacement to an existing motherboard and build, they're going to see just a massive, massive jump in gaming performance."

We're sat on the 26th floor of a hotel that overlooks the Nangang Exhibition Centre, where the Computex show is still thrumming away below us, and McAfee has just explained that it wasn't simply a case of bringing the old manufacturing process out of retirement and spinning up the machines again.

The original way it had stacked the 3D V-Cache on top of the Ryzen 5000-series CPU die was no longer in use at TSMC. "We had to kind of re-engineer, re-qualify, and rebuild that product in a way," says McAfee, "so that it could migrate from that old process that really wasn't around anymore to the newer process."

He notes the changes will make practically zero difference to the relative performance of the newer 10th Anniversary dies compared with the original chips, and that it is most definitely not flipping things upside down in the way it stacked the Ryzen 9000-series 3D V-Cache chips. But it was still a process the company had to go through in order to make the new CPUs a reality.

Given the level of commitment necessary to redesign and re-engineer the old chip, I ask him how difficult a conversation it was deciding to bring this classic CPU up to date.

"Very hard, actually," he says, laughing. " Very, very hard!"

"The tricky part of this particular product is it's also a matter of predicting what's going to happen with memory supply in the market, right? This was also, if you recall, at a time when every major DRAM vendor on the planet is saying, 'Hey, we're shutting down all our DDR4, it's all going DDR5.' And so there was also a lot of work we did with the ecosystem to understand, around the world, is there enough DDR4, like fresh DDR4 that's coming in. Is there simply enough memory to make sure that when we put these products out there they'll sell.

"These were probably the two biggest challenges: making sure the ecosystem could absorb this new DDR4 product that required new DDR4 memory to to absorb it into the market, and then second, is there enough customer demand there to do it? And I think you know one of the best things about the DIY community is we get a lot of feedback from customers, and those opinions on what products they want to see."

With a defined end-point in sight for DDR4 as a newly manufactured product, and increased capacity for DDR5 and other leading-edge memory products coming online in 2028, I wouldn't expect AMD to be creating other new DDR4-based products.

Sometimes the best things at Computex are the things that aren't at Computex; namely, talk of upcoming tech. Case in point here is AMD's Ryzen and Radeon VP and GM David McAfee telling Tom's Hardware that a Zen 5 replacement for the Ryzen 5 7600X3D—meaning surely a Ryzen 5 9600X3D—“may be something that we look at doing… later this year."

AMD launched the former chip way back in 2024, but it was very limited in numbers because it was only sold by Microcenter. There was talk of a 9600X3D as long ago as this time last year, as references to it were found in driver notes, but nothing ever came of it.

The 'X3D' in the name refers to AMD's 3D V-Cache, which is stacked vertically on top of or under the processor—under, in the case of the 9,000-series—giving the chip easy access to much more L3 cache than a normal processor gets. Lots of games really benefit from big processor caches, so naturally, X3D chips make for some of the absolute best CPUs for gaming.

Given that the cheapest of current-gen options is the Ryzen 7 9800X3D, which costs north of $400, a cheaper six-core option would be nice. Though it might not keep the same 3.9 GHz base and 5.4 GHz boost clock of the Ryzen 5 9600X, given (A) the chip would likely be based on a cut-down eight-core rather than a six-core 9600X, and (B) the stacked cache might prevent the chip from reaching its full potential on that front. Still, though, it would make for a formidable cheaper chip.

But we shouldn't get carried away, because the main problem a Ryzen 5 9600X3D might face is competition from other AMD chips. Given how cheap the Ryzen 5 7500X3D and Ryzen 5 7600X3D CPUs are, many might opt for those instead. And if you want to spend more, unless it's priced very generously, it will be hard to convince some to get a 9600X3D over a 9800X3D, which will have more longevity.

McAfee, however, explained the possible market for this chip as follows:

"I think as we go through the rest of this year, I mean we're always looking for ways to, you know, create as many options as we can, and that may become… maybe something that we look at doing as a runway product later this year, simply because you know we know the pressures that are there in building systems aren't going away anytime soon."

That much is true: more options never hurt in this market. And perhaps it would be more of a processor for pre-builts anyway, in which case it could, for instance, offer a new option for high-end builds when system builders want to save a bit of cash and not opt for a 9800X3D.

At any rate, it seems it's not off the cards for AMD.

Never one to be short of hyperbole, at a press Q&A with the Nvidia CEO of course talk of the new RTX Spark SoC was on the menu. The Nvidia RTX Spark (AKA N1X) is a chip built in partnership with MediaTek, combining a custom Arm core (though quite how custom it is, I'm not sure) with an Nvidia RTX Blackwell GPU core. Connected via a super-quick NVLink connection and sharing a pool of LPDDR5x unified memory, the SoC is one of the most exciting things to come out of Computex this year, and Jen-Hsun says it is "100% awesome".

While it is absolutely being aimed at gamers in some form—though probably not in its peak, 128 GB guise—it is also being introduced as the hardware necessary to make your laptop less of a tool and more of an R2D2, a robot assistant. But even though it is being designed to make the AI the UI, it is still just using Windows.

The platform has been designed with Microsoft with Windows-on-Arm very much at the forefront of what Nvidia's doing, but that's still just a very standard Windows operating system. So maybe that itself doesn't make it feel immediately like a transformative moment.

Which is why someone asked the Nvidia CEO if maybe there should be a fundamental change in the UI, or even the OS itself, to encourage people to think about using these RTX Spark devices in different ways.

"The good news is that this PC is 100% awesome at everything everybody expect the PC to do," says Huang. "And it can do more.

"So you can go to that journey at your own pace. I think if we were too abrupt in changing either the UX, too abrupt in changing maybe even the name and its positioning, maybe the people who believe they still need a PC might have a harder time on that journey."

And it is going to be quite the journey, because as Jen-Hsun notes: "N1X has N2X and we've got N3X already planned." We've seen the roadmap, and with an RTX Spark Vera Rubin and RTX Spark Rosa Feynman on the way in 2028 and 2030 respectively there's a long way to go on that journey.

While AMD didn't have much in the way of brand-new PC gaming hardware for Computex 2026—just a cut-down GPU, plus a last-gen X3D CPU and a last-last-gen one—it has at least reminded us of something important for many rig builders. Its current desktop processor socket will still be supported by new chip releases for at least another four years, but that doesn't mean AMD is just leaving things as is.

At the yearly megatech event in Taiwan, AMD reiterated something that it promised back in 2024, in that the AM5 processor socket will be around for a good while yet. Back then, it committed to "2027 and beyond", but that latter part now has a bit more meat to it, with Team Red crossing it out and replacing it with 2029.

With Zen 6 chips expected to make an appearance at some point later this year or early in 2027, this suggests that you'll be able to install Zen 7 CPUs in the same socket. Whether a current X870 or B850 motherboard will happily take such a processor is another question, but those that sport extra-large BIOS chips probably will.

In terms of Ryzen nomenclature, it looks like AM5 will run from the 7000-series all the way through to the 11000 or even 12000-series, assuming AMD keeps to that code scheme (something that the company doesn't exactly have a great track record of).

As a quick note, you might think that AMD names its sockets after what DRAM it supports, e.g. AM5 uses DDR5, AM4 uses DDR4, and AM2 is DDR2. However, it's more of a coincidence than anything else, because AM3 is fine with DDR2 or DDR3, and AM1 (which appeared between AM2 and AM3) is DDR3 only.

With this confirmed support, it does make me wonder if AM5 will outlast the 'legendary' AM4 socket, because that's still enjoying new chips appearing even now, albeit just those being spat out again from Team Red.

Anyway, for some PC builders, all of this is going to be great news, because the idea that you can have a PC with a Ryzen 5 9600X that you could drop a Ryzen 9 11950X3D into is really appealing. I'm in the alternate crowd, in that I keep the same processor and motherboard for four or five years and then change the whole lot, but I fully appreciate why lots of PC enthusiasts prefer AMD over Intel because of the socket longevity.

It's also worth noting that just because AM5 is going to be around for a good while longer, AMD isn't just leaving it alone. In fact, quite the opposite, as the chip giant also announced an upcoming tweak to its EXPO memory overclocking system at Computex. I know that's not the same thing as upgrading a processor socket, but having faster memory does require that the socket's electrical properties are up to supporting the tweak.

Called EXPO Ultra Low Latency, or EXPO ULL for short, AMD is promising up to 4% more performance in games, compared to AM5 PC's using the standard EXPO profile. What resolutions and graphics settings are required for this performance gain isn't clear at all, but you typically need high-end hardware and 1080p to make DRAM the limiting factor in the frame rate.

So EXPO ULL is likely going to be of use to the esports crowd only, and even then, you'll need to have an updated motherboard BIOS to support it, and buy a new set of DDR5 memory that has the updated EXPO profile. You know, the stuff that's just so cheap and readily available at the moment.

After being rumoured to do so for a few weeks now, AMD has confirmed it is indeed bringing back the Ryzen 7 5800X3D with AM4 10-year anniversary packaging, as well as a mid-range AM5 Ryzen 7 7700X3D chip.

From the top, the 5800X3D originally launched in 2022, but the updated model (celebrating 10 years of AM4) comes bundled with a Carbice ice pad and retails at $349. Offering eight cores and 96 megabytes of 3D V-Cache, it's still a decent AM4 CPU, and its compatibility with DDR4 is what may make it worth many thirsty PC Gamers' while. In AMD's press material, it announces the 5800X3D as the "return of the king".

If you're not looking to upgrade to AM5 or get DDR5 memory (thanks, memory crisis), the $349 MSRP is still a little under what you can expect to pay second-hand right now. Still, though, it feels like a lot for an older chip. It's just another sign of how bad the memory crisis has become, but it doesn't seem like a terrible buy if you simply have to build a PC right now.

The 2022 CPU 2022 got an 81% from us in our AMD Ryzen 7 5800X3D review. Our Dave praised its 3D V-Cache and the chip's efficiency, but critiqued its heat and price point (which was at the time $450). The 10-year anniversary edition will be available from June 25 this year, and I'm sure that ice pad will be a welcome addition.

Alongside this, AMD has announced the Ryzen 7 7700X3D, an eight-core 120W AM5 chip coming in at $329. It will be available as soon as July 16, and comes with a boost clock of up to 4.5 GHz. With 16 threads and 104 MB of cache, it seems like a bit of a tough sell so far; it drops 500 MHz from the Ryzen 7 7800X3D, which only fetches tens of dollars more at Amazon (around $370 as of the time of writing).

And this is all assuming you will actually get your hands on the chip at MSRP. As we've seen over the last few years, MSRP is no guarantee. Still, more options on the market, and more ways of using the same hardware are usually better for the customer—If you have more options to choose from, you can actually afford to shop around.

Alongside this, AMD has announced it will continue to support AM5 through 2029, so at least if you do make that upgrade to AM5 with the 7700X3D, you know you've got a few more years with it than you may have otherwise expected.

ByteDance, the company behind TikTok and video editing app CapCut, is developing its own CPUs in a bid to better support its AI infrastructure. Given the AI industry's hunger for silicon, it's hardly surprising another massive tech company is taking hardware into its own hands.

Several external partners have already been approached with regard to design work for ByteDance's chip. Securing capacity at manufacturing foundries was apparently also discussed. The plan is in the early stages and not yet public, but Reuters spoke with a number of anonymous sources familiar with the matter.

Currently, ByteDance is exploring two different chip architectures: Arm and RISC-V. Both architectures are used for chips in data centres throughout the wider industry, though Arm presents a proprietary ISA, with a fixed feature set to go along with it, whereas RISC-V is a modular, royalty-free, open-source architecture. I can imagine not having to pay Arm's licensing fees may be especially appealing to any major player looking to make their own chips.

There's been a lot of excitement about Arm chips in the realm of PC gaming, especially as the hardware team waits for even a whisper of news about Nvidia's still-yet-to-be-announced N1X Arm chip (various reports have insisted it's coming at some point this year). As for the alternate, open-source architecture…well, a few years back someone built a fully compliant RISC-V computer inside Terraria—and then they played Pong on it.

As the AI industry reaches towards CPU-intensive 'inference', ByteDance's planned chip is intended to be deployed throughout its AI data centres and servers. This, in turn, will support ByteDance's agentic AI products, such as the company's development platform, Coze.

Reuters reported earlier this year that Intel had warned its Chinese customers that orders for processors could take six months to fulfil. The same report claims AMD was looking at 10-week delivery times in the same market, and ByteDance currently sources its CPUs from these two partners.

With that in mind, alongside the knock-on effects of the memory supply crisis, it's unsurprising ByteDance would take the route of making its own hardware. It's not the only company to go this way, either, with Google announcing its own Arm-based CPU all the way back in 2024. Microsoft, too, is moving towards designing more custom silicon for its own data centres, and so is Amazon.

With Intel riding high at the moment, thanks to its excellent Panther Lake and Arrow Lake Refresh processors, any news about its forthcoming Nova Lake chips is worth paying attention to. Point in case: it's rumoured that Team Blue will release a 100% E-core mobile chip, with a big integrated GPU, with its next-generation architecture.

Admittedly, there's not an awful lot to go with Golden Pig Upgrade Pack's post on Weibo (via Wccftech), but what's there is interesting. The alleged processor will comprise eight E-cores and a 12 Xe-core GPU, and in the case of the latter, that's actually more than enough to run plenty of games at 1080p with a decent frame rate.

Intel's current E-cores in Arrow Lake are also surprisingly capable, so if we assume that the next (and possibly last) iteration of them is even better, but whatever the funky little chip is using, it could be an excellent low-power option for handheld gaming PCs.

Except there's a wee fly in the ointment in the fact that the leak suggests this chip will be purely for edge computing. Unlike Nvidia, which now classifies everything but AI and data centres as 'edge', Intel uses the term for any computational work that's essentially done where the data is generated. For example, robotic systems in a production line or autonomous driving.

When I visited Intel last year in Arizona, I spent a little time chatting with a development engineer about Panther Lake's NPU being used in AI edge applications. That part of the processor is great for low-power situations, but anything more demanding really needs a GPU behind it. We both agreed that an ideal chip for such situations would have a very low-power CPU section, but a really beefy iGPU.

Well, this is what we appear to be getting here, and I can't help but feel that it would actually be really good in a handheld. Alas, that's unlikely to ever happen, given its intended market, but you never know. One enterprising firm might create something experimental with such an edge chip and surprise us all.

The RAMpocalypse will almost certainly kill off any notion of such a venture, though, because prices for chips in the edge market are considerably higher than the domestic one. The combination of a unique, slightly odd processor that's expensive to buy, coupled with ludicrously pricey DRAM, would just make such a handheld unfeasibly costly. Bah, we never get anything nice these days.

The first true Nvidia CPU has been benchmarked and found to beat, well, just about everything. The catch? While this is an independent test, Nvidia dictated the kinds of workloads that could be assessed.

We speak, of course, of Nvidia's Vera CPU, part of the company's upcoming nexty-gen Vera Rubin AI platform. Vera contains a CPU core known as Olympus, and the reason why it's the first "true" Nvidia CPU is because those cores were designed in-house by Nvidia to use the Arm instruction set. Previous Nvidia CPUs, such as Grace, used CPU cores designed by a third party, such as Arm itself.

The first generation of Vera is an 88 core chip with support for 176 software threads. Nvidia has said it offers double the performance of the Grace CPU.

Anywho, Phoronix.com did the benchmarking and we're not exactly talking gaming here, people. That website's audience is pretty technical, arguably with a bit of an enterprise bent.

