It feels a bit like we're in the long wait before the hardware world returns to normal, doesn't it? Thanks to the global memory shortage, when memory-clad products launch at all they're slapped with dizzying price tags. Given this, hearing that we're only going to see one next-gen GPU next year feels about right. And apparently it won't be a top-end one, either.

That is what we are hearing from prolific and usually very in-the-know hardware leaker Kepler_L2, no less (via VideoCardz). On Anandtech forums they say: "AT2 is the only GPU getting released next year. Everything else from both AMD and Nvidia is 2028."

AT2, if you were wondering, is supposedly the codename for the GPU that is going to be found in Xbox Project Helix, which has been pegged for a 2027 launch for some time. That we'll see at least one new AMD GPU next year, then, isn't news. But what is new is confirmation—assuming Kepler is right—that this GPU is the only new one we'll see next year, especially given it's expected to not be the most powerful of AMD's next-gen line-up.

There's no official word from AMD on this, but rumours have had it for a long while that there'll be an AT0 ('Alpha Triton 0') GPU with 96 compute units (CUs) that will compete with Nvidia's RTX 6080-level GPU. If Kepler's recent statement is true, therefore, it seems AMD will be launching its midrange GPU first, courtesy of console launches.

It would also mean we shouldn't expect any new Nvidia GPUs. Though it's worth noting that at least one hardware leaker—Moore's Law is Dead—disagrees with this and claims we could see an RTX 6090 launch. Kepler disagrees, saying "he's wrong."

Kepler's wording does, however, leave it somewhat ambiguous as to whether Nvidia could release Super versions of current GPUs, given these wouldn't be "new" GPUs but new configurations of the same GPUs. But those have reportedly been cancelled anyway. It's been looking less likely Supers would launch ever since the memory crisis kicked up, given previous rumours had it that RTX 50 Super cards would primarily offer better memory configs.

Nvidia has RTX Spark now, anyway, I suppose. And for my money, given the state of the current PC hardware market, I'm not counting just one new GPU next year as that much of a loss. Roll on 2028, though.

Alas, my series 'I'd do anything for an Nvidia RTX 5090 at MSRP, but I won't do that' has been cancelled after one episode, on account of the fact the card has seemingly disappeared from most major US retailers.

What remains are online listings from third-party sellers, often asking for a price even higher above MSRP, as spotted by Tom's Hardware. Previously, it's been cheaper for me to hop on an international flight to PAX West or QuakeCon, and pick up an RTX 5090 at the event for the $1,999/£1,939 MSRP than buy it anywhere else. This will come as no surprise for those who have been keeping track of our graphics card price watch, where we've seen the RTX 5090 go up 83% in price and even breach the $6,000 sound barrier.

What our Dave most recently recommended as the cheapest option, this Gigabyte Windforce RTX 5090 from Newegg was $5,700 just last Thursday. Now it's a frankly offensive $9050 from a third-party reseller. The same card is sold out on Amazon and Best Buy. But the cheapest new RTX 5090 I've been able to find is the Asus TUF, again from a reseller, for $6,800.

Don't even get me started on how things are looking in the UK—though, for those curious, Dave's tireless efforts did find the Zotac RTX 5090 going for £3,940 from Scan. That's still too rich for my cup of tea.

So, where is all the stock going? Well, some deep-pocketed consumers are definitely snapping up at least a few cards. Recent research suggests a trend that 'defies common wisdom', with folks buying more high-end graphics cards despite rising prices, to the tune of 12.5 million shipped last quarter alone.

Nvidia still enjoys a 90% share of the market according to that same report from John Peddie research. But given company CEO Jensen Huang recently shared that 100,000 Nvidia GPUs were used to train OpenAI's latest model, GPT-6 Astra—not to mention there's already plans to bring quadruple that amount of hardware online—I'll give you three guesses why there are so few RTX 5090s available for consumers.

It's likely not RTX 5090 cards being used specifically to train Astra, but it's still possible that other AI firms are buying up the high-ends cards to adorn their server PCs in lieu of datacenter accelerators or RTX Pro 6000 GPUs. RTX 5090 pallet images shared by HKEPC imply as much (via VideoCardz).

Okay, I'll also give you this for free: Nvidia may not be producing fewer 4nm GB20x Blackwell chips on the whole, but may simply be shifting more of that stock towards its professional products—like workstations—instead. That's still going to leave fewer gaming GPUs to go around, and we're back where we started with runaway RTX 5090 prices.

As GPUs continue to increase in price, it also, of course, becomes increasingly important to feel that those purchases are secure from leaving you with a very expensive paperweight. This is doubly so for the RTX 5090, which now costs upwards of $5,000. Given this, it's important that RMAs go smoothly, but judging by recent comments online, the design of the Nvidia GPUs themselves could throw a spanner in the works.

According to Redditor Willing-Avocado-4830, they recently struggled with Nvidia's RMA process for their RTX 5090 FE because of a faded serial number. They claim the RMA request was rejected. The Nvidia rep's quoted comments seem to have since been deleted or are just not visible for some reason, but the thing to highlight here is the fading itself.

If it were a lone issue, then it might not be as noteworthy, but others in the comments confirm that the serial number on their FE card has also faded—and not only RTX 5090 ones. And indeed, the serial number on our very own RTX 5080 FE, pictured above, appears to have faded somewhat. The number itself is located on the metal part of the I/O side that slots down between the case and the motherboard.

It's probably worth noting that the amount of fading on our version here at PC Gamer might be different to the home user's. We tend to swap between GPUs quite a lot for testing, which could cause more fading for us. Or, if the fading is caused by consistent heat, then it could be that the normal home use would result in more fading. Your guess is as good as mine on that front.

Serial numbers are traditionally found on stickers on the GPU, but some newer RTX FE cards have them put directly on the bottom jut of the PCIe bracket, and they seem to be prone to fading. Presumably the reason this might cause a problem with the RMA process because Nvidia will need to verify authenticity and ensure it's within warranty. And unless there's another sticker elsewhere with it on (on the PCB, for instance, or on the packaging if you still have it), that serial number can be important.

Nvidia's warranty (in its US wording) explains: "This warranty applies only to the original purchases of the Warranted Products from a retailer, mail order operation, or on-line retail store; this warranty will not extend to any person that acquires a Warranted Product on a used basis."

As such, it will be important for Nvidia to verify that the product is, in fact, from an original purchaser.

Looking at what is publicly available from Nvidia about its RMA process, however, it is not clear whether a serial number is required. It's also worth noting that laws surrounding RMAs can vary from country to country.

Whatever the case, if you have a still-legible serial number on your RTX FE card, it might be worth snapping a pic of it, just in case. Though I suppose it's an open question whether that would satisfy Nvidia's requirements if it's faded off the card by the time they receive it for RMA. If not, then surely there must be a better way for Nvidia to plaster those serial numbers on there.

If you've been keeping up with the headlines, then you know gaming hasn't been Nvidia's main priority for some time now. Given that its data centre segment made $89 billion in Q2 of 2026—an eye-watering 93% of the company's total revenue that quarter—it's not super surprising that Nvidia is all-in on AI. What may be surprising though is that, even after this shift in focus, Nvidia is still king when it comes to consumer GPUs.

That's according to the latest report from John Peddie Research; even though Nvidia's market share did decrease by 0.01% last quarter, the company is still holding on to a 90% share of the graphics 'Add-in Board' (AIB) market. AMD has 8%, while Intel enjoys a 2% slice of the pie.

For those a little unclear, an 'add-in board' (AIB) can mean anything you plug into your PC to improve performance, and GPUs definitely fall into that category. When companies refer to AIB GPUs, they're typically distinguishing them from integrated ones like you'll find in laptops, for instance.

The JPR report adds that during the second quarter, the "growth of the global PC-based graphics add-in board market reached 12.5 million units." That's a 10% increase from the quarter before. Not only that, but the likelihood of someone picking up an 'add-in board' in addition to a desktop PC was up to 89%, a 23% quarter-over-quarter increase.

Given that it has been genuinely more affordable for me to hop on an international flight to either QuakeCon or PAX West rather than buying one of Nvidia's RTX 50-series cards absolutely anywhere else, colour me surprised that GPUs remain such a hot-ticket item. I'm not the only one, as Dr. Jon Peddie himself described the trend of high-end GPU sales spiking even as prices rise as straight-up "defying common wisdom."

The report excerpt doesn't go into detail about which GPUs folks are still buying, so there's also a chance that many PC gamers are choosing to forego the latest upscaling and frame-gen tech in favour of an older card that's more reasonably priced. Then again, perhaps not; when we put it to a poll in August, 56% of you were willing to pay over $600 for a graphics card, while 19% would spend over $1000.

Anyway, it's worth comparing and contrasting the GPU increase to desktop PC CPU shipments, which dipped to 14 million units. Considering this is a 10.5% quarter-over-quarter drop, and a 33% year-to-year decrease, it's still pretty stark.

Dr. Peddie himself posits, "Our theory is consumers rushed to buy AIBs before the prices went any higher, as the war in Iran is driving prices up in all segments."

The 2026 Iran War and the Strait of Hormuz crisis has been ongoing since February, making it difficult to secure various components. Constrained supply means rising prices—like those rumoured to be on the way for motherboards. While I've no doubt this could encourage more impulsive purchase decisions, I'm not sure if we'll see a similar shipment spike for motherboards.

Ah yes, melty cables. We've been dealing with them ever since the RTX 4090 pushed the 16-pin 12VHPWR connector (now the 12V-2x6 connector) to its limits. Since then, we've heard of numerous cases of melting GPU connectors—check out Uniko's Hardware on X for many examples—and plenty of attempts at dealing with them.

Now, some modders have said 'screw it' to the new 12V-2x6 design, instead modding an RTX 5090 to run on standard 8-pin connectors (via Wccftech). The modders at TecLab fitted three 8-pin connectors to the back of the PCB using "copper plates, cut by hand", and hooked these up to the front via 24 cables (3x sets of 8).

Once everything was also unlocked to operate at higher power levels, the graphics card was put to the test. And while this was just one batch of tests with one card (and don't try this at home, folks) the results were fantastic.