So, the benchmarks included the likes of code compilation, Python performance, Open JDK Java workloads and so on. That said, AV1 and 7-Zip benchmarks were also on the list, which are relevant to consumer PCs.

As it happens, Vera traded blows with AMD's Epyc server chips for video encoding and pretty much rubbed the Intel Xeon competition out. As for 7-Zip, Vera's per-core performance was getting on for 20% better than any x86 CPU.

Overall, in Phoronix's geomean of results, Vera beat the best AMD Epyc chip by about 10% and walloped Intel's flagship Xeon CPU by over 50%.

Phoronix also claims that Vera easily has the measure of pretty much any Arm CPU for servers, though figures weren't provided. "Vera is much more performant than what we have seen out of the likes of Ampere Computing or the custom in-house Arm solutions at public cloud providers like Google Compute Engine and Microsoft Azure," the site said.

There was no mention of Qualcomm or Apple CPUs, both of which obviously use the Arm instruction set and are capable of very impressive per-core performance.

What can be concluded from all this? Nvidia seems to have a very performant CPU core on its hands in the Olympus core. How relevant that is for PC gaming is less clear.

Nvidia's upcoming N1x chip for the PC, as far as we know, uses off-the-shelf cores designed by Arm, not the new Olympus core. However, if Nvidia is committed to producing Arm CPUs for the PC, it seems reasonable to expect that the company might unify its efforts, just as AMD and Intel do with its consumer and enterprise CPU core designs.

And so Olympus, or perhaps a successor CPU architecture that follows it, could be the basis for future consumer and gaming PC processors from Nvidia. And on this first showing, it looks like Nvidia might just come up with something pretty awesome. Just don't expect to see it in an actual PC anytime soon. The N1x chip isn't even out yet, after all.

Hunting for fresh news of Nvidia's long-awaited, yet-to-be-announced N1 and N1X laptop chips is pretty difficult these days. Reports earlier this year suggested the new chips would be out by March, but here we are in May with not a lot to go on beyond rumour and speculation.

Other than a Lenovo login page mention, of course. That's new (via Videocardz). A dropdown menu on the Lenovo ADFS login page has two entries noting the fabled chip, one labelled "Nvidia N1X Portal PROD" and the other "Nvidia N1X Portal Test."

It's not the first time Lenovo has been linked to potential Nvidia-chipped laptop models. At the start of this year, the "Legion 7 12N1X11" appeared as a supported model for its Legion Space software, while a dataminer also found mention of several upcoming Lenovo models with N1 and N1X designations.

So as secrets go, it's not a particularly well kept one. And in all honesty, test portals being prepared for upcoming laptops using the chips is no great surprise.

But it's been a long time coming, and if the cadence of these leaks is anything to go by, we may be tantalisingly close to an official reveal. Computex 2026 kicks off next week, and if I were a betting man, I'd put good money on at least some mention of the near-mythical Nvidia laptop chip line being made then—if not a full-blown announcement.

Nvidia CEO Jensen Huang has confirmed that the new offerings will be based on the GB10 "Superchip" found in the DGX Spark, although how that manifests in practice is yet to be officially revealed. Previous rumours have suggested the N1X is a 20-core CPU with a large Blackwell-based GPU, potentially with as many CUDA cores as the RTX 5070.

If that's even close to accurate, the Nvidia beasties may well be one of the biggest and most interesting hardware releases of the year—if the X86 emulation is all fine and dandy, of course. Or we could be left waiting once more. Tune in next week to find out?

Intel has been in the news quite a lot in recent times, and for once, all for the right reasons. Panther Lake and the Arrow Lake refresh chips are top-notch, and at long last, orders are coming in for its foundry services. And if a leaked roadmap for its future desktop and laptop chips is anything to go by, Intel will be a headline story for many years to come.

Admittedly, though, deciphering the claimed roadmap from the YouTube channel Moore's Law is Dead takes a fair bit of processing, if you pardon the pun. As mentioned up top, Intel's current chip lineup primarily consists of Arrow Lake (ARL, desktop and laptop) and Panther Lake (PTL, laptop and handheld).

Naturally, these will be replaced by the next generation of microarchitectures, and first up will be Nova Lake (NVL), which is expected to make some kind of an appearance toward the end of the year. That much we already know directly from Intel, but there have been plenty of additional leaks and rumours to fill in the gaps.

After NVL, the alleged roadmap suggests that we'll see Razer Lake, though many SKUs will actually just be rebranded Nova Lake chips. After that, Titan Lake will make an appearance, replacing PTL, and this is where things begin to get interesting.

Ever since Alder Lake in 2021, its 12th Gen Core architecture, Intel has employed a hybrid approach to its consumer processor designs. Alongside the powerful and power-hungry 'Performance' cores, there are clusters of compact and power-sipping 'Efficient' cores. That will still be the case with Nova Lake and Razer Lake, but both this roadmap and previous rumours suggest that the separate core approach will be gone.

While this is being labelled as a 'unified' core design, MLID is suggesting that while there will still be P- and E-cores, Intel will be taking a leaf from AMD's book and the cores will be architecturally identical. It's just that the 'E-cores' will be more compact and lower clocked, as you see with some Ryzen laptop chips that have Zen 5 and Zen 5c cores.

Still with me? Good. On the desktop side of things, the successor to Nova Lake and Razer Lake will be Hammer Lake (cue endless Stop! It's hammer time! memes). Not only will this introduce a second generation of the unified core architecture, but it's also claimed that HyperThreading, Intel's simultaneous multithreaded system, will make a return.

It was dropped for Arrow Lake to help ensure the chips stuck within the targeted power window, and truth be told, you wouldn't notice that it's missing, given how well the Core Ultra 7 270K Plus performs in multicore tasks.

Additionally, the majority of Hammer Lake desktop models will be 'P-core' only, just as AMD doesn't use its compact cores in socketed Ryzen chips. Oh, and speaking of sockets, here's the final head-spinner: Nova Lake, Razer Lake, and Hammer Lake will apparently all use the same LGA 1954 socket, though whether that includes the massive cache, dual compute tile variants is unknown at this stage.

I've only briefly touched on what the roadmaps have to tell us, as the rest of the details are somewhat complex and confusing, and I'll need more than one cup of coffee to make sense of it all.

Although the RAMpocalypse is doing its very best to completely ruin PCs for everyone right now, perhaps by the time Intel's unified chips come to market, things will be a lot better. Maybe even back to normal—whatever normal is these days. If that comes to pass, and you've held off from upgrading until then, you could be in for a treat when it comes to picking your next gaming CPU.

AMD has announced that its first 2 nm CPUs are now in production. It's a server chip, inevitably, but it bodes well for upcoming PC processors. Meanwhile, Intel has mentioned possible future 10A and 7A nodes for the first time. Maybe Moore's Law ain't dead, after all.

"AMD today announced that its next-generation AMD EPYC processor, codenamed Venice, is ramping production in Taiwan on TSMC’s advanced 2 nm process technology, with future plans to ramp production at TSMC’s Arizona fabrication facility," AMD said in a statement.

Venice is based on AMD's next-gen Zen 6 CPU architecture, which will be shared with AMD's next-gen desktop CPU family, codenamed Olympic Ridge and possibly branded Ryzen 10000 Series.

Jumping to TSMC's 2 nm node, branded N2, is an interesting move given that most of AMD's current CPUs, including the Ryzen 9000 series, are built on silicon derived from TSMC's N5 node. In other words, AMD has only made limited use of TSMC's N3 node thus far, and the move to N2 means that the company will skip N3 to a large extent.

It's worth noting in all this that, right now, there are no commercially available chips built on TSMC's N2 silicon. It's likely that Apple will announce products made on that process node in September, mostly the iPhone 18. But AMD is being pretty aggressive with its wholesale move to N2.

One of the more obvious anticipated benefits of N2 silicon is a move to 12 cores in each CPU chiplet (Core Complex Die or CCD), up from eight in AMD's current chiplets. That opens up the possibility of both 12-core gaming-optimised CPUs with 3D V-Cache and monster 24-core dual-die models for heavily multithreaded applications.

As for when we'll actually be able to buy Zen 6 CPUs, AMD has said the architecture will launch in 2026, and at this stage, it's looking like late 2026. However, the chip giant hasn't been entirely transparent about which Zen 6 CPUs will be released this year. Our bet is that it will only be EPYC server processors in 2026, with desktop models to follow early next year.

If AMD is moving to ultra-advanced N2 silicon, what about Intel? It is already selling Panther Lake mobile CPUs using its new 18A node, and the company is bigging up its next-gen 14A node.

But somewhat surprisingly, CEO Lip-Bu Tan recently teased even more advanced production nodes. Speaking at the J.P. Morgan annual tech conference earlier this week and in response to questions about Intel's customer foundry business, Tan said, "I'm starting to look at the 10A, 7A, the roadmap. People don't go to you for just one node. They're looking for the roadmap for the future."

As far as we know, this is the first public mention of possible 10A and 7A nodes from Intel. That said, those nodes may well be largely speculative. After all, Tan confirmed at the conference that Intel's 14A node won't be in volume production until 2029.

But the mere fact that Tan is talking about technology that far out is a big turnaround. It wasn't that long ago that Tan was threatening to cancel 14A with the unsaid implication that Intel might get out of the chip manufacturing game altogether.

Anywho, this all looks pretty positive regarding the broader progress of the PC based on ye olde Moore's Law or at least something like it. AMD N2 chips are definitely coming, and Intel is plotting a roadmap of more advanced silicon for many years to come.

OK, the traditional cadence of Moore's Law involving the doubling of transistor density and halving of costs every two years is arguably dead. But there's still pretty major progress being made.

In a move seemingly designed to make your erstwhile correspondent feel old, it seems AMD is rolling out a special edition of the trusty Ryzen 7 5800X3D gaming CPU to celebrate 10 years of the AM4 socket. Yup, 10 years.

Spotted by X user momomo_us (via Tom's Hardware), the chip has been listed at Indian retailer 99Deals for the equivalent of around $310. The OG 5800X3D is in pretty limited supply these days. Listings on Newegg, for example, are limited to a supplier in Hong Kong for over $400. It's a similar situation for the 5700X3D.

Meanwhile, the closest newer equivalent, the AMD Ryzen 7 7800X3D goes for over $380 and of course slides into the AM5 socket. So, a reboot of the 5800X3D at that price is pretty appealing, especially if it means you don't have to buy any DDR5 memory at whatever bananas price is being charged at any given moment.

Of course, with news that AMD is also adding a Ryzen 7 7700X3D to its line up, there's a healthy selection of 3D V-cache processors from AMD on multiple sockets.

The rebooted 5800X3D seems to come in a natty box with "10th Anniversary Edition: April 16th 2026, 10 Years of Innovation" branding, the 10 years bit referring to the AM4 socket, rather than the chip itself. The 5800X3D is pretty old, but not that old. It was released around this time in 2022.

The 5800X3D is an eight-core model with 96 Megabytes of game-accelerating 3D V-cache and remains a very decent gaming CPU here in 2026. That fact, combined with the AM4 socket's DDR4 as opposed to DDR5 memory support, is where the main appeal is here.

As far as we know, AMD has made no official announcement regarding availability of the chip. But what with this listing and a leak earlier last month that ties in with the date on the box, it looks pretty likely that the chip is for real and could soon be widely available. If you have an AM4 system or even some old DDR4 knocking about, therefore, this chip could be a pretty appealing proposition.

Given the veritable wealth of different Ryzen X3D processors AMD has been churning out in recent times, you'd be forgiven for missing the fact that there's been one notable absence from the full list: a 5700X3D equivalent of the Ryzen 7 7800X3D. However, if one social media post is to be believed, then that box is shortly going to be ticked.

Ordinarily, a single post on X is nowhere near enough compelling evidence for me to make such a claim, but when the account in question belongs to chi11eddog, it's certainly worth paying attention to. They have a pretty stellar track record for dropping early hints on new chips, with correct specs, such as getting it bang on for the 9950X3D2 Dual Edition and 9850X3D.

Anyway, chi11eddog's latest post is for a Ryzen 7 7700X3D, and the stated specs—120 W, eight cores, 16 threads, and 96 MB L3 cache—all point to this being a processor with a 3D V-Cached chiplet that failed to meet the requirements for use in a Ryzen 7 7800X3D.

The process of applying the 3D V-Cache to the CCD doesn't produce 100% perfect results every time, and rather than wasting the failed couplings, AMD just puts them to one side. Some of them might end up in a Ryzen 5 7600X3D or 7500X3D, but since these SKUs are 12-thread processors, two cores have to be disabled.

With a 7700X3D, you're getting the full CCD, but I wouldn't get too excited just yet. Mostly because chi11eddog is claiming that the peak boost clock will be a mere 4.5 GHz. That's a drop of 500 MHz from the 7800X3D, and while that only equates to a decrease of 10%, the combination of the relatively slow clock speed and high TDP of 120 W, suggests that these chips will run quite toasty.

If AMD does decide to roll out a Ryzen 7 7700X3D across all retailers, the point about whether it will be worth picking one up will come entirely down to the price tag. Right now, you can bag a full-fat 7800X3D for $377 at Amazon, so expecting a sub-$300 price for the 7700X3D isn't unreasonable.

The Ryzen 7 5700X3D launched with an MSRP of $249, considerably less than the 5800X3D's $449, but the latter price was perhaps overly high in the first place. Back then, it was the first and only X3D chip, but these days we have a whole lot more to choose from.

With luck, AMD will do the decent thing and pitch the 7700X3D significantly under $300, to help offset the horrendous cost of the DDR5 kit you'll need to go with it, if you're upgrading from an AM4 or older Intel setup.

Intel CEO Lip-Bu Tan sat down with Mad Money host Jim Cramer for a chat about the company's recent success, and the turnaround it's experienced during his tenure. In a wide-ranging discussion, the Intel head honcho credited many reasons for Intel's recovery from its previously shaky spot, including a change in management style.

"In the past, we made a lot of mistakes, and now we correct that mistake, and we simplify the roadmap. From day one I came on board as a CEO, I have all the engineering report to me," said Tan.

"They didn't report to the previous CEO?" asked Cramer. "No," Tan replied. "And so, in a way, they [had] too many silos, too many people needed reporting. And even though they come to the CEO, they are not paying attention to it."

"So I decided the best thing is to really understand where the problem is," Tan continued, "so that I can focus on the engineering, how to redesign, [how to] simplify the products, and then get the real killer product out."

And by the sounds of it (and the look of Intel's most recent financials), Tan's hands-on management style appears to have worked. Addressing Intel Foundry, Tan said: "When I took over, the 18A yield [was] not good. And I had to ask some of the ecosystem partners to help me look at the data, how to improve the industry standard."

"The best practice is to see seven or eight percent yield improvement per month, and now I'm seeing it."

Still, despite Tan's assertion that a change of reporting at the top has made a major difference to Intel's business, he was also quick to credit the teams underneath him—along with Intel's relationship with its customers.

"It's a lot of hard work, it's a lot of teamwork, a lot of talent I brought on board to help," said Tan. "It's a service business. You need to build the trust with the customer."

On the topic of Intel's customers, Tan also said the company would have risk production of its upcoming 14A process underway in 2028, and volume production in 2029. "It will be the same time as TSMC," said Tan. "So that is a major, major breakthrough."

It certainly seems so. Tan's hands-on approach looks to have brought the company success that seemed a million miles away from reality in previous years, and with Elon Musk's Terafab project lining up for its 14A tech, it's looking like a bright future for Intel if it can keep bringing in the wins. Not to mention its partnership with Nvidia. I still can't quite get my head around that, can you?

A new report claims that Intel is pushing its client PC customers to buy its latest CPUs built on 18A silicon. It's said the older Intel 7 node is increasingly being reserved for more profitable server chips and even Intel CPUs made in TSMC's fabs will be in short supply.

Nikkei Asia says Intel's move is in response to "surging" demand for server CPUs thanks to good old AI. Consequently, "Intel is asking its leading PC partners, including those in the US, China and Taiwan, to increase their use of central processing units built with the company's 18A production process," the report says.

As Nikkei Asia notes, the older Intel 7 process node is still used in a wide range of CPUs for PCs and servers. On the desktop, that includes 13th Gen Alder Lake and 14th Gen Raptor Lake CPUs, like, say, the Core i7-14700K.

But the report claims Intel has told PC builders that such chips are "unlikely to see additional supply allocations or accelerated shipment schedules." What's more, that constraint also applies to CPUs which are largely manufactured by TSMC rather than on Intel 7, such as the latest Arrow Lake desktop chips.

All are apparently being deprioritised in favour of either using Intel 7 to knock out more profitable server CPUs, or otherwise upselling PC builders to CPUs based on its latest 18A node.

One executive at a PC maker told Nikkei Asia, "We recently placed an additional order for 100 Intel 7 CPUs. We only got 30, and 10 of them were built with 18A technology. We were told if we don't take the 18A CPUs, they would be given to other PC makers." Yikes.

There are a few interesting implications in all of this. First, in the immediate term, it doesn't imply anything great for desktop CPUs. That's because all of Intel's current desktop chips are either Intel 7 or TSMC silicon. There are no 18A desktop CPUs, not until Nova Lake arrives. That won't be until later this year at the earliest, and 2027 is more likely in terms of wide retail availability.

Frankly, if this report is accurate, it makes Intel's recent launch of Arrow Lake Plus chips pretty odd. The whole point of Arrow Lake Plus, which includes the Intel Core Ultra 7 270K Plus and Core Ultra 5 520K Plus desktop CPU models, is to reposition Intel's desktop CPUs with a better value proposition. But what's the point if you can barely buy them?

Then there's the question of what this might do to laptop pricing. It might not be all bad news given that Intel has recently announced its new budget-specced Wildcat Lake chips. They're essentially cut-down versions of Intel's 18A-based Panther Lake CPUs, with up to two Performance and four Efficient cores. Panther Lake offers double those core counts.

Problem is, you wouldn't want just two Performance cores in a gaming laptop. So, that means high-performance portables will be forced to use more expensive Panther Lake CPUs given Arrow Lake H and HX mobile and older Raptor Lake mobile chips are presumably now hard to source.

The report also has positive implications, if true, for Intel's 18A technology. After all, if yields on 18A were terrible, Intel wouldn't be able to push PC builders to buy 18A CPUs over Intel 7 CPUs.

Anywho, it will be interesting to see how this all plays out. The whole thing with the implications of this report being that Intel's recently launched Arrow Lake Plus CPUs are being heavily de-prioritised is pretty odd, that's for sure. As I write these words, it's no problem buying an Intel Core Ultra 7 270K Plus on Newegg. So, is that going to change? Watch this space.

In the same way that top-fuel dragsters have no bearing on how we use everyday cars, overclocking the twangers off a CPU using extreme cooling methods doesn't have much practical use, but it's super fun to watch. The case in point here is that there's a new world record for the highest clock speed achieved by any desktop processor: 9,206 MHz with an Intel Core i9 14900KF.

You can check out the details of the record over at HWBot, but the first thing of note here is that the chip used isn't a 14900KS, the typical choice of chip for record breakers. That Raptor Lake processor was Intel's last hurrah in its previous era of 'clock speed beats all and power usage be damned'.

That CPU has a peak boost clock of 6.0 GHz by default, so it's not hard to see why it was used to achieve the previous record of 9,118 MHz. Admittedly, top overclockers never use the chips in a default configuration, and that's no different for this new record. Firstly, all the E-cores were disabled, followed by HyperThreading (i.e. simultaneous multithreading), and what appears to be a single P-core, too.

The overclocking team also used a highly-tweaked DRAM configuration of dual-channel DDR5-5792, with timings of 32-47-41-77. Oh, and a lot of liquid nitrogen. Not a couple of litres but a comically huge industrial vat of the stuff. You can watch the team practically vanish behind a water vapour cloud in a Bilibili video on the record attempt.

Only one P-core is actually fired up to the dizzying heights of 9.2 GHz, but even so, it's still one heck of an achievement. What makes it stand out to me, though, is that the fact that it's not all that far off 10 GHz, and when it comes to Intel's processors, that's somewhat of a legendary goalpost.

Back in 2000, when Intel launched the Pentium 4 processor, the chip giant was suggesting that 10 GHz processors could potentially become a reality within the space of five or so years, because of the rapid advances being made in photolithography and wafer process nodes.

Intel employed all kinds of tricks to ramp up the clock speeds of its Netburst architecture, but ultimately ran into a massive power issue at around 3.8 to 4.0 GHz. At those speeds, the transistors were leaking so much energy that no amount of additional voltage or cooling could get the nanoscale switches to operate correctly.

Ultimately, Intel would abandon Netburst and its quest for 10 GHz in favour of a slower but more capable architecture that would come to be known as Core, and as the saying goes, the rest is history. Not that Team Blue would entirely give up chasing clock speeds, as Raptor Lake amply demonstrates.

With Arrow Lake and presumably Nova Lake, Intel has stepped back once more (power being once again a major issue), so we're unlikely to see top overclockers breaking the new world record with one of those chips. Still, you never know: Like mountain climbers of old, staring at the terrifying heights of Everest, K2, or Annapurna, that ten gig mark is now only 800 MHz away.

That's just 9% higher than what's been achieved. Just. Only. We're so close, but still oh so far away.

After years of outrageous power levels, self-destructing chips, and underwhelming laptop options, Intel's pair of Arrow Lake Refresh processors—the Core Ultra 5 250K Plus and Core Ultra 7 270K Plus—are so good that almost all of Team Blue's recent failings can be forgiven. And in the case of the 270K Plus, it's received one more special 'award': it can now be picked up as a delidded option from Thermal Grizzly.

The process of removing the IHS or integrated heat spreader from a CPU (aka delidding) isn't something that most PC users would consider. Partly because you need specialised hardware to do this, and there's a distinct risk of permanently damaging the chip in doing so, but it's mostly because there's no major advantage in doing so when using the processor in stock form.

However, for the overclocking crowd, it's a bit of a must because allowing a cooler to have direct contact with the CPU die (or dies, in the case of AMD chips), can potentially get a serious reduction in operating temperature. This is turn lets you crank up the voltages to achieve much higher clock speeds.

If that sounds of interest, but you don't want the hassle of delidding, you can always head over to Thermal Grizzly and buy a pre-popped processor from there. Naturally, since these are all done by hand and carefully prepared for use with liquid metal, they don't come cheap.

Actually, that's not true, because TG has just added the Core Ultra 7 270K Plus to its list of delidded options, and thanks to Intel's low MSRP for the CPU, a bare butt KP chip will only set you back a fraction over $525.

That's cheaper than a stock Core Ultra 9 285K (currently $560 at Amazon) and even without overclocking, the 270K Plus is better than its bigger brother in games, due to its higher internal bus clocks and other tweaks. Throw some serious overclocking into the mix, and hey presto! You now have the most powerful desktop CPU Intel has made in years.

AMD's Ryzen 7 9800X3D will still be better for gaming, and any AM5 gaming PC you buy or put together now will readily take a next-gen Ryzen with a BIOS upgrade. Intel's LGA 1851 socket only supports the Core Ultra 200S chips and nothing else.

But for sheer all-round brilliance, especially if you game and do a lot of content creation workloads, the Core Ultra 7 270K Plus is very hard to beat for the money. Only the considerably more expensive Ryzen 9 9950X3D and 9950X3D2 Dual Edition are better. I should know because I've not only tested every single one of these chips, but I also use a 270K Plus in my own rig now.

AMD now owns fully 38.1% of overall market revenues for x86 CPUs according to new data from Mercury Research (via HotHardware). That's up from 35.4% for the previous quarter. But AMD's desktop CPU market share actually fell.

AMD is doing particularly well in the server market, where it won 46.2% of revenues and 33.2% of unit sales. That means AMD server CPUs, on average, are selling for quite a bit more than Intel CPUs. Those figures are up by around 3-4% on the previous quarter.

AMD is also making up ground in laptops. Mercury's latest data puts AMD at 28.9% of overall revenues and 28.3% of unit sales. Its figures from the previous quarter are 24.9% of revenues and 26% of unit sales.

Desktop CPUs represent the lone segment where AMD is slipping a little. For the latest quarter, Mercury has AMD pegged at 37.6% of revenues and 33.2% of unit sales. That's down from 42.5% and 36.4%, respectively, for the previous quarter, which were all time highs for AMD.

Intel is, of course, AMD's only direct competitor in specifically the x86 CPU market, as opposed to the broader CPU market, which includes the likes of Arm and RISC-V chips. And desktop aside, it all looks like pretty good news for AMD.

That said, it's worth noting that Intel remains pretty dominant. It still owns 70% of the overall CPU market by unit sales. If nothing else, that's interesting in terms of what it says about how much inertia there is in the CPU market. It's not exactly a pure meritocracy.

After all, AMD has had very competitive CPUs across most market segments, and arguably superior in some, for the better part of a decade now. And yet it's still outsold more than two times over by Intel.

In some segments, that makes sense. AMD has a relatively limited offering in the mobile CPU market compared to Intel's sprawling litany of laptop CPU SKUs. But on the desktop, well, if that was purely meritocratic, Intel wouldn't have over 65% market share, that's for sure.

You'd think business customers building server farms would be pretty hard nosed about it all and go on pure merit. And AMD has indeed done better in the server market than any other segment.

But AMD's EPYC server CPUs really are very good and, I would think, most truly independent observers would argue that they deserve to be selling better than Intel's chips.

Of course, the supply and demand dynamics involved are pretty complex. AMD can't just make as many chips as it fancies on any give day, month or year. It farms out manufacture, mostly to Taiwanese fab TSMC, and that means signing forward contracts and, to an extent, being limited by TSMC's capacity.

That said, while Intel still does manufacture many of its own CPUs, it too needs to plan a long way into the future in terms of fab capacity and so on. All of which means, even if AMD suddenly offered unambiguously the best CPUs in every single segment tomorrow, it wouldn't be straight forward for that to be instantly reflected in sales figures.

Intel has returned to F1 with a new partnership with McLaren. The chip company is providing Xeon and Core Ultra chips to the team in exchange for a decal on the side of the car that won both the Drivers and Constructors championship in 2025.

Intel says its chips will be provided for computational fluid dynamics, aerodynamic analysis, vehicle-dynamics simulation, and real-time decision systems, and race strategy analytics. So, all the important stuff, then. F1 teams are data companies these days, so a major technical partner is a must-have to do much on the track.

We have the usual press release comment from both companies' top brass. Intel CEO Lip-Bu Tan said:

"Formula 1 racing and IndyCar are some of the ultimate proving grounds for high-performance computing. Intel is proud to be McLaren Racing’s compute partner, and to be part of a team that thrives on precision, speed, and innovation. Together, Intel and McLaren will push the boundaries of what’s possible, transforming data into competitive advantage at every turn."

And Zak Brown, McLaren Racing CEO, says:

"Performance in IndyCar and Formula 1 racing is driven by technology, and partnering with Intel strengthens our ability to innovate at scale. Intel has already been an important part of our technology ecosystem, and their leadership in computing will play a critical role in how we design, build, and race our cars. We’re excited to deepen this relationship even further."

Pretty standard word salad. But it is an interesting move from Intel, which has been going through a tough time in recent years. It has more recently has found some success with some products, such as Panther Lake, and investment from the US government and Nvidia with increased interest in US-made chips, which few others can provide. But otherwise it's still on a path to reclaiming its lost leadership position. Perhaps getting back into F1 is a sign of confidence from the company, even if it's still not the money-making machine it once was, or a plan to be seen to be confident again.

The company has previously dabbled in F1 sponsorship, with a big badge slapped on the side of the BMW Sauber team in the early 2000s. This new McLaren sponsorship will also extend to its IndyCar team and its sim racing team in the future.

McLaren also has an ongoing partnership with Google, which is now tipped to be all about AI, because of course it is.

AMD said it would "simply be too expensive" and that "games would not benefit." It was, of course, talking about why it had never launched a dual CCD Ryzen desktop processor with 3D V-Cache on both chiplets. There was no technical barrier, just a matter of pointlessness. And yet here we are, with the Ryzen 9 9950X3D2 Dual Edition. Yours for a mere $899.

For that princely sum of money, you're basically getting a Ryzen 9 9950X3D with 192 MB of L3 cache instead of 128 MB, a tiny 100 MHz knocked off the peak boost clock, and an extra 30 W slapped on to the TDP. The 9950X3D launched at $699, though you can often pick one up these days for around $60 less.

Is it worth spending $200 to $260 more for the Dual Edition? Taken at face value, the answer would appear to be an emphatic 'no'. AMD not only warned that it would be pointless for PC gaming, but it also avoided the topic entirely for the official launch of the chip.

This isn't to say it's a bad processor—far from it, in fact—nor am I suggesting that all that extra cache is useless (check out some of the benefits the Dual Edition can offer scientific workloads and developers in Phoronix's tests). It's just that for high-end PC gamers or modest content creators, buying a Ryzen 9 9950X3D2 Dual Edition would seem to be just throwing money away for no reason whatsoever.

That said, the dual cache nature of the 9950X3D2 does help to avoid one issue that has plagued all of AMD's X3D processors that have twin CCDs. The likes of the 7950X3D, 9900X3D, and 9950X3D rely on drivers to handle thread scheduling in games, so that all of the crucial threads get processed on the core chiplet that has the extra cache.

It didn't work all that well for the Ryzen 7000-series chips, though it was mostly solved for the 9000-series. Since the Dual Edition has extra cache on both CCDs, thread scheduling becomes less of an issue (though it's still important).

Plus, the CCD clocks are no longer asymmetrical. For example, with the Ryzen 9 9950X3D, CCD0 peaks at 5.4 GHz, whereas CCD1 (without the cache) can hit 5.7 GHz. The 9950X3D2, on the other hand, can reach 5.6 GHz with either chiplet.

Does that actually make any difference in our battery of CPU benchmarks? Let's find out.