Even before pushing it as far as it eventually goes, the modders were thrilled with the results (machine translated): "It's getting warmer. It's still colder than a 12VHPWR connector. It's colder than the original connector, guys. Where is Nvidia? Call Nvidia, guys. Put the raw footage out there for people to see, otherwise they'll think it's a lie. This looks like a lie."

Don't worry, TecLab, we've got you. You can see the full footage below.

They even manage to get the GPU running over 3,300 MHz in Port Royal benchmark with just two 8-pin cables, showing that a three-cable setup isn't necessary. (For reference, the stock RTX 5090 boost clock is around 2,400 MHz.) And when it's running at a slightly lower clock speed than this the cables stay very cool.

This is especially important because one of the reasons why the 12VHPWR cable was deemed necessary by many is because it apparently helped prevent having too many standard 8-pin cables connected. But having said that, I'm currently looking at an RX 9070 XT with 3x 8-pin cables and it doesn't seem too overboard to my eyes.

Even with one cable, TecLab gets it running at over 3,200 MHz. Though that single cable does hit 91–92 °C, there's still no melting to be seen.

Saying all this, the modders do push the cables above their recommended spec. Each should have a limit of 150 W, which should mean a maximum power draw of 450 W total. The RTX 5090 has a TDP of 575 W.

But TecLab shows that by unlocking this, the 8-pin cables remain not only stable, but cooler than the 12VHPWR cable, even when they're pushed above spec. They say, "A connector that the industry determines is 150 W but actually it doesn't have a firm specification."

The point here isn't anything to do with the actual overclock they achieved, rather that they managed to push out more power to the GPU than was needed through 8-pin cables—fewer than three of them, in fact—and still have the cables remain cooler than the usual 12VHPWR cable.

Naturally, Nvidia might not want to overrun the official 150 W spec for 8-pin cables. But when there are melty problems with the alternative, and 8-pins seem to be able to step up without issue, at least if this mod is anything to go by, then why not? "Call Nvidia" indeed.

In news that will surprise very few, manufacturers are reportedly planning to raise their GPU prices in China again later this month. This could have a knock-on effect on global pricing for both Nvidia and AMD cards.

Asus looks set to raise the price of its RTX 5070 cards by between 150 and 300 RMB (roughly between $22 and $45) in mainland China, according to Taiwanese outlet United Daily News. Gigabyte has reportedly already raised the price of its RTX 5060 Ti 16GB GPU by 200 RMB ($30).

AMD has increased prices for a number of cards as well; the RX 9070 XT is up by 200 RMB, the price for both 16G and 8G versions of the RX 9060 XT has increased by up to 250 RMB, and the RX 7650 GRE is up by 400 RMB (about $60). This follows reports in July that AMD would be raising the price of its GPU and VRAM partner bundles by 10%.

Your nemesis and mine, the memory supply crisis, is in part to blame. Nvidia has also been reducing its gaming GPU production for some time now, most recently trimming its shipping forecast by between 15% and 20%.

It would be reasonable to assume the company is shifting production capacity to manufacturing even more AI GPUs. For on thing, Nvidia's data centre segment made $89 billion last quarter alone. Nvidia also just announced it plans to acquire AI development platform Hugging Face for an eye-watering $13 billion.

However, the full production picture may be a bit more nuanced. German outlet 3D Center shared some further insight on X, claiming Nvidia isn't producing fewer 4nm GB20x Blackwell chips, but is in fact simply shifting its stock towards the "professional segment" instead.

At the risk of oversimplifying, these Blackwell chips are going towards enterprise products rather than consumer GeForce graphics cards—presumably because the 'professional segment' offers higher profit margins (Is now a bad time to mention that apparently 100,000 Nvidia GPUs were used to train OpenAI's latest model?). As a result, there will be fewer chips to go around for new GeForce cards, meaning shorter supply and higher prices.

Nvidia has done the risk assessment and has determined this is the most sensible way to weather an incredibly volatile hardware landscape. For you and me, though, yes, it's likely going to remain cheaper to fly internationally to QuakeCon or PAX West to pick up a 50-series card rather than buy a bleeding-edge Nvidia GPU anywhere else.

There are reports today that Nvidia is going to be releasing a new variant of its GeForce RTX 5070 graphics card with a cut-down GB203 GPU instead of the GB205 chip it originally launched with. Could this be good news for consumers? Well, no and maybe.

The idea behind using bigger chips to fill out the inventory of smaller cards basically gives manufacturers something to do with graphics processors that don't have enough functional units to power a bigger card. Here we're talking about the same GPU as that used in the far more powerful RTX 5070 Ti and RTX 5080 graphics cards.

The RTX 5070 doesn't currently use the full GB205 die, but its complement of 6144 CUDA cores isn't far off the maximum of 6400 the full chip can offer, so it already has some failsafe headroom built in. Compare that with the GB203 used by the RTX 5080 and that card is using the full silicon package of its chip. That could maybe be seen as quite a gamble on the yields coming from the TSMC 4N process, but it is mature, and Nvidia already has the RTX 5070 Ti, with its 8960 CUDA cores to mop up any GPUs that don't quite fit the full RTX 5080 bill.

But what's a GPU manufacturer to do if there a a bunch of chips from a wafer that can't fit into even an RTX 5070 Ti's cut-down CUDA core requirement? That's where this proposed new RTX 5070 would come in, taking that bigger core, chopping it down to 6144 cores and giving it a new badge.

So yes, don't expect that because it's using a bigger chip that the corresponding RTX 5070 would be any quicker or more powerful, or have a bigger memory bus. Nvidia has done this sort of thing in the recent past—the RTX 5060 has its own GB205 variant—and hasn't given it any greater performance for the privilege of having higher quality silicon at its heart.

GPU price tracking

I've been tracking the lowest price of each GPU in this generation since August 14, 2025

Lowest pricing (Price, $)

Week

Nvidia RTX 5070Nvidia RTX 5060 Ti 16 GB

Lowest pricing Data

Timestamp (Week)	Nvidia RTX 5070 (Price, $)	Nvidia RTX 5060 Ti 16 GB (Price, $)
1	550	430
2	550	430
3	524	430
4	546	430
5	481	430
6	499	379
7	520	430
8	500	430
9	500	430
10	480	430
11	489	400
12	500	400
13	527	420
14	540	420
15	499	430
16	499	430
17	549	430
18	540	450
19	549	450
20	620	520
21	620	520
22	620	520
23	630	520
24	625	520
25	630	550
26	604	550
27	640	570
28	630	515
29	636	500
30	630	500
31	599	500
32	599	515
33	599	490
34	599	550
35	599	550
36	599	550
37	599	570
38	599	570
39	599	570
40	599	565
41	560	560
42	550	570
43	755	650
44	763	780
45	800	773
46	759	570

The recent price spike for both the RTX 5070 and RTX 5060 Ti 16 GB make it tough recommendation against the relatively new RX 9070 GRE.

And just because it's using cast-off GPUs don't expect it to be any more affordable, either. With the RTX 5070 currently retailing for around $800 you can see why Nvidia would be keen to ensure there's good supply of this fourth-tier card when its retailing for so far beyond its $549 MSRP.

Is there any benefits for us PC gamers? If you wanted to be very, very optimistic (as is my wont) then you could say that at least this is going to deliver more supply of the RTX 5070 into the channel. But pricing is unlikely to trend downwards, so there will just be more cards to spend $800 on.

Given that, I wouldn't be surprised to see a variant of the RTX 5060 Ti 16 GB, another ~$800 graphics card, coming out soon using some more of the sweepings off the original RTX 5070 factory floor. If it's good enough for an RTX 5060, it's good enough to turn into an RTX 5060 Ti, after all.

OpenAI began rolling out GPT-6 Astra access to a limited number of organisations earlier this week, with wider subscriber access planned to soon follow. According to Jensen Huang via X, the model was trained on "~100K+ Nvidia Grace Blackwell NVLink72" server racks, and there are already plans to get 400,000 GPUs online in anticipation of whatever comes next.

Greg Brockman, the co-founder and president of OpenAI, told Stratechery that this was "the first run that we’ve trained on more than 100,000 GPUs." That's a staggering amount of hardware (and perhaps leaves little doubt as to how we ended up in a memory supply crisis). But the thing that gives me more pause about Astra's announcement is the Nvidia CEO's assertion that "AGI has arrived."

To recap, 'Artificial General Intelligence' is used to broadly describe an AI model that either matches or surpasses human capability on complex cognitive tasks. Even Brockman admits that "People do have their own definition of AGI, it’s almost this blurry thing." Still, in that same interview he later goes on to add, "Maybe it was the previous model, maybe it’s Astra, maybe it’s the next model, but somewhere in there, I think we’re going to cross most people’s AGI threshold."

According to OpenAI, GPT-6 Astra is "state-of-the-art" on a number of AI benchmarks, including "FrontierMath Tier 4, ARC-AGI 3, and TerminalBench-4.0." A slickly produced trailer attempts to present this performance in more accessible terms, showing scenes of conversational, spoken prompts being used to create presentations, eBay listings, and .STL files for 3D printers. All of the actors in the trailer speak with pretty clear-cut American accents, though personally I'd love to see Astra tapdance for its life against a gorgeous Geordie accent.

As I've already said, access to GPT-6 Astra is limited for now, so it's difficult to look into these 'AGI' claims myself. The timing of this claim isn't lost on me, though—for one thing, rumours continue to swirl that OpenAI's initial public offering isn't that far off. For another, the debut of Astra—a model the company describes as its "most aligned"—follows mere weeks after 'an agentic collective' broke out of OpenAI's isolated testing environment and attacked Hugging Face's servers.

Brockman says, "So much of [Astra's] compute goes into safety and alignment, and we have so much security work that’s gone around it. I think that we’ve done a huge amount of work to deliver this model safely."

Still, I'm not sure I'll be giving GPT-6 Astra access to my eBay account—let alone anything else—for the foreseeable future.