PC Gamer test PC specs

MSI MEG X870E Godlike X Edition | 32 GB Lexar Thor OC DDR5-6000 CL32 | Asus ROG Strix LC III 360 ARGB | Zotac GeForce RTX 4070 | Corsair MP700 2 TB | Gigabyte Aorus Elite 1000 W Platinum | Thermal Grizzly Der8enchtable

Gaming performance

Gaming performance

Cyberpunk 2077 (1080p RT Ultra + DLSS Balanced)Baldur's Gate 3 (1080p Ultra)Homeworld 3 (1080p Epic)Metro Exodus Enhanced Ed. (1080p High)3DMark Time Spy

Avg FPS

1% Low FPS

AMD Ryzen 9 9950X3D2 Dual Edition

121

88

AMD Ryzen 9 9950X3D

111

67

AMD Ryzen 7 9800X3D

112

76

Intel Core Ultra 9 285K

118

95

Intel Core Ultra 7 270K Plus

119

98

037.575112.5150

Cyberpunk 2077 (1080p RT Ultra + DLSS Balanced) Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	121 Avg FPS, 88 1% Low FPS
AMD Ryzen 9 9950X3D	111 Avg FPS, 67 1% Low FPS
AMD Ryzen 7 9800X3D	112 Avg FPS, 76 1% Low FPS
Intel Core Ultra 9 285K	118 Avg FPS, 95 1% Low FPS
Intel Core Ultra 7 270K Plus	119 Avg FPS, 98 1% Low FPS

Avg FPS

1% Low FPS

AMD Ryzen 9 9950X3D2 Dual Edition

146

82

AMD Ryzen 9 9950X3D

147

84

AMD Ryzen 7 9800X3D

146

76

Intel Core Ultra 9 285K

108

67

Intel Core Ultra 7 270K Plus

113

66

037.575112.5150

Baldur's Gate 3 (1080p Ultra) Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	146 Avg FPS, 82 1% Low FPS
AMD Ryzen 9 9950X3D	147 Avg FPS, 84 1% Low FPS
AMD Ryzen 7 9800X3D	146 Avg FPS, 76 1% Low FPS
Intel Core Ultra 9 285K	108 Avg FPS, 67 1% Low FPS
Intel Core Ultra 7 270K Plus	113 Avg FPS, 66 1% Low FPS

Avg FPS

1% Low FPS

AMD Ryzen 9 9950X3D2 Dual Edition

115

64

AMD Ryzen 9 9950X3D

101

54

AMD Ryzen 7 9800X3D

117

55

Intel Core Ultra 9 285K

97

53

Intel Core Ultra 7 270K Plus

101

57

037.575112.5150

Homeworld 3 (1080p Epic) Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	115 Avg FPS, 64 1% Low FPS
AMD Ryzen 9 9950X3D	101 Avg FPS, 54 1% Low FPS
AMD Ryzen 7 9800X3D	117 Avg FPS, 55 1% Low FPS
Intel Core Ultra 9 285K	97 Avg FPS, 53 1% Low FPS
Intel Core Ultra 7 270K Plus	101 Avg FPS, 57 1% Low FPS

Avg FPS

1% Low FPS

AMD Ryzen 9 9950X3D2 Dual Edition

152

95

AMD Ryzen 9 9950X3D

148

91

AMD Ryzen 7 9800X3D

148

91

Intel Core Ultra 9 285K

149

79

Intel Core Ultra 7 270K Plus

143

79

050100150200

Metro Exodus Enhanced Ed. (1080p High) Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	152 Avg FPS, 95 1% Low FPS
AMD Ryzen 9 9950X3D	148 Avg FPS, 91 1% Low FPS
AMD Ryzen 7 9800X3D	148 Avg FPS, 91 1% Low FPS
Intel Core Ultra 9 285K	149 Avg FPS, 79 1% Low FPS
Intel Core Ultra 7 270K Plus	143 Avg FPS, 79 1% Low FPS

CPU index score

Overall index score

AMD Ryzen 9 9950X3D2 Dual Edition

20263

17590

AMD Ryzen 9 9950X3D

17099

17084

AMD Ryzen 7 9800X3D

16116

17052

Intel Core Ultra 9 285K

19293

17430

Intel Core Ultra 7 270K Plus

20252

17548

07,50015,00022,50030,000

3DMark Time Spy Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	20263 CPU index score, 17590 Overall index score
AMD Ryzen 9 9950X3D	17099 CPU index score, 17084 Overall index score
AMD Ryzen 7 9800X3D	16116 CPU index score, 17052 Overall index score
Intel Core Ultra 9 285K	19293 CPU index score, 17430 Overall index score
Intel Core Ultra 7 270K Plus	20252 CPU index score, 17548 Overall index score

Starting with our usual Cyberpunk 2077 benchmark, the Ryzen 9 9950X3D2 Dual Edition drops a surprise right off the bat. Thanks to its 5.6 GHz boost clock, it nips ahead of the 9950X3D and 9800X3D by around 8% on average, but the real shock is just how much better the 1% low frame rates are.

That's possibly down to thread scheduling messing things up for the other chips, but whatever the reason is, it's an encouraging start for the hugely expensive processor. Things fall more in line with expectations with Baldur's Gate 3, but with Homeworld 3, it pulls another rabbit out of the hat.

Our Metro test has reached the end of its usefulness for separating high-end CPUs, as they're all pretty much the same, but repeated tests confirmed that the 9950X3D2 Dual Edition was indeed marginally faster than anything else we've tested of late.

Separately, I did some performance tests with Factorio and got identical results to those for the 9950X3D; the figures aren't shown because it's a new version of our old Factorio benchmark, and it will take some time to go through all of the old chips and retest them.

There's one last surprise in the form of the CPU result in the 3DMark Time Spy benchmark. Traditionally, this is a test that has never particularly favoured AMD processors, but the 9950X3D2 Dual Edition counters that quite remarkably.

So it would seem that having both CCDs identical in terms of clock speeds and cache levels, thus avoiding any thread scheduling headaches, is of some benefit in PC gaming.

Content creation performance

Content creation performance

Cinebench 2024Blender 4.2.07zipHandbrake 1.8.1

Single-core index score

Multi-core index score

AMD Ryzen 9 9950X3D2 Dual Edition

141

2484

AMD Ryzen 9 9950X3D

139

2347

AMD Ryzen 7 9800X3D

133

1307

Intel Core Ultra 9 285K

145

2383

Intel Core Ultra 7 270K Plus

143

2435

07501,5002,2503,000

Cinebench 2024 Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	141 Single-core index score, 2484 Multi-core index score
AMD Ryzen 9 9950X3D	139 Single-core index score, 2347 Multi-core index score
AMD Ryzen 7 9800X3D	133 Single-core index score, 1307 Multi-core index score
Intel Core Ultra 9 285K	145 Single-core index score, 2383 Multi-core index score
Intel Core Ultra 7 270K Plus	143 Single-core index score, 2435 Multi-core index score

AMD Ryzen 9 9950X3D2 Dual Edition

208

AMD Ryzen 9 9950X3D

201

AMD Ryzen 7 9800X3D

101

Intel Core Ultra 9 285K

174

Intel Core Ultra 7 270K Plus

179

075150225300

Blender 4.2.0 Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	208
AMD Ryzen 9 9950X3D	201
AMD Ryzen 7 9800X3D	101
Intel Core Ultra 9 285K	174
Intel Core Ultra 7 270K Plus	179

Compressing (GIPS)

Decompressing (GIPS)

AMD Ryzen 9 9950X3D2 Dual Edition

216

285

AMD Ryzen 9 9950X3D

212

273

AMD Ryzen 7 9800X3D

132

141

Intel Core Ultra 9 285K

173

195

Intel Core Ultra 7 270K Plus

179

200

075150225300

7zip Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	216 Compressing (GIPS), 285 Decompressing (GIPS)
AMD Ryzen 9 9950X3D	212 Compressing (GIPS), 273 Decompressing (GIPS)
AMD Ryzen 7 9800X3D	132 Compressing (GIPS), 141 Decompressing (GIPS)
Intel Core Ultra 9 285K	173 Compressing (GIPS), 195 Decompressing (GIPS)
Intel Core Ultra 7 270K Plus	179 Compressing (GIPS), 200 Decompressing (GIPS)

AMD Ryzen 9 9950X3D2 Dual Edition

135

AMD Ryzen 9 9950X3D

134

AMD Ryzen 7 9800X3D

92

Intel Core Ultra 9 285K

121

Intel Core Ultra 7 270K Plus

123

037.575112.5150

Handbrake 1.8.1 Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	135
AMD Ryzen 9 9950X3D	134
AMD Ryzen 7 9800X3D	92
Intel Core Ultra 9 285K	121
Intel Core Ultra 7 270K Plus	123

Alas, the same cannot be said for our standard content creation benchmarks. While the 9950X3D2 does take a small lead over the 9950X3D, it's unlikely to be due to the extra cache. Instead, it's a case of the symmetrical and higher boost clocks that are helping out here, over the two other X3D chips shown.

But even small gains could be worth taking advantage of to some users, with the time saved accumulating over months of work offsetting the higher cost of the processor. However, countering that argument is the fact that there is an extra cost to using the Ryzen 9 9950X3D2 Dual Edition, and it's all about power.

Power and thermals

Power and thermal performance

Average CPU power consumption in Baldur's Gate 3Average CPU package temperature in Baldur's Gate 3CPU power consumption in Cinebench 2024 multicore

AMD Ryzen 9 9950X3D2 Dual Edition

108

AMD Ryzen 9 9950X3D

108

AMD Ryzen 7 9800X3D

67

Intel Core Ultra 9 285K

108

Intel Core Ultra 7 270K Plus

111

037.575112.5150

Average CPU power consumption in Baldur's Gate 3 Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	108
AMD Ryzen 9 9950X3D	108
AMD Ryzen 7 9800X3D	67
Intel Core Ultra 9 285K	108
Intel Core Ultra 7 270K Plus	111

AMD Ryzen 9 9950X3D2 Dual Edition

69

AMD Ryzen 9 9950X3D

64

AMD Ryzen 7 9800X3D

51

Intel Core Ultra 9 285K

61

Intel Core Ultra 7 270K Plus

67

020406080

Average CPU package temperature in Baldur's Gate 3 Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	69
AMD Ryzen 9 9950X3D	64
AMD Ryzen 7 9800X3D	51
Intel Core Ultra 9 285K	61
Intel Core Ultra 7 270K Plus	67

Avg package power (W)

Peak package power (W)

AMD Ryzen 9 9950X3D2 Dual Edition

245

245

AMD Ryzen 9 9950X3D

197

200

AMD Ryzen 7 9800X3D

109

116

Intel Core Ultra 9 285K

235

243

Intel Core Ultra 7 270K Plus

240

253

075150225300

CPU power consumption in Cinebench 2024 multicore Data

Product	Value
AMD Ryzen 9 9950X3D2 Dual Edition	245 Avg package power (W), 245 Peak package power (W)
AMD Ryzen 9 9950X3D	197 Avg package power (W), 200 Peak package power (W)
AMD Ryzen 7 9800X3D	109 Avg package power (W), 116 Peak package power (W)
Intel Core Ultra 9 285K	235 Avg package power (W), 243 Peak package power (W)
Intel Core Ultra 7 270K Plus	240 Avg package power (W), 253 Peak package power (W)

For the 9950X3D2, AMD increased the thermal design power limit (TDP) to 200 W, compared to 170 W for the 9950X and 9950X3D. Interestingly, the PPT (package power tracking) limit is 250 W, which is only 20 W more than those other processors.

As you can see in our Baldur's Gate 3 benchmark, the Dual Edition chip doesn't come anywhere near to using that amount of power, but it's still a good deal more than the Ryzen 7 9800X3D, which gets by with a mere 67 W. AMD's full-spec dual-CCD models demand a fair bit more juice, but that's to be expected having twice as many core chiplets.

The downside to having all those cores and cache is that the Ryzen 9 9950X3D2 is quite a hot little chip. In games, it's not so bad, but when fully loaded up in Cinebench 2024, consuming over 240 W of power in the process, the CPU temperature reached 94 °C.

In fact, the cores frequently bounced off their thermal limits in that benchmark, and I suspect a more substantial cooling system than the Asus ROG Strix LC III 360 ARGB would help the processor to achieve an even better score. It doesn't do the chip any harm running at that temperature, and Ryzen 7000-series users will know exactly what I mean by that.

But if those kinds of numbers leave you uncomfortable, you're perhaps going to need a 420 mm AIO cooler or even a dedicated water loop to keep the chip's temperatures right down.

Overall verdict

With my brief time with the Ryzen 9 9550X3D2 Dual Edition, I've come to the conclusion that AMD wasn't entirely right by saying dual 3D V-Cache is of no benefit to PC gaming. There are some scenarios where it does help, though it might be due to the symmetrical nature of the chip rather than the extra cache, and naturally, your gaming experience will only improve if it's currently limited by CPU, rather than GPU, performance.

That said, AMD was absolutely right when it said that it would be too expensive. While $899 isn't the priciest processor that Team Red has ever released for desktop PCs (the Athlon FX-62 from 2006 was over $1,000), it's arguably too much of a price hike over the 9950X3D to be worth it for a gaming PC.

Not that this will put off some gamers from forking out a huge sum of money just to have the most powerful gaming CPU in their rigs. Everyone else should just stick with the Ryzen 7 9800X3D or, if content creation is more important to you, just go with the Ryzen 9 9950X. Just think of all the DRAM you could buy with that extra cash. Hmm, maybe not.

I love a supercomputer reveal, and this one's a little different than most. China's National Supercomputing Center in Shenzhen has announced its LineShine system, a machine that aims to deliver 2 exaflops of performance once it's fully constructed.

That's an impressive figure, and would make LineShine one of the fastest exascale supercomputers in the world (via HPCWire). This one, though, is planned to be an all-CPU system—in contrast to virtually all of the others on the Top500 list, which make use of AMD or Nvidia GPUs to accelerate proceedings.

This will be a supercomputer of sheer processor-induced heft, and some of the stats are jaw-dropping. The first construction phase will link 100 Huawei Kunpeng servers together, with 12,800 cores spread between them. The second phase, according to reports, will involve the addition of an extra 47,000 individual CPUs spread across 92 compute cabinets.

A pre-print of a paper entitled "Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials" gives us some more potential specs of China's new toy. The paper says that the total system will be made of 20,480 computing nodes, each equipped with two Armv9-based LX2 processors.

Each of those LX2 CPUs has two compute dies with 304 cores in total, which run alongside eight 32 GB HBM stacks. Every node is linked by a "LingQi high-speed network with dual-plane multi-rail fat-tree topology, offering 1.6 Tb/s bandwidth per node."

I know what at least some of those terms mean. Anyway, if these specs and performance figures are accurate, the LineShine would take its place as the fastest all-CPU-powered supercomputer ever built—although El Capitan currently holds the record for the fastest overall in the Top500 listings, with a 2.74+ exaflop Rpeak result.

As for its purpose? Apparently, it will deliver "internationally leading performance in large-scale applications such as molecular dynamics, fluid simulation, life sciences, and AI large-scale model training and promotion." Y'know, supercomputer stuff. A bit boring, isn't it? Imagine the Geekbench score, folks. Surely it's worth a run just for giggles?

Intel's Enthusiast Channel VP and GM Robert Hallock has told PC Games Hardware that we shouldn't underestimate the impact software improvements could have on gaming performance. In other words, that perhaps our focus as gamers shouldn't be so laser-focused on hardware:

"I truly believe that the general PC gaming market and especially enthusiasts... are significantly underestimating the importance of software to the PC experience, like really, really seriously."

"There is no game on earth that is as fast as it's going to be purely through hardware. That doesn't exist anymore. That used to be the case in 2010, 2015. That is not how gaming works anymore."

He doesn't deny that hardware is an important factor, but wants to remind us that it's not the only one:

"Yes, you can make the game faster with a faster piece of hardware, but there's always going to be 10, 20, 30% performance hidden behind the fact that that game was just not optimized for your CPU."

In one sense, this shouldn't come as a surprise, because in pretty much any CPU review we've done we see that some games perform better on AMD chips and some on Intel ones. Often, game devs will simply optimise their games for whichever CPUs they can spend the most time testing them on.

While the 10–30% figure is referring to game optimisations and, more specifically, console ports, games are only one part of the software picture. Apparently Hallock also couched some of the discussion in terms of things like CPU scheduling, which is an operating system task that helps direct the right processes to the right CPU cores and threads.

Scheduling and similar fine grain CPU management has been incredibly important for Intel most especially since the company decided to split its cores into P-cores and E-cores (and, more recently, LPE-cores). And while the company got rid of Hyper-Threading for a while, multithreading should be back in future, given CEO Lip-Bu Tan admitted. "Moving away from SMT put us at a competitive disadvantage." Which means optimised CPU scheduling will become even more important.