A developer has vibe-coded a new ReShade add-on file that offloads the DLSS Neural Rendering workloads to a second GPU, almost negating the otherwise massive DLSS NR performance hit. Given the original DLSS 5 shots we saw when Jen-Hsun first showed off the feature around GDC and GTC in the Spring were running on two graphics cards, that's not wholly surprising, but this mod certainly does open the the potential for keeping your old GPU around for neural coprocessing in the future.

Like almost all of the current DLSS 5/Neural Rendering implementations we've seen where the user gets to pick the game, the model, and the weight of the effects, the Neural Coprocessor add-on file—mgpu_bridge—is designed for ReShade, an app used to inject different visual effects into PC games. The files exist on Github, can be downloaded and added into a game that's already been modded with DLSS 5 support, and could dramatically improve the performance.

This is, for me, the biggest issue with both the homebrew mods for DLSS Neural Rendering and the only release implementation we've seen from Nvidia and the developers themselves: the hefty performance penalty from turning it on.

Nvidia itself has touted a 50 - 60% frame rate hit as a blanket figure—regardless of game and how much AI filtering it's doing—as what it is expecting to see from enabling DLSS NR. Andy benchmarked NBA 2K27 ahead of launch and found that sometimes the frame rate impact can be even higher than that.

I mean, it's not as bad as when you try and run DLSS NR on AMD GPUs, but I've noted these kinds of performance hits when you run the new feature through ReShade, too. But Marcelo Guibout has coded a ReShade add-on file that allows a second GPU—here an RTX 5060 Ti 16 GB—to take the final finished frame and perform the neural rendering pass on that GPU and output on its own monitor. That means it doesn't have to hand the frame back, which adds its own performance hit and latency.

They initially used a local model—Qwen3.8-27B-Q6_K—running on LM Studio using the same pair of RTX 5060 Ti 16 GB cards they eventually tested the feature on. This created the "initial add-on scaffold" which they then took into Claude to write most of the C++ core, though Anthropic's AI "also produced several confident, well-formatted, wrong diagnoses along the way."

Looking through the Github documentation Guibout makes it clear this is not a production-ready add-on, and is mostly presented for further research and development. They note it has been run on exactly one machine with three games, and encourage everyone to read the source code before running the files. But it's still a really interesting development in the DLSS 5 modding story.

"The reason it is worth doing: neural post-processing saturates whatever device it runs on," writes Guibout. "Run it on the render GPU and lowering your render resolution just frees capacity the neural stage immediately eats, so upscaling stops paying for itself. Move it to a second GPU and the render GPU is genuinely freed."

DLSS NR on AMD

AMD GPUs running DLSS 5 looks great, but runs horribly.

DLSS NR Off

DLSS NR On

This is almost certainly the reason behind Nvidia originally running the feature on two GPUs for the original demos, and the fact that it can get it running on a single GPU is still impressive from a technical perspective, but having a coprocessor handling the load should give you the best of both worlds.

Though there are obvious caveats to this.

The first being that it's expensive enough to buy one graphics card right now. You look at this mod and the fact it's been created and tested on a pair of RTX 5060 Ti cards and think that's kinda mid-range. They are the fifth-tier cards of Nvidia's RTX Blackwell stack. But those are two GPUs that now cost something around $800 apiece, so we're still talking about needing $1,600 worth of graphics silicon to get this working.

The other caveat is that we're running on with two monitors here as a necessity. Guibout notes in the Github readme file that, "The second monitor is a requirement, not a nicety - the neural output is displayed by the card that produced it. Headless works and is meaningfully slower."

The third caveat here is that this ReShade implementation is a final step process. It is handled after the final frame has been rendered and works with none of the depth and motion vector information the official Nvidia implementation runs with. That means the second-GPU is going to have to be more tightly integrated into the actual rendering process. Not quite as tightly bound as with SLI, perhaps, but it will surely still need to get involved before the final post-processing is done.

But I could potentially see this as a way to do something useful with your current-gen card when you next upgrade. And, on a purely mercenary level, it might be a smart thing for Nvidia to actually make as a feature as it would have the effect of locking someone into the GeForce ecosystem even more securely.

Though times have changed and SLI and CrossFire have died, and so motherboards are even less likely today to have multiple PCIe x16 sockets available to stick secondary GPUs into. Especially if you've got multiple NVMe SSDs taking up precious PCIe lanes.

I'm loving the rate of development happening outside of Nvidia now that the feature has fallen into the hands of the community at large. And regardless of how you feel about the technology, I would encourage you, if you have the hardware, to take a look at how even just the ReShade implementation looks, and how you can alter it. It gives a fascinating insight into the technology, into its limitations as well as its unwelcome excesses.

Though it's not always easy. I've spent the morning trying to get this ReShade add-on working with the test rig, because I was interested to see if it could be run asymmetrically, as I don't have a pair of RTX 5060 Ti 16 GB cards. Sadly, with an RTX 5080 and RTX 5060 Ti as the neural coprocessor, I could not get the add-on to arm when in Cyberpunk 2077. So, I've got more work to do to get this thing running.

Hello, and welcome to another episode of 'I'd do anything for an Nvidia RTX 5090 at MSRP, but I won't do that.' This time, 'that' is hopping on an international flight from the UK to Seattle, Washington in the US, so I can buy a 50-series card at PAX West.

Nvidia's self-described GeForce Evangelist Jacob Freeman wrote on X, "Hello PAX West! VPA IRL is here! Come find me if your looking for a GeForce RTX 5090, 5080 or 5070 AT MSRP! While they last."

Unlike QuakeCon earlier this year, there didn't appear to be a Nvidia booth or any boxed Founder's Editions purchasable at the event itself. Instead, Freeman explained that he was giving out cards with "a code that you use to purchase from the Nvidia web store." For context, 'VPA' stands for Verified Priority Access, which in the past has been used as a draw system that only granted gamers the chance to buy a full-price GPU.

Now, it's a little late for me to hop on a plane as PAX West ran from September 4 until September 7. But if I'd gotten my act together, hopping on a short-notice flight to the West coast for approximately £1,000 (about $1,352), plus grabbing the most expensive PAX day-pass for $89, then picking up the RTX 5090 for $1,999 MSRP would have totalled $3,440—and that's still significantly cheaper than the $5,500 price tag we've most recently seen at Best Buy as part of our graphics card price watch.

That is a ridiculous state of affairs, but given that Freeman was out of 5090 and 5080 cards by the end of day three, clearly the maths is maths-ing for some. I barely remember the last time Nvidia mentioned gaming at all in an earnings call, but this runaway pricing doesn't appear to be hurting the company.

For a start, Nvidia's AI profits completely overshadow what it makes from gaming. In fact, the company is dining so well on AI's demand for hardware like Vera Rubin, that it recently announced its intention to acquire Hugging Face for $13 billion. Purchasing this development platform will likely further shore up the company's prime position within the AI industry.

It's old news, but Nvidia hasn't been a gaming company primarily for some time now. As such, it can kind of do what it likes with its gaming business (Notwithstanding the memory supply crisis). As such, I suspect we're far from seeing the last of 'VPA IRL' stunts like those at QuakeCon and now PAX West. Meat Loaf, anyone?

Another day, another DLSS 5 story that actually kinda doesn't involve Nvidia itself doing anything, because it seems now AMD GPU owners can get in on the act. A new homebrew app has been released on Github that will allow you to run the DLSS Neural Rendering feature on a Radeon GPU. I've tried it and it works, but damn does it hurt the poor Radeon silicon.

From having DLSS 5 .dll files plucked from an early access version of NBA 2K27, we've had modders working hard to create surprisingly polished ways to essentially get the tech working in practically any game, even if it doesn't natively support traditional DLSS itself.

I've been playing around with it a lot this past week, even to the point of jamming it into Football Manager 26 just to see what would happen. And, y'know what, I kinda like the result so I'm keeping it. I wasn't expecting that, and I also wasn't expecting this technology, created by Nvidia, to appear so quickly on AMD GPUs. Historically we've found that where Nvidia goes, so AMD follows, at least in the GPU feature space, but it seems like AMD might not actually have to do much this time. Well, apart from actually make the silicon a little more tensor-friendly.

Because I have spent the afternoon trying to get the new app from user danielblnc on Github to work on my Radeon RX 9070 XT in Cyberpunk 2077. Essentially you download the DLSS-NR-on-AMD file, drop it into the relevant folder (the bin/x64 folder in the Cyberpunk 2077 directory in Steam, for example) and run it. So long as you have the necessary nvngx_dlssnr.dll file that's been doing the rounds installed in the same place it ought to work.

It didn't.

Initially Windows flagged it as containing a virus, but I did some checking around with VirusTotal and seemingly the Defender flag is a machine learning false positive. Probably. But even then I still struggled to have it running happily.

I am not entirely sure what I did wrong, but when I enabled the feature in-game all I got was a black output where the 3D world ought to be, even though the HUD and game kept on running. After some file deletion, redownloading, verifying of integrity (mine and Cyberpunk's), and a rollback of my AMD Adrenalin drivers, I finally got it working. Kinda. I mean it's still kinda crashy, and I can only now launch Cyberpunk 2077 directly from the .exe file rather than from Steam, but it works.

But my poor GPU wished I hadn't. Now, we know that DLSS 5 is a hell of a resource hog on an RTX 50-series GPU that it's designed for, with a touted 50 - 60% performance hit officially noted by Nvidia, so I should have known it would trouble the AMD GPU.

DLSS NR on Nvidia

DLSS 5 was designed for GeForce GPUs, so is a little more capable, but still a massive resource hog.

DLSS NR Off

DLSS NR On

Still… running at 4K RT Ultra settings is tough enough, and I was getting just 61 fps on the RX 9070 XT even with FSR 4 set to Balanced. And then flicking the DLSS Neural Rendering toggle to 'on' made that figure drop to just 9 fps.

And I thought the 73% drop Andy found with NBA 2K27 was rough.

DLSS NR on AMD

AMD GPUs running DLSS 5 looks great, but runs horribly.

DLSS NR Off

DLSS NR On

Now it's worth noting that the fact it is running at all on AMD GPUs is impressive on its own, but it's not going to make any Radeon graphics card owners any happier about it when they're getting unplayable frame rates out of their new ill-gotten pretties. An 85% performance drop is not sustainable. Nice for screenshots though, eh?

The issue is that AMD's RDNA 4 doesn't have the level of dedicated silicon for the sort of matrix calculations the feature is running… in the same way that RTX 20-series chips don't.

The Turing architecture does have Tensor Cores, but they're far weaker compared to those in the RTX 40- or RTX 50-series, and the shared resources AMD plucks from the standard RDNA 4 Compute Units for these sorts of calculations aren't as robust either. Hence the weaker performance.

Will they be able to squeeze more performance out of this feature on AMD chips? I don't know, but YouTuber Ancient Gameplays noted the app in a recent video, and states that performance has gone up 12% since the first version of the DLSS-NR-on-AMD software first hit Github, so maybe there's a chance. But, realistically, I think there's a hard hardware limit in there which is going to stop DLSS Neural Rendering from being a realistic part of AMD's armoury in this generation.

If you happen to pick up a copy of NBA 2K27, in your eagerness to try out the first official implementation of Nvidia's DLSS 5 AI image tech, you won't find it labelled as such in the settings menu. Instead, it appears as a simple on/off toggle under the label "DLSS Neural Rendering." And while that might seem like a small change, I think it makes a huge amount of difference in terms of public perception.

Calling it DLSS 5 in the first place was, at best, confusing. Traditional DLSS, as in the upscaling/super sampling technology, is well established at this point. While some still refuse to get on board, many have become accustomed to the idea that it works well at its primary task, running your games at lower resolutions to achieve better frame rates while outputting a high-resolution image.

So much so, that DLSS updates are usually pretty well received. Slightly better image quality, slightly better performance. Ray Reconstruction niceties. That sort of thing. Given that DLSS is now a near requirement for smooth performance in many games (and I'm not opening that can of worms any further right now, believe me), the gradual, iterative improvements have been pleasing to see.

But DLSS 5 changes things. Quite literally. And while Nvidia has gone to great pains in recent months to point out that it's both developer controlled, and a toggled option that can be safely ignored if you don't like the imagery, shoving the tech underneath the DLSS iterative umbrella felt like a poor move.

Something in the association inferred, even if it was later made clear that it wasn't the case, that AI image alteration tech was just how games were going to be going forward, and you were going to have to get on board if you wanted to keep using DLSS in future.

And what happens to DLSS 4.5, the current version of the upscaling tech? Does it simply update to DLSS 4.6, 4.7, etc, until we start getting into more and more decimals to avoid the big 5.0?

I asked this exact question to several Nvidia staff at my Gamescom visit, and received little in the way of a satisfying answer. It was made clear, however, that regular DLSS upscaling will continue to receive consistent updates, separate from DLSS 5.

Now it seems like we might have a clue to the solution of this branding mess. In NBA 2K27, it's been labelled as DLSS Neural Rendering, and it's an on/off switch that can be used independently of the upscaling, or ignored entirely for those who don't want to get on the AI image alteration train.

Our Dave first noticed this change in the leaked DLL files last week, which refer to it as "Nvidia DLSSNR." But the fact it appears in an official game settings menu under the "Neural Rendering" title feels like a step in the right direction. Nvidia seems painfully aware that it needs to rehabilitate its new tech, and separating the nomenclature would be a good start.

Of course, the naming scheme is only part of the story when it comes to the controversy. There are many who are dead set against the idea of any sort of AI generation tech entering their games, amid a wider AI backlash. Nothing's changed there, I would wager.

Separate from that, whether you like the look of the AI-altered imagery is still a matter of taste. It's developer controlled, as Nvidia has repeatedly pointed out—though if you start modding it into games yourself, like Dave has, you do get your own sliders to play with.

DLSS 5 in Cyberpunk 2077

The effect on the lighting of the character model is impressive and easy to mod in. Just computationally expensive...

DLSS 5 off

DLSS 5 on

Still, I've got a feeling we're going to see some relatively subtle implementations to start with—and inevitably, some goppingly ugly, AI beauty filter-like imagery as time goes on. There's no accounting for taste in a general sense, and there's certainly no accounting for taste when it comes to graphics.

It's also, in its current iteration, a massive performance hog. Nvidia claimed a 50-60% performance hit, but my early testing in NBA 2K27 shows it can hamstring an RTX 5080 by up to 73% at certain settings. That's a very powerful (and expensive) graphics card, held back significantly by Nvidia's new toy.

All of which means the DLSS 5/DLSS Neural Rendering controversy is likely far from over. But I can't help but feel that if the tech was unveiled as "DLSS Neural Rendering" in the first place, as just a new optional feature and not presented as the entirety of DLSS 5, a fair amount of the confusion—and perhaps even initial backlash—could have been avoided.

DLSS 5 is finally here. Well, its been officially released in one game, anyway, after modders got their hands on it last week and made the actual launch somewhat redundant. Still, I've been testing Nvidia's new AI image-enhancing tech in NBA 2K27, the latest hoop-shooting game in the long-running series, and I have bad news: even on a high-end RTX 5080 rig, it's seriously punishing when it comes to performance.

Which isn't a huge surprise, given that Nvidia itself estimated a 50-60% DLSS 5 performance hit in the briefings last week. Still, my early testing shows it can tank your frame rate considerably harder than that, even on a PC that costs well over $3,000 in 2026.

I threw a pre-release version of NBA 2K27 at one of our benchmarking rigs, an AMD Ryzen 7 9800X3D, RTX 5080-equipped, 32 GB DDR5 behemoth. And while this particular PC can make mincemeat of the game's relatively lightweight hardware needs overall, turning on DLSS 5 can drop the average frame rate by up to 73% in certain scenarios.

It's capable of looking very pretty in this particular game, that's for sure. But my jaw has dropped several times during my testing, and not always for the reasons Nvidia may have hoped for. To the benchmarks, then.

DLSS 5: 1440p NBA 2K27 benchmarks

Given the RTX 5080 is Nvidia's second-fastest consumer graphics card, I had no qualms about turning up the settings to Ultra, including enabling ray tracing. I recorded my results during actual gameplay within NBA 2K27's quick match mode, making sure to play for long enough that cutaway shots to the player's celebrations and the crowd triggered several times.

It's worth mentioning, however, that given the lack of repeatability within this mode, these results are potentially more representative of the DLSS 5 average frame rate hit, rather than game performance overall.

DLSS 5 | RTX 5080 | 1440p gaming benchmarks

Results recorded during gameplay in NBA 2K27 at Ultra settings, ray tracing enabled, using an RTX 5080.

NBA 2K27 (Ultra preset)

Avg FPS

1% Low FPS

Native

230

93

Native + DLSS 5

93

37

Quality upscaling + 2x FG

363

153

Quality upscaling + 2x FG + DLSS 5

168

75

0100200300400

NBA 2K27 (Ultra preset) Data

Product	Value
Native	230 Avg FPS, 93 1% Low FPS
Native + DLSS 5	93 Avg FPS, 37 1% Low FPS
Quality upscaling + 2x FG	363 Avg FPS, 153 1% Low FPS
Quality upscaling + 2x FG + DLSS 5	168 Avg FPS, 75 1% Low FPS

And my goodness, what a performance hit it is. The difference between the Native 1440p average frame rate and the average rate with DLSS enabled is a whole 137 frames. A 93 fps average still results in very smooth gameplay overall, but the drop to just 37 fps in terms of 1% lows is very troubling.

This is an RTX 5080 running at 1440p, after all. A 4K-capable graphics card that will currently cost you $1,550 thanks to the ongoing memory/component pricing crisis. And that figure is based on the cheapest we can find—I've seen them listed for $2,000+ in the past month or two.

Of course, Nvidia will tell you that upscaling and frame generation are here to save the day when it comes to smooth performance, and that proves to be the case. Still, watching a 363 fps average result plummet to 168 fps at the push of a DLSS 5 button is pretty jaw dropping.

Nvidia has confirmed that DLSS 5 performance improvements are coming in future, and by the looks of it, it's in desperate need of some. It's often very pretty at these settings, though, as I'll go into below. So... that's something, I guess?

DLSS 5: 4K NBA 2K27 benchmarks

There are no awards for guessing that, at 4K, the effect is even more pronounced. At Native resolution with no upscaling or frame gen assistance, the RTX 5080 spits out an admirable 183 fps average (still at Ultra with ray tracing enabled, no less), which duly plummets to 49 fps as soon as DLSS 5 is enabled.

That represents a staggering 73% decrease in average frames, and an instantly crunchy experience. Sure, 49 fps is perfectly playable, but with the knowledge that NBA 2K27 would be capable of so many more frames with the AI-enhanced pretties turned off, I'm not sure I could grin and bear it.

DLSS 5 | RTX 5080 | 4K gaming benchmarks

Results recorded during gameplay in NBA 2K27 at Ultra settings, ray tracing enabled, using an RTX 5080.

NBA 2K27 (Ultra preset)

Avg FPS

1% Low FPS

Native

183

91

Native + DLSS 5

49

31

Quality upscaling + 2x FG

255

118

Quality upscaling + 2x FG + DLSS 5

92

53

075150225300

NBA 2K27 (Ultra preset) Data

Product	Value
Native	183 Avg FPS, 91 1% Low FPS
Native + DLSS 5	49 Avg FPS, 31 1% Low FPS
Quality upscaling + 2x FG	255 Avg FPS, 118 1% Low FPS
Quality upscaling + 2x FG + DLSS 5	92 Avg FPS, 53 1% Low FPS

What I would do, however, is turn upscaling and frame generation on to compensate. And here, the news is slightly better. A 92 fps average is more like it, with regular old DLSS set to Quality, and 2x frame gen enabled. I feel better already.

That was until I realised that, with DLSS 5 turned off, I'd be getting a 255 fps average at the same settings. That's a 163 fps drop with DLSS 5 enabled, for those counting. Which is a lot of frames to come tumbling out the bottom of your very expensive graphics card.