Given all this, Hallock claims that gamers asking for Intel to focus less on software and more on hardware is sort of like saying "hey, just leave 20% performance behind."

When thinking about software and hardware improvements, though, I can't help but think we should be echoing Tulio and Miguel on their Road to El Dorado: Both. Both is good.

Intel has reported its first-quarter 2026 financial results, and they read as something of a victory lap. It wasn't too long ago the company appeared to be down on its knees, but CEO Lip-Bu Tan, now just over a year into his posting, struck a defiant tone in the opening remarks.

"Intel is now a very different company than when I first joined over a year ago," said Tan. "We have taken, and continue to take, deliberate steps to rebuild Intel into a more competitive and more profitable company.

"A year ago, the conversation about Intel Corporation was about whether we could survive. Today, it is about how quickly we can add manufacturing capacity and scale our supply to meet enormous demand for our products," the Intel chief continued. "This is a fundamentally different company today, and we still have a lot of work ahead."

The company reports first-quarter revenue of $13.6 billion, which CFO David Zisner notes, "would have been meaningfully higher, but demand continues to outpace our growing supply." AI-driven businesses are said to now represent 60% of Intel's revenue, and grew by 40% year-over-year.

Data center and AI revenue was reported as $5.1 billion, a seven percent sequential increase and 22% year-over-year. ASIC revenue was reported to have nearly doubled compared to the same quarter in 2025, while client computing group revenue is up by $300 million quarter-over-quarter, representing 33% of total revenue.

Zisner notes that the client computing revenue increase was down to "improved product margin, sales of previously reserved inventory, better 18A yields, and lower operating expenses." It's AI server-fuelled boom time in the processor industry, although reports have indicated that some Intel chips are already in short supply, and that the industry is waiting for more 18A offerings to plug the gap.

Speaking of 18A, the company reports "better" yields and "great progress" for its previously-struggling chip process, although Intel Foundry's operating loss was reported as $2.4 billion. That's a $72 million improvement over the previous quarter, and something Zisner explains as being "mostly offset by increased operating expenses associated with an intentional step-up in Intel 14A investments."

CEO Lip-Bu Tan said that the "maturity, yield, and performance" of Intel's in-development 14A process is already outpacing 18A at a similar point in time, and that the company is continuing to develop process design kits (PDKs) with "multiple customers".

One of which is Tesla. CEO Elon Musk recently announced that Intel's 14A process will be used to make AI chips for the company, as part of a partnership with Intel for his Terafab project—which represents a huge potential boon for Intel. In fact, Musk was given special mention in Lip-Bu Tan's opening remarks:

"Elon and I share a strong conviction that global semiconductor supply is not keeping pace with the rapid acceleration in demand," said Tan. "We are excited to explore innovative ways to refactor silicon process technology, looking for unconventional ways to improve manufacturing efficiency that will eventually lead to a dynamic improvement in the economics of semiconductor manufacturing."

And there's always that Nvidia/Intel collaboration to consider, too. Rumours have suggested that the team green/team blue efforts will be named "Serpent Lake", and potentially be a combination of Nvidia's next-gen Rubin GPU tech and Intel's own silicon. The exact details are mostly speculation at this point, though, and Nvidia was only tangentially mentioned in the earnings call.

Still, Intel's certainly been busy racking up partnership and collaboration agreements in the past year, and by the sounds of it, the AI boom has played a significant part in its impressive performance, too. The company's share price jumped by 20% after the earnings and forecasts were revealed, and that's likely to be news that makes Lip-Bu Tan, and Intel's investors, very happy indeed.

Earlier this month, Intel teamed up with Elon Musk to "help refactor silicon fab technology" for Musk's ambitious Terafab project. Now, in a Tesla earnings call, the tech mogul has said he will use the company's upcoming 14A node to make chips for this project.

Terafab is Musk's (one might say) bold venture into making his own chips for xAI, Tesla, and SpaceX to power cars, robots, and data centres in space. It's apparently necessary, given the world seemingly doesn't produce enough chips for these companies' needs. Which isn't as absurd as it might sound, to be honest, as CPUs are reportedly already feeling a shortage squeeze.

There's been hope that one solution to this shortage might be Intel 18A, the process that the previous Intel CEO Pat Gelsinger said he'd "bet the whole company on." We have 18A chips already on the market in the form of Panther Lake mobile processors, and later this year, we'll be seeing them again in Nova Lake for desktop PCs.

14A, on the other hand, is still a couple of years out. Previously, Intel had said that it's not in any hurry to get the process node into full swing "until we know that we have the customers that will accept that demand." Well, how's Tesla as a first customer for you? That should be enough to get the ball rolling, if Musk's Terafab goals are anything to go by.

When he reported on Intel's partnering up with Musk, our Nick hinted that Intel's Lip-Bu Tan would most likely have had foundry and packaging wrapped up in part of that agreement, and this has turned out to be true. That's because Intel didn't exactly have the largest list of external customers for its foundries.

Recently, however, Intel has opened the door to having external customers for 18A, and now with Musk on board for Terafab, we have a first customer penned in for 14A.

Musk expresses some optimism towards the timing of the process:

"We plan to use Intel's 14A process, which is state-of-the-art and in fact not yet totally complete. But given that by the time Terafab scales up, 14A will be probably fairly mature or ready for prime time. 14A seems like the right move. And we have a great relationship with Intel. A lot of respect for the CEO, the CTO, and the new team there. So we think it's going to be a great partnership."

All good news for Intel, then, and presumably for its shareholders too. The company's market cap is now higher than it's been in over 25 years, thanks largely to the Terafab deal plus a multi-year deal with Google.

It all makes a refreshing change from the previous few years of stagnation and seeming issue after issue for Intel. And speaking of refreshes, the latest Arrow Lake Core Ultra 200-S 'Plus' refresh is pretty great, too. Hopefully, things continue in that direction; lord knows the PC hardware market needs as many wins as it can get right now.

When even a mid-tier pre-built gaming rig costs further above a grand than anyone would like, there's no ignoring it: PC gaming is an extremely expensive hobby. Thankfully, Intel may be moving to make high-tier performance just a little more accessible for those on a budget.

Robert Hallock, the VP & GM of Intel's enthusiast channel business, recently hinted that the company would offer more overclockable CPUs at lower price points. "What you will see is more and more unlocked SKUs over time. That is the goal," Hallock said during an interview with PC Games Hardware.

"[Overclocking] should not be a feature that is exclusively reserved for the people paying the most amount of money," Hallock explains. "Not everyone can afford the most amount of money [...] and that doesn't make them any less of an enthusiast than the person who can spend $500 on a CPU. They are still PC enthusiasts, and they deserve the same level of features, and that is what we intend to deliver in our roadmap."

In other words, although affordable Intel CPUs with overclocking haven't traditionally been available for those on a budget, Intel is looking to change that in the future. In current and previous hardware generations, overclocking has only been possible for CPUs with a 'K' designator, which are usually higher-end, and more expensive chips than their non-K designator counterparts.

That said, it's worth noting the Intel Core Ultra 5 245K is definitely not the tippy top of the CPU tier list, but it can also be overclocked. Perhaps Intel plans on K chips even lower down than the Core Ultra 5 ones being part of its future hardware line-ups.

Now, whether you'd even want to overclock low-end chips will be another question entirely. Eking out the most performance only makes sense if power and thermals permit it (and, obviously, if there's a big enough gain to make it worth it).

In the not so distant past, Intel was the king of the budget overclocking chip. You could pick up an affordable processor and, even if it had locked down multipliers for whatever reason, you could tweak the front side bus and mess around with it to your heart's content; you could almost guarantee an extra 1 GHz on a ton of old Intel processors. It's great to see at least some folk at Intel talking about loosening the artificial restrictions on its chips, and maybe going back to those good ol' days of CPU overclocking.

We'll have to wait and see what actually pans out in upcoming Intel generations—Nova Lake is what's next, and it's expected to launch later this year, though Robert Hallock has not yet specified whether this particular generation will get lower-end chips capable of OCing.

I know that plenty of PC gamers are still having a grand old time with almost decade old hardware (to the GPU crowd still rocking a GTX 1080 Ti: honestly, I'm so happy for you). But it's fair to say that most gamers don't feel like they're getting sufficient bang for their buck these days.

The memory supply crisis isn't just causing RAM and SSD prices to increase, it's having a knock-on effect on an array of components. After all, recent reports claim that the global CPU shortage is 'more acute for processors than memory'. While we wait for Intel's 18A chips to plug that gap, product shortages often mean higher prices on available stock. Higher prices mean fewer consumer purchases, and that could spell trouble for more than just Intel over the longest term.

While a commitment to making key components more affordable in the future should be music to my ears, obviously any PC is more than just its processors; a cheaper CPU is one thing, but if you can't find a decently priced, suitable motherboard to mount it into, what's the point? With component prices reaching for the sky, it's no wonder this is a longer term goal for Intel.

A Digitimes report on the ongoing CPU shortage says that industry sources have described the situation as 'more acute for processors than memory'.

While memory is still available in limited quantities (and with massive price increases, thanks to the ongoing RAMpocalypse), several models of chips made by both AMD and Intel are said to be effectively unavailable, regardless of price.

This shortage is hitting PC makers building enterprise-level machines the hardest, with many of those suppliers relying on the availability of Intel chips. And with some Intel CPUs already said to be in short supply since 2025, the industry looks to be waiting for Intel's 18A chips to plug the gap.

Intel's 18A process has had a difficult history. While its Panther Lake mobile CPUs, which use the node for the compute tile, are already available in some of the newest laptops (and hopefully, eventually, handhelds), yield progress was initially rather slow—although it appears to be steadily improving.

So much so, in fact, that CEO Lip-Bu Tan is now supporting its use for external customers, suggesting that those yield improvements are bearing fruit and that more capacity may be on the way.

In the meantime, some analysts have warned that consumer demand may drop in 2026 thanks to higher prices for PCs and components, even during sales periods. And while a check of the consumer component listings suggests many gaming CPUs are still available for reasonable sums, some have claimed they expect the CPU shortage to hit the market hardest in the April-June quarter.

What's bad for enterprise system builders is often bad for the gaming PC market, too, although it remains to be seen whether we'll be writing about the "CPUpocalypse" alongside the memory crisis for the latter half of the year. It's not got the same ring to it, I'll admit.

AI server demand really is throwing a spanner in the works for all sorts of components these days, and most indications suggest it may well get worse before it gets better. I like to end an article on a positive note, but... uh, it's not looking too cheery right now, is it?

Last October, Intel launched its Panther Lake mobile processor architecture, and when we got our hands on a Core Ultra Series 3 CPU several months later, we were greatly impressed by its gaming chops. Now it's back again with Wildcat Lake: all the best parts of Panther Lake, but cut down to a kitten-sized little chip.

Officially called Core Series 3 (no Ultra here), the new processors utilise the same Cougar Cove P-cores and Darkmont E-cores as in Panther Lake (PTL), but there are considerably fewer of them, even in the top-end model. For example, the Core 7 360 sports just two P-cores and four E-cores, for a total of six threads.

By contrast, the PTL-powered Core Ultra X7 358H houses four P-cores, eight E-cores, and four Low Power E-cores, which adds up to a total of 16 threads. However, Wildcat Lake (WCL) chips are aimed at a rather different market than PTL ones.

While they're both purely for laptops (though we'd love to see a handheld gaming PC with a beefy PTL chip), the Core Series 3 range is aimed directly at the entry-level/budget market of notebooks. This is why the integrated GPU in WCL is very small: just two Xe cores at most.

That's actually half the number of cores you'll find in a Core Ultra 7 270K Plus, and with only six PCIe lanes in WTL's controller tile, you're unlikely to see the chip used in a laptop with a discrete GPU. So you're really not going to be doing any graphics-focused gaming with a Wildcat Lake laptop.

Instead, Intel says what you will be doing is enjoying "all day battery life and everyday productivity, with up to 2.1x faster creation and productivity, up to 64% lower processor power, and up to 2.7X AI GPU performance versus previous generation Intel Core 7 150U processors."

While I have no doubt that Core Series 3 chips are very capable little things, given what I've seen with Panther Lake, I do wonder just how many businesses, organisations, and other sectors will be upgrading to new WCL laptops.

All laptops have risen in price due to supply issues with DRAM and flash memory, and the multi-tiled nature of Intel's latest processors does mean that they're more expensive to manufacture than AMD and Qualcomm's monolithic chips. Intel doesn't list market prices for its mobile processors, since you can't buy them directly, but every PTL laptop I've seen so far has been anything but cheap.

It will also be interesting to see how a Wildcat Lake laptop fares against the MacBook Neo. They're unlikely to compete on price (the Neo starts at just $599), but if the chip is considerably superior in terms of performance and battery life, it might go some way to helping customers see past the higher price tags.

Then again, every cent counts in business, and in these RAMpocalypse-trying times, Intel's little chip could be left out in the wild.

Patent law is often confusing, especially when the alleged patent infringement is from none other than Intel. In just one of many twists in a story spanning almost a decade, the US appeals court has restarted proceedings for VLSI to take Intel back to court, continuing a $3 billion feud.

As reported by Reuters, the decision was given on April 14 in California. This decision is not inherently proof of wrongdoing and is not a judgment from the courts. It simply suggests that Intel will once again be fighting its case against VLSI in the future.

Effectively, VLSI owns many patents for techniques for handling speed, power, and architecture in processors. The company has alleged since 2017 that these are infringed upon in Intel's modern processors. These patents are ones that VLSI managed to get its hands on from the likes of Dutch chipmaking company NXP Semiconductors and Freescale Semiconductors.

The case currently being fought here is for patent 8,566,836, which was originally filed by Freescale Semiconductor Inc in 2009. VLSI sued Intel for infringement regarding this patent and seven more since 2017, but '836 is the issue of the most recent appeal. As described in the case, '836 "relates to choosing one or more cores of a multicore processor to execute a particular task, for example, based on whether the task must be executed on a single core or can be executed across multiple cores."

However, this has been a long and winding road for the two entities, with both going blow for blow. In 2021, VLSI won a court case against Intel for the sum of $2.18 billion, with $675 million coming from patent 7,725,759, and the rest coming from 7,523,373. This case was reversed later, and Intel was not sent that bill.

Intel was originally stopped from appealing this due to an ongoing legal dispute, but was allowed to do so when it joined two separate petitions against VLSI's patents filed by other companies. In other words, it managed to get around the appeals processes by tacking itself onto others' complaints.

In 2022, Intel was once again hit by a big bill from VLSI (at $948.8 million), which Intel also challenged.

The new appeal from the court rejects Intel's previous arguments to appeal the original '836 verdict (yeah, I know. It's confusing). The conclusion of the decision argues that further proceedings are necessary in this case. All this is to say that despite many back-and-forths between the two companies spanning almost a decade, we don't seem to be done yet.

And though Intel was making a net income of $5.9 billion by the last quarter of 2020, it failed to make a profit last year. With Intel betting big on the likes of AI and its latest processors, legal battles over chip architecture patents are an expensive affair. It'll just have to hope it wins in the end.

This September will mark the tenth anniversary of AMD's 'legendary' AM4 desktop processor socket, one that's still going strong today. And if one leak is to be believed, it would seem that Team Red is planning to mark the occasion by releasing a special version of another classic: the Ryzen 7 5800X3D. Except there doesn't seem to be anything special about it at all.