Given all of Nvidia's benchmarks are with a lower resolution DLSS setting and the full 6x Multi Frame Generation bean, on the one hand I expected DLSS 5 to actually need MFGto achieve smooth frame rates at all, or require Ultra Performance-level upscaling. So, the fact that I can achieve very playable frame rates with just 2x frame gen and upscaling set to Quality is still pretty impressive. But on the other, NBA 2K27 runs so well without DLSS 5 enhancement, the thought of slashing my frame rate by such a huge amount for some prettier, AI-assisted graphics stops me in my tracks. Speaking of which...

DLSS5: Image quality comparison

There's one thing I will say for DLSS 5: it looks pretty great in this game. I spent a good hour or two zipping around in the photo mode, marvelling at the visages of my ill-fated basketball team.

The photo mode, however, is where you'll see most of the benefits of Nvidia's new toy. In terms of actual gameplay, you'll spend most of your time in NBA 2K27 in a zoomed out perspective, cursing loudly at the fact you can't time a well-placed three-pointer for toffee.

Sure, there are cuts to player celebrations, slow motion replays, and half-time team talks. But it's when things are paused that you'll see most of the DLSS 5 difference—so that's exactly what I've done for the comparisons below.

NBA 2K27

A comparison of two screenshots in NBA 2K27 photo mode, one with DLSS 5 on, and one with DLSS 5 off.

DLSS 5 on

DLSS 5 off

My early concern about DLSS 5 was that it seemed to produce overblown, AI beauty filter-like imagery, which I personally find very ugly. However, as I pointed out in my DLSS 5 sneak peak, the implementation in this particular game is often much more tasteful.

The player above looks more realistic with DLSS 5 turned on. The highlights on the nose and forehead are shinier without looking completely blown out, which, combined with the more realistic shadowing on the facial features overall, creates a very pleasant image.

His beard looks a little more lifelike, as does his hair. And I'd pay special attention to his jersey, which has extra wrinkles in the fabric in the DLSS 5-ified image that give it a more 3D-like effect. Because you'll be paying attention to your team's ironing mishaps in your new basketball game, obviously.

NBA 2K27

A comparison of two screenshots in NBA 2K27, one with DLSS 5 turned on, and one with DLSS 5 turned off.

DLSS 5 on

DLSS 5 off

Switching over to two players together, I think this particular comparison shows off DLSS 5 at its best. Both of their faces look muted and poorly shaded in the non-DLSS 5 version, but swing the slider over and they suddenly come to life.

As your graphics card screams bloody murder, perhaps. But as much of a cynic as I am regarding the idea of AI "enhancement" of any sort, I can't deny that DLSS 5 in close up is capable of producing some lifelike human beings.

Once game developers have tweaked it for themselves, of course, and those settings sliders have been reined in from some of the horrors we saw in the earlier demo footage.

It's far from flawless, though. While some character models look noticeably more realistic, others get some heavy-handed AI treatment that isn't quite so seamless.

Above: NBA 2K27's photo mode, with DLSS 5 turned off to start, and then on and off again in sequence.

Take the footage above, for instance. While I would say there are some definite shading improvements once DLSS 5 is turned on, there's also a kind of overblown, over-contrasted effect to some of the facial features in this particular example that looks slightly off to my eyes.

NBA 2K27

Two screenshots showing the crowd in NBA 2K27, with DLSS 5 turned on and off.

DLSS 5 on

DLSS 5 off

And finally, here's a comparison between the crowds. You're going to need to fullscreen and zoom in on this one, but it's fairly representative of how you'll see the match attendees when you're actually playing the game. Aside from the odd camera cut, they're mostly there for background atmosphere.

And yes, the folks in the cheap seats get a miniature dose of DLSS 5 image enhancement, too. The effect here is very subtle, but it's probably most noticeable on the courtside commentator's face. Which will go past in a flash as you actually play, but hey—check out that ever-so-slightly more chiselled jawline.

So, is DLSS 5 actually worth turning on in this game? Well, I can't help but be impressed by the imagery (for the most part), and this relatively light-touch implementation adds a lot of character to individual members of your team. They're also noticeably sweatier, too, if that's your sort of thing.

But you're going to want plenty of frame rate headroom before you even think of touching the DLSS Neural Rendering checkbox, and that means running a very powerful rig—or cranking up the DLSS upscaling and frame generation to compensate. NBA 2K27 is a very fast-paced game, and the experience suffers quite badly when things get crunchy. You'll want smooth frames for this one, believe me—and if that means turning off DLSS 5, I say go for it.

Still, I can absolutely see why Nvidia picked this game for DLSS 5's first official outing. NBA 2K27 feels well optimised with the AI frippery turned off, and it's not the sort of thing that would otherwise be bothering most modern rigs.

But if my testing is any indication, DLSS 5 can look really good, but still be a massive drain on performance even in a properly optimised game. And I really question who among us will be prepared to make the frame rate trade off, even in a relatively lightweight game like this.

The real test will be in something properly demanding, and it's there where I think the wheels will come off. Better image and lighting quality is all very well, but for this sort of performance cost? I dread to think what it'll do to the already-addled frame rate of something like Borderlands 4, for example.

DLSS 5 is capable of looking much better than I expected, which is something of a relief. But when it tanks the frame rate this heavily, I can't help but feel that it's a technology designed for the next generation of graphics cards, and not the versions we're running today. Even the $1,500+ ones—and how many of us have the spare cash right now for another one of those?

Ever since the Control mod appeared I've been enjoying jamming the released DLSS 5 .dll files into a bunch of different games, to greater or lesser effect. Some games it works well in, some it doesn't quite take and you end up with visuals that look a lot like the characters have worms writhing around under their skin.

Though, I have to say when I stuck it into the review version of The Blood of Dawnwalker and it failed to hook the DLSS pipeline that look kinda worked for my semi-Vrakhiri. Bit of extra undead chic with the game failing to run the DLSS neural rendering feature with actual motion vectors or scene depth.

With the necessary files being released in the early access version of NBA 2K27, they've now been unleashed across the internet and modders have been busy getting game-agnostic versions ready for people to squeeze into any title they fancy. And while the first iteration was focused on Control and then games running DLSS already, there have been mods released—particularly the RenoDX and specifically the ShortFuse mods—which allow you to experiment with pushing DLSS 5 into practically any game you like.

And it's surprisingly easy to do, though not without its awkwardness, too. Well, such is the way with unreleased features shoehorned into games they're not meant to be in, but also such is the way with PC gaming.

The thing that has stood out to me—and something that Andy first noted in his hands-on look at the official technology at Gamescom—is that DLSS 5 really can produce some outstanding visual effects… if used sparingly. Pick the right model of the three (for me, that's Model C) and drop the intensity of the effect down to around 60 - 70% and you end up with scenes and character models that look better than the originals, but without the over-saturated, overblown very AI look of it dialed up to 11, nor without it looking like it's grinding away or adding its own geometry.

But it's also clear it's one hell of a resource hog. Nvidia has been very clear about this from its recent presentations at Gamescom and a briefing I had at the end of last week; DLSS 5 represents about a 50 - 60% drop in frame rate regardless of the game. Though, sometimes we've seen that performance hit be even greater, especially at 4K.

Running the Control mod on an RTX 5090 I was seeing DLSS Balanced 4K Ultra Ray Tracing running at around 110 fps in standard mode, which then dropped to under 50 fps as soon as DLSS 5 was brought to bear on the game, which is about where Nvidia is pitching the performance hit.

That, in all honesty, is to be expected. When Nvidia first showed off the tech around GDC it was running on a pair of RTX 5090s, so the fact that it can be run at all on single lower-spec GPUs is impressive. But it was also going to be punitive for current generation hardware, which only have tensor cores enough for standard DLSS upscaling.

DLSS 5 in FM26

Okay, it's not photorealism, but I think it looks better.

DLSS 5 off

DLSS 5 on

What I wasn't expecting, though, was to be able to stick DLSS 5 into any game, regardless of whether it had DLSS in there in the first place. Which meant that I, with my 33 year Football Manager addiction, had to see what it does to FM26.

And, well, I'm going to be leaving it enabled. It's not that it suddenly makes the game look photorealistic—as Nvidia itself has said, you need a really good input source to get a really good output from DLSS 5—but on its own it adds in a bunch of things that you otherwise have to mod in yourself from a bunch of other places. And because it's just added in last, and FM isn't hugely demanding, I'm still happily locked to 60 fps anyways.

Those little computer people don't need a whole lot of processing.

But it does improve the environmental lighting of the matches. It makes the pitches look better, there's a tiny amount more detail on the tiny people, on the pitch and in the crowd, and it adds in some improved shadows. But it's not just in the 3D engine, because it also works its neural rendering magic on the player profile pics, too. Which is especially helpful for making the regenerated players look less like plastic Action Man figures (or GI Joes, for my American cousins).

DLSS 5 in FM26

You're getting better regen faces without copy/paste mods.

DLSS 5 off

DLSS 5 on

Normally you'd have to use a mod to get more realistic player faces, and even then they're still just a bunch of AI generated profile pics which get disseminated among the database so you still get identical twins from different parts of the game world. With DLSS 5 I'm just getting a little extra facial detail and realism for my new potato-faced folk.

They're still not exactly photorealistic, but I'll take it.

How to enable DLSS 5

DLSS 5 in Cyberpunk 2077

The effect on the lighting of the character model is impressive and easy to mod in. Just computationally expensive...

DLSS 5 off

DLSS 5 on

To begin with it was all a case of finding a Discord invite to the RenoDX channel, finding the right .dll files and downloading the necessary RenoDX DLSS 5 mod. Then you had to play around with getting it and Reshade installed in the right directories and crossing your fingers.

Now, it's far more straightforward. The cleanest installation method I've found is to use the ReShade HDR Installer (RHI) from Github. This does all the hard work for you—run it as Administrator and you can just pick from a list of installed games on the left hand side of the UI, install ReShade, and then pick one of the DLSS Tools from the Shaders/Addon button at the top, or from sliding over to the specific Neural Rendering section that has just been added. This should install the necessary .dll files in the right places for your game to work.

But, it must be noted, that this is all predicated on an as yet officially unreleased feature being jammed into unsupporting games. So you might find your mileage varies on just how much messing around you might need to do to get it running perfectly. If at all.

When you boot into the game you can then hit 'Home' to bring up the ReShade in-game menu and slide over to the RenoDX tab to access the DLSS 5 options. Now, these are options and sliders that aren't going to be available in the release versions of DLSS 5. That's something important to note, as officially DLSS 5 is going to be an on/off toggle switch so that you just get the version of DLSS 5 the game devs themselves have tuned.

The ReShade option puts those sliders in your hands, but with less fine grain control over where it's deployed and how intensely. Which is going to make the whole maintaining artisitic intent thing a tricky argument. There has been resistance to DLSS 5's AI filtering, with some folk suggesting that it tramples on the original artistic intent of the developers. So Nvidia has spent a lot of time since its announcement talking about the level of control devs will have themselves, and that it will only be in their hands with no sliders for us gamers. These mods, however, change that giving users a level of control not officially sanctioned, which means gamers themselves will get to choose how much of the devs artistic intent they want to retain.

Anyways, you ideally want to use the Upscaled or DLSS version of DLSS-NR rather than the 'Present' method (if you're using the ShortFuse mod). Present is the version that runs when the game frames are presented rather than inside the actual graphics pipeline. That means you lack the motion and depth information DLSS 5 relies on to be effective. If it's saying 'Waiting' then something's gone wrong.

I had this with a couple of games—it allows you to test the overall look, but it's terrible in motion—and had to mess around putting files in different places to get the mod hooking in properly. Switching between the two different RenoDX DLSS 5 tools generally helped.

I walked into Nvidia's DLSS 5 demo last week with some preconceptions, I'll be honest. I hadn't enjoyed the images it had produced to date, and I was sceptical about the benefits it could bring to our beloved games. Not to mention the troubling notion of AI-altered images entering into our hobby, even if you can turn them off with a simple checkbox.

The good news is, DLSS 5 is capable of looking pretty good once the settings have been tweaked by actual game developers. The very bad news is, Nvidia says turning DLSS 5 on will currently reduce performance by 50-60%. And that's across the board—no matter the game.

Nvidia was keen to point out that, just six months ago, DLSS 5 required two RTX 5090s to run. So, a 5x improvement since then is something to hold within our collective hats. But tanking frame rates is very unlikely to improve DLSS 5's perception within the PC gaming community, that's for certain.

Looking at the basic performance charts Nvidia has provided us so far, the ultra-high frame rate figures are based on 6x Multi Frame Generation. So, while the little RTX 5060 can churn out a 258 fps result in NBA 2K27 at 1080p Ultra settings with ray tracing and DLSS 5 enabled, the game underneath will actually be running at around 43 fps.

At 4K, the same game settings result in a 370 fps average on an RTX 5090, or a 233 fps average on an RTX 5080. That's with Performance-level upscaling, though, and with 6x MFG doing some seriously heavy lifting once more.

If you've spent thousands of bucks on a high-end graphics card, I'd imagine that running games with the upscaling and frame generation settings near their limits to achieve high frame rates is probably not what you had in mind.

All of which makes me wonder if the AI-ification of DLSS 5 is really the biggest barrier to its acceptance. I have my own deep, dark concerns about DLSS 5, AI in general, and its hooks making their way into our hobby. But if the tech is capable of slicing the frame rate in half, all of this might be a moot point.

The Nvidia folk I spoke to were keen to point out that performance improvements are already on the way. But given that so many gamers are stuck with older graphics cards, thanks to the ongoing pricing crisis that seems destined to become much worse before it gets better, I do wonder who the tech is currently for.

When ray tracing first arrived on the scene, its performance-tanking qualities were at least (partially) offset by the introduction of upscaling. But all these years later, I still debate for myself whether it's worth turning on when I play games on my own, relatively high-end gaming PC. Our rigs might be better equipped on average, but ray tracing is still a performance hog in 2026—and many gamers I speak to still turn it off for smoother frames.

And while DLSS 5 is certainly more... transformative than ray tracing in terms of the images it creates, it's also far more controversial. Even if you're prepared to make the jump into AI-enhanced games (and it's quite the leap, let's be honest), putting those concerns in tandem with massively compromised frame rates is a seriously big ask for even the most curious of gamers.

I expected to be writing about the controversy surrounding Nvidia's new tech in terms of the wider atmosphere of AI discontent. But here I am getting down to brass tacks. In its current form, DLSS 5 seems to hit your graphics card like a freight train—and whatever your opinion on AI in gaming, nobody wants that.

I think it's fair to say that DLSS 5 is the most controversial Nvidia creation… well, ever. After the huge amounts of backlash it received in reaction to its unveiling earlier this year, Nvidia's AI image enhancing tech has been mostly kept under lock and key.

However, myself and a select group of journalists were ushered into a small room in the Nvidia booth at Gamescom last week, amid some furtive glances and hushed whispering. As the lights went down, it became obvious as to why. Here was the latest version of DLSS 5 running in front of us, the controversial graphics tech in the flesh. And the message Nvidia wanted to get across was deliberate, and strongly put: It's all about developer control.

I was shown a character sitting within the Nvidia Zorah demo, the same shown at SIGGRAPH 2026 earlier this year. As he sat, looking around at his environment in a bored fashion, I leaned forward in my chair. DLSS 5 was obviously altering the image to a significant degree, but seeing the results in person, it was clear this was nowhere near the "yassified Grace" level of AI enhancement shown in the first reveal.

Our digital example, casually inspecting the back of his hand, looked a lot like the original character model—but with more photorealistic lighting, and plenty of finer details that seemed fitting for the scene. It was nowhere near as brash an image as I was expecting, I'll say that much for free.

At the side of the screen was the DLSS 5 developer control menu. Two sliders, marked "structure intensity" and "tone intensity" were set somewhere around the midway mark—and at these settings, the character looked pretty damn good.

There are more tools for developers to play with, too. The masking feature allows devs to select individual items in a scene (say, a fruit bowl on a table) and give it the AI-enhanced treatment, while leaving everything else DLSS 5-free. The varying AI models can also be switched in and out (or on and off) on a whim.

According to Nvidia, the fine-tuning abilities of DLSS 5 are designed to be as developer-friendly as possible, allowing them to dip in and out of its features on a model-per-model basis to suit their artistic intent. Seeing the demo running in front of me with my own eyes, I could feel myself coming around to the idea. My arms uncrossed.

And then my hosts began messing with the sliders and the model presets, and an essential truth emerged. While a light dusting of DLSS 5 can cause a substantial visual leap in terms of photorealism, turning it up to 11 can still provoke an unwanted, gut-level reaction.

There are three image-transforming AI models for devs to choose from (at least in the demo I was shown): A, B, and C. While models A and C looked pretty good to me at various settings, model B made me wince when both sliders were pushed up. The HDR-like shadows became overblown, the skin textures too shiny, the facial features too keenly drawn.

To my eyes, characters with model B turned up can look substantially different from the model underneath, in a way that's reminiscent of the AI beauty filters we've all become far too familiar with. In short, it's still capable of giving me the ick.

I'll call it as I see it: DLSS 5 turned up to the guns, particularly using model B, can look downright ugly. But Nvidia's point makes a whole lot of sense. It's the developers, the artists behind the game, that choose how far to push the technology. If it oversteps the mark? Devs can adjust the settings and rein it back in, and you can turn it off entirely (separate from regular DLSS upscaling) if you disagree with their choices.

On that note, some Nvidia staff members I spoke to took issue with the idea that DLSS 5, even in the experimental version shown earlier this year, is capable of transforming the facial geometry of a character model. I argued the contrary, because well, look at the screenshots below.

Apparently, this is more of a perceived effect created by the lighting than an actual, DLSS 5-induced facelift. This eventually turned into a philosophical debate, where I argued that, even if it simply looks like the model is creating markedly different facial geometry from the original character models, the end result is functionally the same.

Because that's what it looks like, and what it looks like is the whole point. This argument resulted in a stalemate, so I'll let you be the judge for yourselves.

Anyway, the longer the demos continued, the clearer it became that DLSS 5 is capable of creating some stunning imagery when handled with a delicate touch, and some seriously overblown results when pushed to the limit. DLSS 5 can enhance—but it can also alter an image with all the subtlety of a sledgehammer.

The example above shows a DLSS 5-enhanced scene of some foliage against a building, with ray tracing turned on and off. It's supposed to demonstrate that the more lighting information you give the models to work with, the better the visual result. The DLSS 5 and ray-traced image combined, though, looks seriously oversaturated to my eyes. Realism has gone out of the window, and the end result is… lurid, to say the least.

What is beautiful to some is garish to others, but all I'll say is that the term "better" is far from my mind. There's no accounting for taste, is there?

What we're going to be relying on for the most part, though, are the tastes of the game developers themselves. I was taken into a separate room to try out NBA 2K27, the first game you'll be able to play with DLSS 5 enabled. Officially, at least, as it can apparently be modded into any game.

This was a very locked down session, in which the devs refused to be drawn in on the details of exactly what sort of DLSS 5 implementation I was looking at. However, I'd guess that, by Nvidia's standards, everything I saw was achieved by using the lower end of the DLSS 5 spectrum.

The sweat running down the players' arms looked… sweatier. More glistening. The facial features looked more realistic, with better lighting in the nooks and crannies of their various mugs, and less of the "uncanny valley" effect I'd seen at higher settings. The crowd in the background transformed from low-poly models into a more passable depiction of basketball match attendees, without straying too far from the look of the original assets.

To my surprise, I didn't experience the AI image ick. It looked more like the difference between a game running on a console, and one running on PC with a ton of extra graphics options added on top. DLSS 5 doesn't look quite so revolutionary at these settings, perhaps—but the results impressed me more than I expected.

I was careful to look for signs of ghosting, or any morphing AI faces under fast motion. One of my critiques of the previous DLSS 5 footage was the temporal handling, which looked a little blurred and smeared to my eyes. However, there was no hint of that here. I played virtual basketball, I sucked at it, and I enjoyed the look of the game overall.

In all honesty, it felt like playing a next-gen game with cutting-edge graphics technology, and not an ill-advised AI gaming revolution. This, beyond anything else, was perhaps the best demonstration that DLSS 5 can be used to genuinely enhance the look of our games, rather than turning them into an AI-slathered mess. If developers tune it wisely, there are real visual benefits to be seen.