The leak in question comes from HXL on X, with what looks like a photo of a presentation slide that shows all the key pieces of information behind this puzzle, i.e. the processor, the year, and a 10-year anniversary logo for the AM4 socket.

AMD's Ryzen 7 5800X3D first appeared in 2022, and despite being somewhat pricey and hot-running, it set all kinds of performance records in certain games, thanks to its massive 64 MB of L3 cache bonded to the top of the core chiplet. Better known as 3D V-Cache, the additional SRAM returned with future generations of Ryzen chips, turning the 7800X3D and 9800X3D into the best gaming CPUs money can buy.

Since those particular processors use a newer socket, it makes sense that AMD would return to its best AM4 gaming chip to celebrate the decade-old system. What's puzzling, though, is that if the photo is 100% genuine, it would seem that there's going to be absolutely nothing special about this anniversary edition.

Well, other than the fact that you can potentially buy it, which is sadly not the case for the original Ryzen 7 5800X3D. That's because it doesn't seem to be manufactured any more, and along with the 5700X3D, a slightly cut-down version of the original, it is incredibly hard to find new, in its retail packaging, and at a sensible price.

But other than that, the special edition 5800X3D has the same specifications as a standard 5800X3D: eight cores, 16 threads, a 4.5 GHz boost clock, 100 MB of total cache (4 MB L2 + 96 MB L3), and a TDP of 105 W.

It might come with a fancy box or maybe even a plush stand, so you can show off the chip on your gaming desk, but the leaked photo provides no hints of this. Likewise, there is no indication of price and market availability. My guess is that if all of this is genuine, then it's going to be very pricey and pretty hard to find in stores.

I hope not because the kind of birthday party where you only get invited if you're rich and popular is no party at all.

Anyone who does a lot of data analysis will know that the longer you collect data for, the easier it becomes to spot trends. Sometimes, though, a pattern is so stark that you don't need to be a data expert to spot it, and in the case of German retailer Mindfactory, its CPU sales figures tell you everything you need to know about PCs right now.

The numbers I'm about to show you come courtesy of YouTube channel TechEpiphany, via its X account, where it routinely collates and posts weekly sales numbers for CPUs, GPUs, and motherboards. The data mostly refers to Mindfactory, but TechEpiphany does similar tracking for Amazon and others.

Anyway, the above chart shows the number of AMD and Intel CPUs sold at the German store, starting week 3 of this year, up to week 15. In some cases, successive weeks have identical sales, but that's because TechEpiphany has posted a value for a group of weeks, and I've taken the average weekly figure from that.

When I first saw the numbers in raw form, it didn't strike me as being particularly bad in any way, but once visualised in a graph, the drop in sales is pretty stark. So much so that it drove TechEpiphany to say, "In 10 years of tracking retail CPU (and related) sales, I have never seen such a steep decline."

Now, I'm only showing you Mindfactory's data, but I'm sure the tech tracker has other sources of data to make them make such a statement. Not that this will come as a surprise to any PC enthusiast who regularly checks out the prices of components or follows tech news, because no matter how great today's CPUs, GPUs, and motherboards are, increasingly fewer people are considering doing a full upgrade.

The outrageous price tags that DRAM kits and SSDs currently sport are a colossal barrier: You're looking at spending at least $160 at Amazon for a basic 1 TB NVMe drive, and should you be feeling like throwing caution to the wind and sticking 32 GB of snappy DDR5-6000 into your rig, you'll be handing $380 to Newegg for the privilege.

Add them together (i.e. $540), and you have a sum of money that's actually enough to buy an RTX 5060 graphics card, with a Ryzen 5 7600X taped to its heatsink.

So, naturally, PC shoppers are making the sensible decision to forgo doing a full platform upgrade until memory and storage prices come back down into the realms of affordability. Put it like this: 12 months ago, you could buy a 1 TB PCIe 4.0 SSD and a 32 GB kit of fast DDR5 for less than $200. Sadly, it'll probably be more than 12 months before we see that kind of price again.

You can always tell when an exciting new bit of PC tech is getting nearer to launch, because all the various leaks and rumours get increasingly more comprehensive and detailed. That's very much the case with the latest drop for Intel's Nova Lake chips, because now we have a look at what the full range of Core Ultra 400-series processors is going to look like.

Well, that's according to Videocardz, which says it's been shown a new roadmap for NVL-S, the next generation of desktop processors from Intel, which will be officially listed as Series 4, or Core Ultra 400S. As there has already been a wealth of leaks about the forthcoming chips, much of this information isn't especially new, but seeing all tied together and with snippets of details for the range of SKUs is interesting.

Just as with the current ARL-S (Arrow Lake) chips, and every other generation of Intel processors before it, Core Ultra 400S CPUs will come in a wide range of varieties for different segments: 125 W for general scenarios and gaming, plus 65 W and 35 W models for office and embedded applications.

PC gamers are most likely to be interested in the usual Core Ultra 9, 7, and 5 chips, and at first glance, it just seems to be the same as it has always been. The series leading 9-class models will sport nine Coyote Cove P-cores, 16 Arctic Wolf E-cores, and four Low Power E-cores (LPE).

Dropping down a tier to the 7-class will mean saying goodbye to four E-cores, and going further down the order to the 5-class results in losing two P-cores as well.

So, just as with Arrow Lake, then (though these don't have LPE-cores). However, where things get spicy are the bLLC variants, of which there is one apiece in the 9 and 7-class tiers. These are generally expected to be Intel's answer to AMD's X3D processors, as the bLLC is a monstrous slab of 'last level cache', purported to be up to 144 MB in size (for comparison, the 3D V-Cache slice in the Ryzen 7 9800X3D is 64 MB).

Whether a 7-class CPU gets that much additional L3 cache is anyone's guess, but I would reckon that Intel is looking to have a better separation between its Nova Lake SKUs than in Arrow Lake, so the full 144 MB might be reserved purely for Core Ultra 9 processors.

Now, so far, I've not said anything about the 'dual compute tile' models. That's because Videocardz's source for all of this information is apparently saying that Intel has yet to declare the brand nomenclature for the 44 and 52-core CPUs. This means these are unlikely to be sold as Core Ultra 9 chips, for example, and it's possible that Intel may choose to revive its X-series branding from 2017/2019.

The likes of the Core i9 10980XE Extreme Edition were huge, very expensive processors, with up to 18 cores and with quad-channel memory support. Purely for the HEDT (high-end desktop) market, such chips were for workstations and applications that required more threads than a 'normal' CPU could offer, but less than you'd get from a full-blown Xeon chip.

This is precisely what AMD's Threadripper chips are for, though as impressive as they are, they're really not for gaming at all.

Anyway, Videocardz also says that the leak confirms some other aspects about the Core Ultra 400-series, such as that the processors will still only have a maximum of 24 PCIe lanes available for graphics cards, SSDs, and other components (the same as Arrow Lake), but Nova Lake will support Thunderbolt 5, rather than 4, with up to 120 Gbps of transfer speed and 240 W of charging.

Assuming all of this is correct (and at this stage in the rumour mill, it almost certainly is), the only important pieces of information that are unknown are clock speeds, prices, and the all-important release date. Take your best guesses at those, folks.

After several years of uncertainty, job cuts, issues with self-destroying CPUs, and spending billions of dollars on its foundries, Intel has made big strides in moving into a better place. Thanks largely to its latest products and deals with other tech and AI giants. Investors certainly seem to think so, because its market cap is now the highest it has been in more than two and a half decades.

As reported by Tom's Hardware, you can see the figures yourself and easily spot that the last time Intel's market capitalization was above the $300 billion mark was all the way back in October 2000. According to CompaniesMarketCap, the site tracking the figures, this makes Team Blue the 47th most valuable company in the world, though just by market cap and nothing else.

In comparison, AMD is currently enjoying a cap of $385 billion, and Nvidia is miles ahead of either company, with a market cap that's approximately $4.5 trillion in size. However, this is all somewhat of an unfair comparison, because neither of those firms have their own manufacturing plants for CPUs and GPUs, whereas Intel does.

You'd think that this would result in a higher market cap, but for the majority of Intel's history, it's only ever made chips for itself, and those fabs have been a serious financial drain in the past. But now that it's agreed to be part of Elon Musk's Terafab project, there's a chance that Intel's numerous foundries could be busy creating or packaging AI chips for SpaceX and Tesla.

Intel has already recently agreed to a chip deal with Google, and along with how well its Panther Lake and Arrow Lake 200S Plus chips have been received, it would seem to suggest that investors believe that the 57-year-old company is on the right track to regaining its former glory.

It's certainly faring better than other US-based chip companies, such as Texas Instruments and Qualcomm, which are either relatively static or currently on a bit of a slide, in terms of market cap.

Chasing down manufacturing orders seems to be what investors are most interested in Intel doing right now, probably because no matter what its desktop CPUs are like (or what they do to themselves), Intel's overall processor market share is still pretty dominant. AMD is certainly making headway in the world of server chips, though, and when it comes to PC gaming and consoles, Team Red rules the roost.

A healthy AMD and Intel is good news for consumers, because having just one company utterly dominate a market isn't great for competition (*cough graphics cards *cough). Investors are perhaps less concerned by this, but they're certainly happy with how things are going at Intel right now.

In the world of tech, if there's no competition in a particular sector, you can pretty much charge whatever you like for your product. Which is why AMD has set the official price tag for its new Ryzen 9 9950X3D2 Dual Edition processor, with 192 MB of L3 cache, at a seemingly outrageous $899.

To put that into perspective, the launch price for the Ryzen 9 9950X3D was $699, though you can pick one up for $676 at Newegg right now. So what exactly are you getting for your extra $200?

Well, compared to the 9950X3D, you're getting a small decrease in the peak boost clock (5.6 GHz instead of 5.7 GHz) and a 30 W bump to the TDP. Oh, and you're also getting a 64 MB slice of 3D V-Cache on both Core Complex Dies (CCDs), for a grand total of 192 MB of L3 cache.

Is the 50% increase in cache worth $200? For gaming, absolutely not, because you won't see any performance games in the vast majority of games out there. That's because the threads that games generate are generally processed on just one of the eight-core CCDs.

You won't see much in the way of any gains for content creation tasks, either. AMD suggests that in video editing, offline rendering, image manipulation, or code compilation, the 9950X3D2 will only be around 5 to 8% faster than the 9950X3D.

However, for any scenario that is heavily dependent on cache or memory performance, rather than thread count, the new chip will certainly stand out. For example, scientific analysis, engineering modelling, or anything that involves lots of data read/writes, CPUs with lots of cache are much sought after.

For those roles, you'd be typically looking at an AMD Threadripper/Epyc or Intel Xeon processor, but they're all very expensive ( So, for the right person, spending $899 (£830 in the UK) on a Ryzen 9 9950X3D Dual Edition actually makes a lot of sense.

Naturally, there will be some PC gamers out there that will still go out and buy one, purely to have the 'best' AMD Zen 5 CPU money can buy. All the sensible gamers, though, will just get a Ryzen 7 9800X3D (currently $469 at Amazon) instead. It really is the best gaming CPU you can put in your rig.

Forget Arrow Lake Plus, Nova Lake and even Razer Lake AX. The future for Intel is the past in the form of Raptor Lake CPUs. Well, kinda. The company has just said that the Raptor Lake generation of chips is "not going anywhere" and will "continue to be abundantly available."

Speaking to Club386, Intel VP Robert Hallock (who our Nick chatted to recently about Intel's new Arrow Lake Plus chips) explained that Raptor Lake is, "still really, really good, even with multiple generations of hardware from other vendors coming after it, so it’s not going anywhere. I want people to understand that Raptor Lake will continue to be abundantly available."

Exactly how long Intel will keep cranking out Raptor Lake chips isn't clear. All Hallock will say is that Raptor Lake is "an extremely fast product and is not going away any time soon."

However, it seems part of Intel's thinking here involves support for DDR4 memory to complement DDR5. "You’ve also seen some new motherboard announcements that support both DDR4 and 5 on Raptor Lake, as kind of like a bridge between worlds for people. That is reflective of our overall confidence and expectations." Hallock says.

Clearly, DDR4 support gives gamers more options in the current context of acutely constrained memory supply thanks to the AI boom. Right now, 32 GB DDR4 kits can be had on Newegg for around $200 and up. But you're looking at more like $350 plus for DDR5. That's quite the delta. Given SSD and GPU prices are also pretty painful, that's a worthwhile saving for anyone desperately trying to keep the cost of a new rig build in check.

Of course, DDR4 prices are as volatile, too. But having more options is surely a good thing. If DDR4 spikes, maybe DDR5 will then become an option, and so on. And, of course, some gamers may have DDR4 memory to hand in an existing rig, sidestepping the need to buy memory altogether.

Given the memory crisis is widely expected to continue at least through 2027 and perhaps beyond, you'd have to think Intel has a similar time horizon in mind for Raptor Lake production. In which case, we'll be in a scenario where Intel launches Nova Lake and has fully three generations of CPU in production: Nova Lake, Arrow Lake and Raptor Lake.

Indeed, could it even be possible that Intel sunsets Arrow Lake but keeps Raptor Lake in production? Given Arrow Lake doesn't support DDR4 and hasn't been a roaring success, you have to think that's a possibility. It would actually make sense.

Nova Lake provides the modern platform, the latest features and, at the high end, the best performance. Raptor Lake, meanwhile, offers a budget option that still delivers decent performance. Makes sense to us, anyway.

The slight caveat to all this involves the widely publicised stability and reliability issues with 13th and 14th Gen Raptor Lake chips. But those can probably be considered resolved, at least for chips being bought new as opposed to any existing CPUs that were damaged.

Rumours around Intel's next-gen Nova Lake CPUs have multiplied rapidly in recent days. Highlights include an IPC performance advantage over AMD's upcoming Zen 6 chips, a core count upgrade for some models, and a massive APU variant to take on AMD's Strix Halo or its successor.

First up, those IPC claims. IPC refers to instructions per clock, and it's essentially a measure of how much work a CPU core does in each clock cycle. It's not a measure of overall performance, because IPC combines with clock speed to determine actual performance.

In other words, a processor's performance hinges on how much work a core does per clock cycle multiplied by the number of clock cycles per second. Anyway, established tech leaker HXL on X recently posted the following superficially cryptic message: "IPC: CYC>Zen6. Clock: Zen6>CYC".

For reference, "CYC" here refers to Coyote Cove, which is the generation of Performance core used by Nova Lake CPUs. With that in mind, the leaker is claiming that Nova Lake will beat AMD's Zen 6 when it comes to IPC, but Zen 6 will have a clockspeed advantage.

To be frank, with just those two data points, anything is possible. For instance, should AMD's clockspeed advantage be sufficient, it could still beat Intel regarding single-core performance. Exactly how high AMD will clock its next-gen Zen 6 cores isn't known, but some rumours point to figures in excess of 6 GHz thanks to the use of TSMC's most advanced N2P silicon.

Next up, Intel has apparently upgraded its mid-range Nova-Lake model from 42 cores to 44 cores. These variants have previously been said to be based on two Nova Lake compute tiles, each with 7 Performance, 12 Efficient and 4 Low Power Efficient cores. But the full-sized Nova Lake compute tile contains eight P cores. And now all eight are said to be enabled for this SKU according to another established X leaker, @jaykihn0.

Of course, the top Nova Lake chips have long been said to get 16 P cores, 32 E cores and four LPE cores, making for a grand total of 52 cores and a right old monster of a CPU. This new rumour does nothing to change that. What it does do is imply that whatever process Intel is using for Nova Lake, yields are probably decent.

After all, presumably the reason why Intel had previously disabled a P core in each tile on the previous 42-core model was to enable some chips with a dead P core to be used. If Intel has decided it doesn't need to do that, maybe yields are particularly good.

Another rumour involves a chip that is variously referred to as either Nova Lake AX or Razer Lake AX. Either way, what we're dealing with here is a monster APU that's said to use an absolutely huge socket known as LGA4326.

Indications of just such a socket have been found in shipping manifests on NBD Data attached to the NVL-AX code, which seems to be a pretty clear reference to Nova Lake AX. Some sources now refer to this chip as Razer Lake AX, and the implication here is that an earlier Nova Lake-based APU has been cancelled in favour of another APU using the newer Razer Lake architecture.

The most important aspect is that an LGA4326 implies an absolutely massive CPU socket with 4,326 contacts. @jaykihn0 on X says it's actually a BGA rather than LGA socket, but the point is that 4,326 contacts or pin outs is a massive step up from the 1,851 in Intel's current LGA1851 socket for desktop Arrow Lake CPUs.

Previously, Nova Lake AX was rumoured to run 28 CPU cores and 48 GPU cores, the latter based on Intel's upcoming Xe3P architecture. It's not known if the shift from Nova Lake AX to Razer Lake AX changes that, but given that this big APU is expected to be a chiplet design, it's quite possible that Intel could switch out the CPU tile but stick with its plans for the GPU tile.

On paper, a GPU tile with 48 Xe3P cores would be hugely powerful. Intel's current Arc B580 GPU has just 20 Xe2 cores. So, this rumoured APU has well over double the cores, and presumably each Xe3P core is more powerful than an Xe2 core. That said, @jaykihn0 nevertheless claims that there will be a version for laptops, which is intriguing, even if most observers think Razer Lake AX will be targeted at running local AI workloads rather than gaming. We'll just have to wait and see.

Taken overall, what we probably can conclude is that Intel's next-gen CPUs look like a much more exciting family of chips than anything Intel has produced in recent years. We're talking very powerful individual CPU cores combined with high core counts and maybe even the option of an APU variant with a properly powerful GPU.

As ever, rumours are just rumours until products are launched. But we've haven't seen rumours this exciting about upcoming Intel chips for ages.

As with any new piece of PC technology, months of leaks and rumours ensured that Intel's launch of its Core Ultra 200S processors wasn't especially surprising. More cores, higher clock speeds, and lower price tags. But once I put the Core Ultra 7 270K Plus and Ultra 5 250K Plus chips through our battery of benchmarks, I honestly couldn't believe how good they were, especially in light of them being refreshes of CPUs that had so disappointed me back in October 2024.

But after recently chatting with Intel about Arrow Lake Refresh, the shock has been replaced by an understanding. Let me explain by going back to the launch of the first Arrow Lake processors, the original Core Ultra 200S series. For those chips, Intel held various press briefings to explain everything about the new architecture, what to expect, and why certain decisions had been made.

Time has a funny way of changing one's memories, but I can certainly recall Intel's staff coming across somewhat hesitant. Responses to questions were a little guarded, and when we all finally got our hands on the CPUs for testing, it all became very clear: Other than for a few specific applications, Arrow Lake was merely 'okay' and AMD's Zen 5 Ryzen chips were still going to be the number one choice for any PC user.

Fast forward 16 months, and it was a rather different Robert Hallock, vice president and general manager at Intel, who fielded my questions about the new chip designs. Joined by Erin Maiorino (director of CPU technical marketing) and Thomas Hannaford (desktop/workstation communications), the trio were cheery and chatty, and perhaps more importantly, confident.

Given that the 250K Plus, at just $199, runs amok in so many benchmarks (to the point that it's now our top recommendation for best mid-range CPU), such assurance is understandable. The 270K Plus, only $100 more expensive than the 250K Plus, is one of the most potent desktop CPUs Intel has made in a long time. However, there was more to it than just 'Yeah, we're happy because our new chips are really good.'

Hallock pointed out that much has changed at Intel, from staffing to internal structures, and new ways of doing things; he often remarked about how much the nearly 60-year-old company was "listening to customers".

It's not just the Arrow Lake refresh that's really good, as the Core Ultra 300 series (Panther Lake) is a great processor for laptops and would be superb in a little handheld gaming PC. I say 'would' simply because the global memory crisis has pretty much put paid to any hopes of seeing a Panther Lake-powered handheld, because it needs fast LPDDR5X DRAM to properly shine, and that stuff is super expensive now.

I've also been somewhat unfair to the first iteration of Arrow Lake, because thanks to a whole host of microcode, BIOS, driver, and operating system updates, they run an awful lot better than they first did. For outright gaming, you're still going to be choosing an AMD X3D processor, and likewise stick with Team Red for an ultra-budget chip (i.e. the $83 Ryzen 5 5500) or a workstation ultra-threaded processor, but for everything else, Intel's offerings are solid choices.

With Panther Lake, updated Arrow Lake, and the 200S Plus duo, it does seem like Intel is indeed listening to what the consumer market wants: fast, broadly capable chips that don't use too much power and don't have an outrageous price tag. Intel's even saying its future desktop PC sockets will last longer than just one or two generations of processors.

One important thing to consider is whether this is just a flash in the pan. Are the prices for the 270K and 250K Plus indicative of what to expect from Nova Lake when it rolls out later this year, or has Intel been happy to accept a lower profit margin for a short while, just to get a brief edge over AMD? Will I really be able to fit a 2029 Core Ultra 600S processor in a 2026 Z990 motherboard? Is it really listening and responding or just saying things that consumers and the market want to hear?

I suspect nobody outside of Intel has the actual answers to these questions, and as someone with a pretty poor track record of correctly predicting the future of PC tech, I'm hesitant to put my chips down (if you pardon the pun). But it does feel like this is a 'new' Intel. Not entirely new, of course, but for an enormous, multibillion-dollar company, with tens of thousands of employees around the world, it's about as new as you're going to get.

Over the decades, I've sat through countless press briefings, tech presentations, and architecture conferences, and the recent ones from Intel really do stand out. Naturally, such events should be vibrant, full of energy, and packed with promises and tech hype, but they don't always come across that way (for example, AMD's lacklustre launch of RDNA 3).

Talk is cheap, of course, and a handful of punchy or lousy presentations ultimately don't matter if the end product is rubbish or brilliant. Few PC enthusiasts aren't really going to care what somebody said on a stage one afternoon, many moons ago, when they're out shopping for a new processor or graphics card.

But they probably are going to remember, or at the very least, be influenced by reports of a company doing the right things. When it re-entered the discrete graphics card market in 2022, with its Alchemist-powered Arc range, Intel promised that it would have frequent driver updates and support. Despite having the tiniest of slices of that market, Intel has stuck to its word, though this hasn't helped shift any more Arc cards, sadly.

Just as important, the chip giant is learning from past mistakes and directions. It's only been a handful of years since we had the Intel that made Raptor Lake, a processor design with an overwhelming thirst for power and launched with some inherent operating and manufacturing flaws.

High-end desktop versions had a tendency to obliterate themselves in certain circumstances, and laptop variants were incredibly hard to keep cool. Now, we have desktop and mobile chips that sip away at power, sport excellent GPUs, and are highly versatile.

Somewhat obviously, Intel had to do this, though. With AMD constantly increasing its share of the x86 market and Arm processors making headway in the world of desktop PCs, sticking to the old ways of doing things simply wasn't an option. But regardless of what's really going on behind the scenes, at this moment in time, Intel's making the right products at the right prices.

If we all shout 'Yes! More of this, please!" and then, rather crucially, go out and buy the products, perhaps the new Intel will continue to listen for many more years to come.

AMD compute and enterprise AI VP Madhu Rangarajan told tech news site CRN that they are expecting next-gen AMD Venice servers to increase the performance lead over next-gen Intel Xeon Diamond Rapids, in part due to Intel's decision to ditch for the upcoming server CPU generation.

The AMD VP said, "We understand that our competitor has made some interesting choices on multithreading and so on. That’s going to help us gain even more market share in a broader enterprise market where that has implications to your licensing and many other costs associated with it."

Rangarajan is referring to Intel's decision to ditch HyperThreading—i.e. simultaneous multi-threading (SMT)—in the Panther Cove P-cores that will lie at the heart of the upcoming server processors. This isn't to be confused with Panther Lake (PTL), Intel's recent snappy laptop chips, which use Cougar Cove P-cores, though these also lack HyperThreading. Different cores, different markets, same absence of multi-threading.

AMD's upcoming Venice server architecture is expected to include SMT, with up to 256 cores and 512 threads per chip.

Intel fellow Srini Krishna responded that the company sees "positive momentum with our data center customers... with the performance, security, and efficiency of [the] Intel Xeon 6 family... Ultimately, winning in the data center is about the aggregate value we deliver across security, efficiency, consistency, quality, and overall platform experience."

Last year, Intel's lead x86 CPU architect, Stephen Robinson, explained the decision to ditch HyperThreading with PTL:

"We didn't lose a lot in client because of hybrid and the core count, but we gained a bit in our design and execution... Now you've kind of got something that's maybe a little bit easier and less expensive and maybe can go a little bit faster... If we can get more workloads running on an E-Core then we can bring more efficiency to the whole platform."

The upcoming Xeon server chips will also lack E-cores, however. This is something we've recently heard might be the case for Intel's upcoming tiny mobile chips, the Core 300-series (sans 'Ultra'), though they will use low-power versions of Intel's Darkmont E-cores.

After Diamond Rapids, we can expect SMT to probably return in server chips, however. This is because Intel CEO Lip-Bu Tan explained last year that this is the desire:

"In data center, we are focused on regaining share as we ramp Granite Rapids while also improving our capabilities for hyperscale workloads. To support this, we are reintroducing simultaneous multi-threading (SMT). Moving away from SMT put us at a competitive disadvantage. Bringing it back will help us close performance gaps."

Diamond Rapids is expected to launch later this year and be built on Intel's 18A node. What we've seen of 18A so far, with PTL chips, is pretty impressive, so there are no concerns on the node front. Notably, though, Intel's next-gen Nova Lake desktop chips are expected to be made on a mix of Intel 18A and TSMC N2 processes, not full 18A.

In the PC gaming market, AMD chips hold a pretty significant lead against Intel ones, despite new impressive additions like the Core Ultra 7 270K Plus and Core Ultra 5 250K Plus. There's a reason five of the six CPUs we recommend for gaming right now across all budgets are from Team Red: Intel just doesn't have the same value offering when it comes to pure gaming.

However, when it comes to market share, things change somewhat. Despite AMD climbing significantly, Intel still holds about 70% of the desktop CPU market. And the picture is similar in the server market, with AMD making some serious gains but Intel still retaining the lion's share.

We're only a couple of months off the back of Intel's Core Ultra Series 3 'Panther Lake' laptop chips, which our Andy called "fast, feisty, and fabulous" after extensive testing, and now we've got some concrete specs for a new generation of Intel mobile chips—assuming X user Jaykihn's leaked table is legit, that is. As these things go, this leaker is reliable and often right about these things, so there's good reason to take them seriously.

You might have noticed the lack of an 'Ultra' designator, here, and that's because these chips aren't actually direct successors to Panther Lake ones. Instead, they're low-power ones, meant for thinner and cheaper laptops.

You'll also notice in the reproduced and simplified table above that there's a conspicuous lack of E-Cores—there are only LPE-cores, which are the low power variants. And these LPE-cores will be Darkmont ones, the same that we get in Panther Lake. We'll still be getting some more performant P-cores, of course: two of them alongside four LPE-cores.

That's for all six models listed, from the Core 3 304 up to the Core 7 360. Another spec that's the same for all chips in the line-up is a 35 W turbo TDP (base 15 W). That's far from the TDPs of high-end Panther Lake chips, and is more in line with what we see in some of the best handheld gaming PCs, such as in the AMD Ryzen Z1 Extreme.

We're also unfortunately only getting a maximum of two Xe graphics cores, at least in these variants, which means they likely won't be as good as the chips in current handheld gaming PCs, so it's unlikely these will be picked for that purpose.

Still, given how much Panther Lake has impressed us on the performance and efficiency front, these Wildcat Lake chips will hopefully make for some mean little efficiency machines—just probably not for anything near serious gaming.

There's also the chance that further models could launch which do have more Xe cores, with an X designator such as a Core 7 360X, just as we have with Panther Lake chips.

The company behind the popular Geekbench software tool has taken another pot shot at Intel's BOT software.

Primate Labs says that Intel BOT "only supports a handful of applications, meaning BOT-optimized benchmark results paint an unrealistic picture of how a CPU performs in practice. This makes Intel processors appear faster relative to AMD and other vendors than they would be in typical, real-world usage."

Following the announcement last week that all Geekbench results posted courtesy of Intel's new Arrow Lake Plus CPUs would receive warning flag in the online database, Primate Labs has a new blog post detailing the results of in-house testing carried out with Intel's BOT software.

Primate Labs actually carried out their testing on a Panther Lake laptop rather than one of the new Arrow Lake Plus chips. Turns out Intel's Panther Lake CPUs also support BOT, something Intel didn't initially flag.

Anywho, Primate Labs says, "Intel’s public documentation on BOT is limited, so we decided to dig in ourselves to understand how it works and what optimizations it’s applying to Geekbench."

Primate Labs claims it discovered that BOT is going well beyond Intel's characterisation of how the software operates.

"Based on the instruction counts, it’s clear BOT has performed significant changes to the HDR workload’s code. The number of total instructions is reduced by 14%. Most of that reduction comes from BOT vectorizing parts of the workload’s code, converting instructions that operate on one value into instructions that operate on eight values.

"This is a significantly more sophisticated transformation than simple code-reordering. Intel’s public documentation only discloses the simpler code-reordering techniques, not the vectorization transformations observed here," Primate Labs says.

If Primate Labs is correct, that's certainly something of a concern. To quote myself from last week, "my understanding is that what Intel's BOT is doing essentially amounts to re-ordering instructions so that they fully utilise the Arrow Lake Plus pipeline. All the actual calculations are the same. In other words, enabling BOT doesn't mean skipping any work."

I said that based on Intel's description of how BOT works, but if Primate Labs is correct, there's quite a bit more going on. Primate Labs also found that BOT adds a time penalty to application start up. "When running Geekbench 6.3 with BOT enabled, the first run has a 40-second startup delay before the program starts. Subsequent runs are faster, with a 2-second startup delay. The startup delay disappears when BOT is disabled," the blog post explains.

Ultimately, this all comes down to how many applications end up supporting BOT. My understanding is that application support will require Intel's Labs specifically doing optimisation work and adding the results of that to the tool. Quite how many apps Intel will choose to optimise is unclear, but the fact that BOT is never going to just work with any given app is a clear negative. Primate Labs probably has a point here, therefore.

AMD has announced a new X3D processor, the AMD Ryzen 9 9950X3D2 Dual Edition, set to launch on April 22. This chip will have 3D V-Cache on both of its eight-core chiplets, something we've not seen before. Though don't expect this to mean twice the gaming performance—far from it.

"Together," AMD Computing and Graphics SVP & GM Jack Huynh says of the two chiplets with V-Cache, "they deliver an incredible 208 MB of total on-chip cache, the most we ever built into a Ryzen processor."

"208 MB of cache means more game data, more assets, and more working data sitting right next to the CPU cores, that dramatically reduces latency and unlocks new levels of responsiveness across demanding workloads."

That's not the only way the 9950X3D2 goes big, though. The Ryzen 9 9950X3D2 has the same 16 cores as the Ryzen 9 9950X3D but demands a 200 W TDP, 30 W higher than the more gaming-focused chip, because it has 192 MB of L3 cache.