All of this might be a moot point, though, if performance is a concern. The Nvidia devs I spoke to estimated a 50-60% performance hit with DLSS 5 turned on, which doesn't vary on a game-by-game basis. Essentially, it's still a massive resource hog as things stand, although it no longer requires two RTX 5090s just to run. Goodie.

Further performance improvements are forthcoming, says Nvidia, but my eyebrows jumped so high at this news I'm surprised they didn't shoot off the top of my head. Graphics card prices are outrageous right now, and I can't help wondering who's going to want to slice their frame rate in half in return for some AI-enhanced imagery, good-looking or otherwise.

All of the performance charts I was shown used a combination of 6x Multi Frame Generation and various levels of upscaling to bump up the numbers. And while I'm not as much of a frame generation sceptic as some, I doubt this will come as good news to those curious to try out the tech.

If the performance impact was minimal, I could see some gamers experimenting with DLSS 5 for themselves and potentially changing their minds. With this sort of massive dip, though, I think DLSS 5's GPU-hogging ways are a much bigger issue for adoption than the AI-ified nature of the tech itself.

What's left, then, is the thorny issue of the ethical debate. The elephant in the room, and there's no getting around it, is that generative AI usage in games is a massive hot topic at the moment. In fact, AI usage in general is causing all sorts of debates in all sorts of arenas, from the potential effects on the job market, to ecological concerns, to discussions around the very nature of truth.

And this, I feel, is the real hurdle Nvidia is trying to jump. While it's clear that DLSS 5 can be beneficial to the image quality of our games in the right hands, whether we, as gamers, want to participate in the AI-ification of everything in this way is much more doubtful. And it's on this level where I feel far more conflicted about the tech.

My criticisms of the AI revolution, in a general sense, are well documented. And while I can absolutely see the potential benefits of DLSS 5, I left the demo booth with mixed feelings. Is this where the games industry is heading? Will I be spending most of my gaming time looking at more AI-enhanced imagery in the future?

Will developers make use of the tech wisely, or will major releases eventually go for the full yassified approach, transforming our games into something unrecognisable from what they are today? Will devs jump in with both feet, or will they be more cautious? And crucially, is AI-enhanced imagery really what we want, even if it's capable of looking nice?

In principle, the demos I was shown should have me jumping up and down with excitement at the potential of better gaming graphics. But the AI association, perhaps even infiltration, into a form of media I truly adore is enough to keep me awake at night. In practice, DLSS 5 is certainly capable of creating some good-looking imagery in the right hands.

But it also represents a major leap towards an AI-enhanced gaming future, and one that still makes me, in the depths of my bones, more than a little uncomfortable. Whatever the case, DLSS 5 will be launching on September 3—and we're going to have to decide for ourselves if we want to engage with it.

We know that DLSS 5 is due to arrive with us in the autumn time, but that release date feels a lot closer, with specific files being picked out of an early access version of NBA2K27 and dropped into Remedy's Control (via Notebookcheck). The files in question have been found in the 2K ball-tossing title's PC version—which is currently playable for gamers who purchased the Deluxe or Ultra editions—and a modder from the RenoDX team has managed to inject it into Control with sliders.

And, while I'm likely to be pilloried for saying so, I'm kinda impressed with the changes and the level of… ahem… control you have over the visuals with the number of sliders at play in the mod. It's also impressive that this has been knocked together using an early DLL file from a completely different game and dropped into Control.

The mod has been released in the RenoDX Discord channel and offers up seven different styles, which seem to range from conservative to full-on yasification. So far it's just aimed at the character models, but given the suggested performance hit from enabling it, you are going to want to limit its reach.

Looking into the Reddit chats about it, people are suggesting drops on an RTX 5070 Ti that take Control from 71 fps to 35 fps. That's a big hit, though we are talking about a pre-release feature in a game that it's not (yet?) designed for. When ray tracing still gets a lukewarm reception due to its performance impact, yet another resource-hogging feature will likely struggle for traction even outside the AI antipathy in gaming.

And that's without taking into account the current ludicrous pricing situation for every Nvidia GPU from the fifth tier of its RTX 50-series and up.

There are various images of the lead character, Jesse Faden, but I think one of the most interesting is linked in the original X post, where the conservative preset is seemingly being used. The character model doesn't look markedly different, but as the sliders are adjusted, you can see how the filter is layered on without negatively impacting the original design or geometry.

Then there's another uploaded image where you can slide between the DLSS 5 on and off versions of the character model, where it's far more extreme. But still, I don't think it looks bad. If you'd given me that Faden model from the outset natively, I'd have marvelled at Remedy making such a detailed character, especially given the company's obvious desire for photorealism in its characters. It still obviously retains the geometry, but this is far more of the AI beauty filter fear that large parts of the gaming community are concerned about with this feature's release.

I think the fact this has all been pulled together from one DLL file taken out of an early access game is impressive, and makes me think that the instant DLSS 5 gets released, we're going to see a whole lot of these mods out there giving gamers access to all the controls for a whole bunch of games… whether those game developers like it or not.

And that's going to be a whole other debate.

One final thing to note is that the DLL file in question is noted as DLSSNR in the description, which suggests that DLSS 5 as the catch-all name for this AI game filter is going to change. Rebranding it as DLSS Neural Rendering feels far more suited to this feature, and might have led to less of a backlash at the start, where calling it DLSS 5 from the outset made it feel like this is just what the next version of DLSS is in its entirety.

Apex Legends seems to be having trouble working with the AMD Radeon RX 9070 XT right now. As relayed by the Apex Legends account ApexCommunityJP on X (machine translated), "We have confirmed an issue causing crashes for players using the AMD Radeon RX 9070 XT GPU, and we are currently investigating the cause."

I can confirm on my end that Apex Legends crashes when I try to launch it using a rig with an RX 9070 XT. I get the following game/engine 'fatal error': "The application's device failed due to badly formed commands sent by the application. This is a design-time issue that should be investigated and fixed."

The aforementioned X account does recommend trying to roll back your AMD graphics driver; however, some replies to this claim that doing so doesn't resolve the problem.

One suggestion in the replies that does seem to have worked for me as a temporary workaround is to add the following to the game's Steam launch options: "+staticprop_color_use_bindless 0 +staticprop_depth_use_bindless 0 +r_drawworld_depth_bindless 0". To do so, on the game's Steam page, click the cog, then Properties, then paste those commands into the launch options.

These seem to be options related to bindless rendering, which is where the GPU accesses shaders without mediation from the CPU. It looks like these commands disable that newer tech, which could mean a bit of a performance hit. I can confirm that adding these commands to my game's launch options did allow me to boot up and have a quick game, so it looks like it works.

Of course, it's only a workaround, and we'll have to hope AMD or Respawn come up with a proper fix for something a little cleaner and more permanent. It's unclear at this time what is causing the crashes and what can or will be done about it, so we'll just have to keep our eyes peeled for updates.

Nvidia's new DLSS 4.5 Ray Reconstruction feature is laser-focused on trying to fix the second most annoying thing about ray traced or path traced games: visual artifacting. Yes, the lighting effect does deliver the most accurate-looking lighting and shadows in games, but in motion there is a definite visual weirdness added into scenes. And that weirdness generally comes from the denoising pass designed to turn the low-fidelity path traced image, with its limited number of rays, into a fully realised frame.

Leave a game's viewport still and the scene will pretty rapidly come together to make a sharp, convincing image. Shift that around, or have that scene filled with moving objects or characters and you're going to see the sort of visual trails you might only have been used to in the warehouse parties of your early twenties. Just me?

The issue is that it's way too expensive in terms of raw compute to trace every single photon/pixel that would otherwise make up a fully traced scene. Even with the current method of tracking a far smaller representative sample of rays bouncing around the geometry of your game world it's still incredibly demanding of your PC's hardware.

This is, by the way, the most annoying thing about ray traced or path traced games: the amount of graphical horsepower the tech demands over pre-baked lighting techniques. The huge performance hit it involves is punitive for all but the most powerful GPUs, and we are still no closer to solving that right now.

So we still have to use small numbers of tracked rays to create an approximate scene which looks something like this:

But that doesn't look great if you're trying to make a convincing world, though I guess there's something to that aesthetic in a sort of 3D ASCII kind of way. Anyway, to realise this image into a production ready frame you need to use a denoiser to fill in the blanks between the sampled rays, and the hand-tuned denoisers track and group pixels across multiple frames, to give them access to more samples. They then use these neighbouring pixels to interpolate and blend the scene together.

As I said, that's fine for a static image, but those pixels don't stay representative for long and in motion they break down, adding in artifacting to a scene and smearing detail away from textures. This is where the original Ray Reconstruction feature came in, replacing the hand-tuned denoisers with an AI-based denoiser working alongside Nvidia's upscaling models, to generate more accurate pixels in between those sampled rays.

And now we're finally getting an updated version of Ray Reconstruction which uses the sharper second-gen Transformer model initially brought into DLSS 4.5 Super Resolution back in January.

And it's about time, because for ray or path traced games, even if you were using the improved transformer model of DLSS 4.5 Super Resolution, Ray Reconstruction was still using the old model, leaving you those tell-tale artifacts.

Nvidia is touting four main improvements with DLSS 4.5 RR:

• Efficient Denoiser: The new model delivers 35% more compute capability, and processes 20% more parameters, while maintaining similar performance to the previous model.
• Enhanced Super Resolution: Building upon the advances from DLSS 4.5 Super Resolution, the model has deeper spatial awareness across every scene, and more intelligently uses game engine pixel sampling and motion data. The result is improved lighting accuracy, better temporal stability and clearer motion in ray-traced and path-traced content.
• Expanded Training Dataset: Trained on a larger dataset, the new model is even better at image reconstruction. This intelligence gives the model even better awareness to pick the most accurate engine data to reconstruct scenes closer to ground truth.
• Finer Developer Control: The new model provides developers with finer control for temporal accumulation, providing precise tuning of model response for even better image quality.