According to AMD, the new chip gets a 5–10% performance boost across the creative and productivity tasks compared to the 9950X3D. But no gaming figures in sight.

We already suspected such a chip might have been on the cards, especially given motherboard maker ASRock recently accidentally spilled the beans on it. Though it wasn't certain whether that was for a planned-and-cancelled release or a planned-and-still-going-ahead one.

We also already knew that such a duo-cached processor probably wouldn't be a massive benefit to gaming. That's because games still don't efficiently use a ton of CPU cores, and the higher number of threads that could benefit from the added L3 cache won't benefit hugely if they have to traverse across to another chiplet. The latency there will likely offset a lot of the benefit. Plus software will have to be programmed to call for such things effectively and efficiently, which isn't a small ask.

AMD agrees:

"You still get the incredible gaming performance you expect from a Ryzen 9 X3D processor, but where the 9950X3D2 really shines in workloads that thrive on ultra-fast data access: large software builds, game engine compiles, AI models, 3D rendering, and complex content creation pipelines."

The idea with this chip is instead to have a step up in professional or creative performance without sacrificing gaming performance. It might make an ideal processor for a solo game dev, for instance, who wants to use a computer for both developing and gaming.

"You'll no longer have to choose between a gaming or creator CPU. With a Ryzen 9 9950X3D2, you get both."

For context, the much more expensive AMD Ryzen Threadripper PRO 9955WX has 16 cores and 64 MB of L3 cache, though those are of course very different cores that are tailored to professional workloads. The current Ryzen 9 9950X3D has 16 cores and 128 MB of L3 cache.

It's a little of an apples to oranges comparison, of course, but AMD seems to be positioning the new double-cached 9950X3D2 as kind of approaching Threadripper territory thanks to that extra cache that does help some professional or creative workloads. Except you are, of course, getting all the gaming chops of your usual X3D chips. Which is to say, a lot of gaming chops.

There's no word on price as of yet, but with the Ryzen 9 9950X3D costing $700, it's safe to say this won't be a cheap chip. It should be cheaper than a Threadripper, though, and that's probably what will matter the most for the niche market it's tailored to.

A week ago or so, I flagged a report claiming that CPUs about to become the next chip class walloped by an AI-instigated shortage and spiralling prices. Now another source is making essentially the same claim but this time via at least two gaming PC maker.

Nikkei Asia reports that a supply crunch is hitting the PC and server CPU markets, leading to "an average price hike of between 10% to 15%" with some CPU model prices spiking even higher.

The outlet further claims that, "Intel and AMD have recently told clients they will increase prices for all series of CPUs from March and April, respectively." Lead times for CPU orders have likewise reportedly increased from one to two weeks to eight to 12 weeks, with some orders taking as long as six months to be filled.

Notably, Nikkei Asia says, "an executive with a gaming PC company said that the shortage of CPUs is expected to become worse in the April-June quarter." It quotes the executive saying, "Intel and AMD have prioritized capacity for server CPUs, and the supply for PCs has become less ... what PC players can get in Q2 is much less than the volume we got in Q1."

The same executive voiced fears that even with escalating prices, supplies will be tight. "What we worry about is that even if we pay more we still cannot get more. The CPU shortage is getting more serious day by day, no less than the memory chip situation," the source said.

What's more, Nikkei Asia also spoke to Jose Liao, general manager of systems business at Asus. Liao said mid-range x86 chips will be subject to the most acute supply crunch, with Intel tilting production in favour of high-end chips. "The supply gap is indeed widening and is expected to continue," he said.

Inevitably, the cause behind all this is said to be "surging" demand for AI computing. Nikkei Asia also says Intel commented that it has, "informed customers of planned pricing updates on select products, reflecting sustained demand, increased component and material costs, and evolving market dynamics."

As I reported in the story linked above, Intel has spoken previously of its difficulties keeping up with demand, in part thanks to the yields at it own chip production fabs. As for AMD, it is constrained by production capacity at manufacturing partner TSMC.

Arguably, what's slightly odd in all this from a gaming PC perspective is that Intel has just launched two new Arrow Lake Plus desktop CPUs and what's most remarkable about them is their notably low prices. The new Intel Core Ultra 7 270K Plus and Intel Core Ultra 5 250K Plus officially go on sale March 26 for $299 and $199, respectively.

If nothing else, it will be interesting to see what those chips are actually listed at by retailers, and if it is MSRP how long that lasts.

Primate Labs, the company behind Geekbench, has announced that it will be attaching a warning to benchmark results produced by Intel's new Arrow Lake Plus CPUs. The benchmark specialist says that Intel's new BOT tool, which only runs on the freshly-launched chips, as well as Panther Lake parts, "can boost Geekbench 6 scores by up to 8%, but those results aren't comparable with standard runs."

Primate Labs actually says that individual workloads can be improved by as much as 40%. As for a justification for the move, the company says: "Since the tool modifies the benchmark, and it is unclear to both Primate Labs and the general public how these changes occur, results generated with the tool are not comparable to results generated without it."

The warning will be applied to all Geekbench 6 results for the new Intel chips because Primate Labs currently has "no way to detect if a Geekbench 6 result was run with or without the Binary Optimization Tool." Hence, for now, all Arrow Lake Plus benchmark results will have the following warning attached:

"This benchmark result may be invalid due to binary modification tools that can run on this system."

Primate Labs concludes that, "while the Binary Optimization Tool only supports a small number of Intel CPUs, this is an important step to ensure scores reported on the Geekbench Browser remain trustworthy. Intel lists the supported CPUs on the Binary Optimization Tool webpage. We expect this list to be dynamic and that it will change over time. Primate Labs’ warnings will be updated accordingly."

The immediate question is whether this is fair. Some observers have rightly pointed out that Geekbench runs aren't entirely comparable across different platforms in the first place. More to the point, what Intel's BOT tool is doing is arguably no different to what various CPU architectures do internally.

The Geekbench binary is dynamically linked, which means the actual instructions being run differ depending on the CPU and OS combination in question. Intel's BOT simply makes sure that the instructions generated are absolutely optimal for the underlying micro-architecture. Moreover, when you dig right down, all CPUs effectively translate software binaries into microcode specific to their architectures.

My understanding is that what Intel's BOT is doing essentially amounts to re-ordering instructions so that they fully utilise the Arrow Lake Plus pipeline. All the actual calculations are the same. In other words, enabling BOT doesn't mean skipping any work.

If there is a valid reason to flag results that are BOT-enhanced, it's that they are unrepresentative of how the Arrow Lake Plus architecture performs without BOT enabled. That matters because BOT doesn't automatically work on all software.

Instead, Intel must first assess the software or application and create application-specific profiles for BOT. In an ideal world, you'd have results both with and without BOT enabled, giving a full picture of how the CPU performs both natively and with the tool switched on.

You could also argue that Intel has something of a track record when it comes to, well, tweaking benchmark results. To dredge up just two examples of many, in 2024 SPEC invalidated Intel benchmark results using a certain Intel compiler due to "unfair" optimisations, and in 2009, Intel's ICC was found to be crippling performance on AMD CPUs by deliberately removing all optimisations for the competing CPU architecture.

The other concern is that Intel's BOT sparks an arms race of optimisations that effectively make all of the results unrepresentative of actual CPU performance. With all that in mind, flagging BOT-enhanced results probably makes sense.

With the launch of its Arrow Lake refresh processors, Intel didn't just release some of the best CPUs it's made in years, it also introduced a new software package called Binary Optimization Tool, or BOT for short. Over the past week, I've been testing BOT to assess whether Intel's claims have merit and whether it will become increasingly important over time or fade into obscurity as a neat idea but not worth the time and effort.

But what exactly is BOT? The simple answer is that it's similar to how GPU vendors release drivers with fixes/support for specific games, where certain shaders are replaced by ones that ensure the game renders correctly on the graphics chip. What Intel is doing is rejigging certain thread instructions from a game, such that the CPU can process things more efficiently, making full use of cache, core execution units, and so on.

At the moment, BOT is only supported on desktop and mobile 200S Plus chips, such as the Core Ultra 5 250K Plus and Core Ultra 7 270K Plus, and Panther Lake laptop processors, and while there's a chance it could be brought to older processors, part of the changes in the Arrow Lake refresh were an increased number of hardware hooks in the compute tile.

These give Intel's engineers a better insight into what's going on deep inside the processor, and without those, it's possible that a BOT which supports earlier CPUs just won't be able to offer the same kind of uplift, if any at all.

However, it's also worth noting that Intel has only BOTified 12 games so far, and given that none of the testing, optimizing, and rewriting of instruction scheduling is a simple automated process, we're not going to see Intel churning out BOT updates on any kind of regular schedule.

So if you were disappointed that the tool isn't being offered for your 14th Gen Core processor, for example, you're not missing out on anything if you don't play any of the currently-supported games.

Binary Optimization Tool has been integrated into the already-existing Intel Application Optimization Tool (APO), but as the two things are entirely separate, they can be independently enabled in games that support both options.

Enough jibber-jabber, let's get on with what's important here: how much more performance do you get from using BOT? Well, before we see any figures, let me emphasise something important here: Binary Optimization Tool is all about the CPU, so it won't result in games being rendered any faster, as such, because that's almost entirely down to the graphics card in your gaming PC.

But the CPU still needs to process graphics instructions for the GPU to crunch through, and if those threads aren't being handled quite right for Intel's Arrow Lake refresh architecture, then they might be causing a bit of a performance issue.

However, the rate at which a PC produces frames for display on a monitor is controlled by a number of factors, and the performance of the entire rendering pipeline is determined by the slowest stage in that chain. If that's not the CPU, then no amount of BOTification will improve the frame rate.

The first game I tested was Shadow of the Tomb Raider (SOTTR), as Intel claimed to achieve the biggest improvement in performance with the use of BOT here. The basic hardware setup for all my tests comprised a Core Ultra 7 270K Plus, 32 GB of Corsair Vengeance DDR5-6000 CL30, and a Zotac GeForce RTX 4070. Most tests of this ilk are done with an RTX 5090, but using a mainstream GPU gives you a more real-world view of what BOT can achieve.

SOTTR has a really good built-in benchmark tool that separately tracks the speed at which the CPU processes the game's main engine, as well as rendering instructions, and how fast the GPU churns out frames.

For the sake of simplicity, I've enabled APO and BOT together (the former doesn't make a huge difference with Arrow Lake, as the E-cores are really very good), and the effect is quite dramatic. Together, the optimizations lift the 'CPU game' performance by 35% on average, with the 1% low figure increasing by 41%.

I've done multiple test runs for each game, and in some instances, the SOTTR 'CPU game' improved by as much as 58%, which is quite incredible to see. However, the overall benchmark never gains anything more than 4 or 5% to the average frame rate, and that's because at 1080p High, SOTTR is relatively GPU-bound on an RTX 4070.

But look closely at the above images, specifically the frame times for the GPU and CPU (green line in the bottom graph, lower is faster). Notice how much APO+BOT speeds up the rate at which the CPU crunches through everything in the second half of the benchmark. It transforms the test from being 52% GPU-bound, on average, to 97% GPU-bound; that's exactly what you want from a game.

However, that's just a fancy benchmark, right? It's not the game itself. To that end, I tested Marvel's Spider-Man Remastered, which is a game that Intel only offers BOT for, and it has no built-in benchmark, so the results are 'pure' gameplay.

Over the course of five minutes of swinging around the city, I recorded an 8% higher average frame rate and 7% better 1% lows with BOT enabled, at 1080p High. Those figures would be lower at a higher resolution, of course, but they would also be better with a more powerful graphics card than an RTX 4070.

Alas, these were the highlights of my testing. Cyberpunk 2077 at 1080p High only improved by 2% on average, as did Borderlands 3 (also 1080p High), though the latter's 1% lows actually decreased by 4% (but that game's been twitchy from the get-go). And in Assassin's Creed Mirage and Hitman 3, I experienced almost no improvements with APO and BOT enabled.

In the case of the Mirage, the issue there was certainly the RTX 4070, despite running the game at 1080p High. Dropping the graphics preset to Low and enabling DLSS Ultra Performance saw the combination of APO and BOT improve the average and 1% low frame rates by 3% apiece.

So, what to make of all of this? To start with, Intel's Binary Optimization Tool clearly works as claimed for the most part, and for games that have CPU instructions written in such a way that the Arrow Lake refresh chips aren't handling them as well as they could, BOT is literally a game-changer, as it pushes the performance limitation to where it should be, i.e. on the GPU.

Having your graphics card be entirely responsible for the end frame rate means you can fully control the performance you want by changing graphics settings, using upscaling and frame generation, and so on. However, if a game is already GPU-bound on your gaming PC, then no amount of BOTtastic tweaks will make a difference.

This is most evident in Cyberpunk 2077, Hitman 3, Borderlands 3, and Assassin's Creed Mirage, with an RTX 4070, whereas Spider-Man Remastered picks up a decent enough performance increase with BOT. While 8% isn't overwhelming, the fact that you're getting this from a mere software override is quite impressive.

There is a category of games, though, that would really benefit from Intel's BOT: esports, especially competitive shooters. Pro gamers typically use a combination of the lowest graphics settings and 1080p resolution to get the maximum possible frame rate, and this typically results in the game being almost entirely CPU-bound.

Unfortunately, you're not likely to see Counter-Strike 2, Valorant, or Overwatch getting BOT support anytime soon. That's because Intel's system has to inject itself into the game's code to override instructions, and any software that does that in a competitive game will generally result in the player getting a very swift ban.

Intel says it's working with studios and developers to find a way of implementing BOT that won't cause issues like this, but I suspect that some companies are wary that cheaters will find a way to use BOT to implement a hack and cheat away to their heart's content.

Having spent a long time developing Binary Optimization Tool, Intel certainly isn't going to drop it if esports publishers ultimately say no to BOT, and I reckon that the feature will become even more prevalent when Nova Lake finally makes an appearance (currently on track for later this year, but who knows for sure?).

Right now, though, you can only take advantage of it if you have a Core Ultra 200S Plus chip, and while they're really very good processors, that's going to be a very small pool of PC enthusiasts who will get to see it in action. Current Arrow Lake owners might not be tempted to splash out $200 or $300 on a new CPU, just to use BOT, but you will see some pre-built PCs sporting a 250K Plus or 270K Plus chip over the coming months.

But even so, that's still not going to result in many gamers testing out BOT and giving feedback to Intel about its usefulness. Which is a shame, because the benefits are definitely there.

If you do happen to buy a new Arrow Lake refresh chip, then simply download the very latest Intel Platform Performance Package (IPPP), install that, and you should be able to dive into BOTifying the small number of games currently supported.

A word of caution, though: When I first tested BOT, on a PC with a fresh Windows installation and the motherboard handling the automatic installation of drivers, I found that BOT actually made things a lot worse (up to 25% lower frame rates).

After discussing these issues with Intel, I tried a new installation of Windows, disabled the motherboard's auto-driver feature, and just installed IPPP and the GPU drivers. That resulted in a perfectly working BOT, so if you do have an Arrow Lake PC at the moment and you slap a 200S Plus into it, you might find that the only way to get BOT working properly is by doing the same 'fresh Windows + IPPP-only' installation that I did.

Intel's had a couple of rough years of late, and while it's not totally out of the woods just yet, the likes of Binary Optimization Tool and the Core Ultra 200S Plus processors show that all the internal changes it's made (new staff, restructured divisions, 'listening to the customer') are starting to bear fruit. Let's hope that this new Intel carries on this way for good.