The new version of the feature is releasing today, and jamming the second-gen transformer model into supported games is done via the Nvidia App rather than from inside your games' settings screens. You just need to select a title from within the Nvidia App itself, navigate to the 'DLSS Override - Model Presets' entry and select 'Preset F' from the 'Ray Reconstruction' dropdown menu. Apply it and you should be set.

While complete image fidelity across all ray or path traced games is by no means a 100% solved problem, it's clear that DLSS 4.5 Ray Reconstruction is able to deliver a more stable image when it's in use. And it's not just about giving you a more stable image with less 'boiling' textures and trails, either, it's adding back in more detail to a scene that would have otherwise been lost.

You can see that especially in the below video for Resident Evil Requiem's opening scene. The textures on the white car at the beginning are still shimmering behind the rain effects using the older model, but are almost entirely stable with DLSS 4.5 RR enabled. But it's the extra detail on the ground that highlights what the new second-gen transformer model is capable of here.

DLSS 4.5 Ray Reconstruction on left, DLSS Ray Reconstruction on the right

Of course we're going to highlight what it looks like in Nvidia's poster child for all things DLSS: Cyberpunk 2077. In the two videos below you can see where that extra temporal stability really makes a difference. In this first video, inside a store, you can see the striping on the ground as the viewport shifts, which is completely absent from the improved DLSS 4.5 RR version.

You might have to watch it loop through a couple of times to see what I mean, but if you focus on the older version on the right you will see, when the view moves to the floor, that there is an after image of essentially the edge of the screen imprinted on the floor. This fades quickly, but is noticeable.

DLSS 4.5 Ray Reconstruction on left, DLSS Ray Reconstruction on the right

Outside, looking up at the ceiling of this underpass you can see where the textures resolve far, far quicker than before. The different areas of the ceiling using the previous version of RR almost seem like they can't make up their mind what they should look like, while DLSS 4.5 RR comes in with far more certainty.

DLSS 4.5 Ray Reconstruction on left, DLSS Ray Reconstruction on the right

My final example comes from Alan Wake 2 and its own heavy use of path tracing and the denoising steps that are vital to the game's fidelity. To be honest, the strange artifcating of Alan Wake 2 almost gets a pass from me, because the dreamlike trails around people kinda add to the weirdness of the game. But it's not meant to look like that.

You can see from the dancing man in the Elderwood Palace Lodge bar that those trails are almost entirely gone from the latest version of Ray Reconstruction. But, notably, they are not completely resolved. There is still some level of artifacting around his feet as he shuffles around the wooden floor.

Though it is worth noting here you can also see the enhanced level of detail the second-gen transformer is adding back in. If you look to the wooden floor in the room behind the dancing dude you will see the planks are more clearly represented.

DLSS 4.5 Ray Reconstruction on left, DLSS Ray Reconstruction on the right

So yes, there is still some work to be done before we can say that path tracing visuals have been perfected, but this is still a worthy upgrade. And honestly, we probably need a mixture of both improved AI modelling and more powerful hardware to be able to add more traced light bounces into a frame. But in these tech-starved times of ours, hoping for more horsepower when the datacentres want to jealously hoard all the compute for themselves is a fool's game.

But DLSS 4.5 Ray Reconstruction is a definite improvement over Nvidia's previous versions. Sure, a lot of the improvements are only really telling when you're specifically looking for them, but equally some of that smearing and the 'boiling' textures are absolutely things I can do without.

Ray Reconstruction also remains RTX agnostic in that it will function across all generations of GeForce RTX GPUs, and isn't really any more draining upon your PC's resources than the previous version. Though, as a whole, it remains true the transformer-based DLSS features weigh more heavily upon your system's performance on older generations of RTX hardware than the RTX 50-series.

But I would say that's less of an issue for Ray Reconstruction than Super Resolution. One is for helping resolve high-end path tracing where a massive performance hit is expected, and the other is specifically designed to get you more performance. You're only really going to want to be doing heavy path traced gaming on high-end cards of the Ada and RTX Blackwell generations, whereas Turing and Ampere GPUs could do with all the help they can get.

A couple weeks back I noted that I'd been tracking GPU prices on an almost weekly basis for a year now, and that things have never been worse. Well, PC Partner—one of the biggest manufacturers of graphics cards you've probably never heard of—suggests it's going to get even worse. And probably quickly.

PC Partner makes a ton of graphics cards for its consumer facing brands Zotac, Inno3D, and Manli. I've visited one of its main factories in Dongguan, China, and it's production lines are big. So when Singaporean outlet, The Straits Times, notes the company as saying that the supply of graphics cards is going to get real tight in the coming months it's worth listening.

And the worst part about it is that apparently it's going to be the entry level cards that will be the most adversely affected. With severe shortages comes increased prices, and while Nvidia cards with VRAM capacities of 8 GB haven't had the spiking price tags of their more heavily memory'd cousins, it looks like it won't be long before they do.

Benchlife, reporting on the PC Partner announcement, has itself confirmed that supply of Nvidia's RTX Blackwell GPUs in the third quarter of this year is below the second quarter, and that supply at the end of the year will be even lower than that. It also notes that this isn't purely down to the constrained supply of GPUs and memory owing to the demands of the data centre elites, but that other parts of the graphics card manufacturing process—such as PCBs and power componentry—are also suffering from long lead times compared with last year.

Given that Nvidia RTX 5060 Ti 16 GB and RTX 5070 graphics cards are both now batting around the $800 mark, it's mighty concerning that the more affordable end of the GPU spectrum look set to get slammed with higher prices, too. The doom disciple in me fears for a world where an RTX 5050 costs $500.

Though it can only cost $500 if someone is willing to pay that and surely, surely no-one would. Would they?

But it's not just going to be Nvidia cards though. It has been GeForce GPUs which have been affected more by the most recent price spike (though RX 9000 series cards have also had their fair share of price increases immediately after launch), but PC Partner is also the contract manufacturer of Sapphire GPUs, one of AMD's biggest graphics card makers, so I don't doubt it was talking to those cards, too. Especially if part of the supply constraints are around other components all GPUs need to live.

And if AMD cards get priced out of the reach of us mere PC gaming mortals, then there will be no alternatives for us.

We've been tracking graphics card prices for the past twelve months and after yet another pricing surge last week, there's been a wave of depression sweeping the PC Gamer offices. It's Nvidia GPUs that have been hit the hardest, with the ludicrous situation where the cheapest RTX 5060 Ti 16 GB we can find is more expensive than the cheapest RTX 5070.

What's even more ludicrous—and surely entirely unsustainable—is that both those cards are both close to $800. These are the fourth and fifth tiers of Nvidia's RTX 50-series lineup and that pricing is utterly punitive for PC gamers.

So, that had me wondering what this might mean for the next generation of cards, and how much you dear people, our PC Gamer readers would accept paying for your next graphics card. So we've been asking: How much would you be willing to spend on a new GPU? And honestly, the answers have surprised me; you're willing to spend way more than I could!

How much would you be willing to spend on a new graphics card?

Given the recent price hikes, here's what the PC Gamer audience be okay spending on a new card

Votes (%)

$600 - $800

22

$400 - $500

18

$800 - $1000

15

$300 - $400

10

$1500+

10

$1000 - $1500

9

Up to $200

8

$200 - $300

7

Votes (%) Data

Product	Value
$600 - $800	22
$400 - $500	18
$800 - $1000	15
$300 - $400	10
$1500+	10
$1000 - $1500	9
Up to $200	8
$200 - $300	7

The recent pricing surge is all down to the AI-backed memory crisis, and these 16 and 12 GB cards respectively are seeing their pricing rocket as a result of the healthy dollop of GDDR7 memory attached to the GPUs. But it's not just production constraints pushing up the price of cards going into retail, it's also going to partly be down to AI enthusiasts looking for Nvidia GPUs with high VRAM capacities to run their local AI systems.

And yes, once more PC gamers are suffering at the hands of folk using both the technology and our graphics cards for things other than gaming and massively spiking the prices as a result.

But I'm surprised that a hefty 56% of readers who responded to our poll would be willing to spend $600 or more on their next graphics card, though I think the number that got me thinking most was just how many of you would actually drop $1000 or more on a new GPU. 19% is a not statistically insignificant percentage, and that is going to make Jen-Hsun's face light up as he considers pricing the new RTX 6070 Ti at $1,000, despite looking at dropping the VRAM capacity to 4 GB of GDDR5 because of the ongoing memory crisis. DLSS 5 will take up the slack, right?

Statistically speaking, it's also interesting to see the parity between the 10% of readers looking at the $300 - $400 segment and the other 10% waaaaay down the other end of the spectrum happy to spend $1500 or more.

But the $600 - $800 segment is the most popular among our readers, with 22% of the vote, and $400 - $500 just behind that with 18%. That's probably where I'd find myself, to be honest.

Given the state of the market, maybe you guys are just being realists in terms of what new GPUs will cost. But it's still kinda surprising to see so few people around the budget end of the market; that's something I've always seen as a strength of PC gaming, but it's also been an underserved colour in the GPU spectrum of recent times.

Let us know where you fit in the spectrum in the comments below!

This week we want to know how you feel about monitors...

Prices for graphics cards are still on the rise. Just last week, we noted prices were at the worst we've seen them in the past 12 months, but after further checks this morning, they're looking worse again still.

Among the worst offenders, the RTX 5060 Ti 16 GB. With memory being the main driving factor of price hikes in the past 12 months, it's perhaps no surprise that this memory-laden card is rising faster than most. Yet prices are beyond the pale.

In my search for a brand new card from a reputable retailer, this was the cheapest I could find: an MSI Shadow RTX 5060 Ti 16 GB for $780. Not only is that more expensive than the best value RTX 5070 I could find at $763, but it's far in excess of AMD's RX 9070 XT. That's just $720. It's also in excess of the RTX 5060 Ti 8 GB, which I've found at $430. A $350 premium for the 16 GB model.

There's little reason to purchase an RTX 5060 Ti for gaming at that price. It wasn't even a card I liked much at MSRP. I like lots of VRAM as much as the next guy, and it makes a difference to the overall performance, but there's a limit to how much I'd be willing to spend to get it.