Phantom Blade Zero, the highly anticipated wuxia action RPG, is finally out in the world, and if you're planning on playing it on an Nvidia graphics card, maybe consider skipping the latest Nvidia driver update. Yes, that's the opposite of developer S-Game's pre-launch advice.

Today, the Phantom Blade Zero devs took to the Steam page with a post titled "Reason and Madness" recounting the history of the dev team, previous works, and how Phantom Blade Zero came to be. This is followed up with three announcements: a recommendation to download the 5 GB day one patch, encouragement to try the game on handhelds and lower-end/mid-range PCs, and, finally, advice not to use the latest Nvidia driver, 621.01.

It says, "Based on our testing, version 617.42 currently offers the best experience in Phantom Blade Zero. If you’ve installed a driver newer than 617.42, we recommend rolling back to 617.42 and choosing the clean installation option. Please stay on this version until Nvidia releases its next driver update, which we expect to arrive soon."

Game Ready Driver 621.01 launched on Wednesday, October 21, and, in addition to providing some bug fixes, it is intended to better support DLSS and RTX in games like Phantom Blade Zero, Call of Duty: Modern Warfare 4, and NBA 2K27.

The Phantom Blade Zero team does not elaborate on how the previous drivers are better than the current one, but the recommendation implies it's significant enough to negatively impact your experience.

Luckily, it's very easy to roll back Nvidia drivers in the Nvidia App. Simply boot it up, click on 'Drivers', then scroll down to whatever previous driver you want to roll back to, click the three dots on the side, and then click to install.

If you don't want to install the Nvidia app, you can roll back drivers by opening up the Device Manager, clicking on "Display Adapters", then right-clicking on your GPU, hitting "Properties", navigating to the "Driver" tab and then rolling back drivers there.

We don't yet have a timeline on when Nvidia's next driver is due to be released, but S-Game seems to think it will be soon, so you can always hold out for that if you won't be playing the game in the near future.

Galax has come out in defence of its continued existence for the second time this year, following a report that incorrectly claimed the graphics card brand was closing down its factory.

In a report by El Chapuzas Informatico, citing industry sources, Palit was claimed to be closing the Galax factory. This is said to be occurring at the same time as a major reorganisation of the Galax and KFA2 brands.

Galax, a subsidiary brand of Palit, has expressly denied this claim. In, the company says its business operations are "continuing as normal".

"Galax continues to operate as usual," a statement under both Palit and Galax letterheads, published on the Galax website, says. "No manufacturing facilities have been shut down, and Galax products remain available through our sales channels in both domestic and international markets.

That bit about internal reorganising of the Galax and KFA2 brands has some truth to it. We know that from Palit's previous statement on a rumoured Galax closure. Yep, Galax has been rumoured for chop before, and not too long ago. It last happened in April, 2026.

Back then, it was speculated that Galax was shuttering and Palit was taking over. Palit said no, Galax is still operating and we've already taken over (Palit has owned Galax since before it was officially called Galax, buying the company then named Galaxy in 2008); however, Palit did note at the time that it was integrating management of the brand, alongside KFA2 and HOF, under Palit's HQ.

So, the reports of Galax's death are greatly exaggerated, but the ones about putting it under new centralised management are closer to the mark. Importantly, the brand's cards are still on the shelves, including the brand-new Galax RTX 5080 HOF Extreme encrusted with Swarovski crystals.

Nvidia's RTX 5090 has been a subject of much speculation recently, as it was rumoured to be discontinued last month when it vanished from the company's official marketplace, only for Nvidia to confirm it wasn't discontinued. The plot has thickened, though, as AIB partners—the companies actually designing and manufacturing RTX 50 series graphics cards—suggest even they can't get their hands on the parts required to make the card.

This is according to an Nvidia AIB partner who spoke to YouTuber Der8auer. They say, "No, Nvidia did not inform us that RTX 5090 is EOL. But currently we cannot purchase GB202 [the GPU] together with 32 GB GDDR7 as [a] complete set, and we also don't know if ordering will be possible again in the future."

The unnamed board partner clarifies that in previous shortages, they could place backorders with Nvidia, but this is "currently not possible anymore."

This is all to say that this AIB partner is unable to get the GPU and memory required to make the RTX 5090, and doesn't know when that supply issue will be fixed. Though this isn't the card being discontinued, for many partners, not getting the GPU and VRAM is as good as doing so.

Whether it's the GPU or the memory that's causing the biggest headache isn't specified, though in the current context, one can assume it's likely both to some degree.

As of right now, companies are reliant on GPU and VRAM bundles from Nvidia before they can make their own cards. There have previously been rumours that Nvidia would stop bundling its GPUs and VRAM together, but even if it did that, this would only partly solve the issue of the memory crisis for Nvidia. Those who might have been able to buy just the GPU would still be subject to a price squeeze on the VRAM when trying to source it direct from one of the major memory manufacturers.

Der8auer also spoke to an RTX 5090 cooler manufacturer who said, "Two of our partners already reduced quite much the order forecast for RTX 5090 based heatsinks in Q1 2027, when compare[d] with Q3 2026."

"For Q2 2027, until now we still did not receive any order for RTX 5090 model heatsinks from our partners."

Nvidia still remains steadfast in saying the card has not been discontinued, though it acknowledges that limited supply chain capacity has led to a "somewhat constrained" output. Its German branch says, "Nvidia will continue to ship it, and this is very important."

The RTX 5090 (an already expensive card at its previous MSRP of $2,000) has been prohibitively expensive for months. It's so pricey, in fact, that we stopped tracking it in our graphics card price watch. It has been incredibly difficult to find any of the cards available from reputable sellers at even reasonable prices (with some listed for upwards of $8,000 right now) and therefore we could never recommend actually picking one up. Though it may not be officially discontinued, it seems like the likelihood of picking one up soon at a decent price is getting slimmer and slimmer by the day.

Consumer RTX 5090 stock in the US (and many other countries) seems to be virtually non-existent at the moment, despite Nvidia's claims that production is still ongoing. So, what's happening to all those brand new consumer cards?

A leaked video reportedly taken from inside a Chinese factory may have the answer. Posted by HKEPC on X, the footage shows what appears to be a huge number of consumer RTX 5090 cards being dismantled en masse for AI server conversion (via Tom's Hardware).

Workers are shown dismantling the GPUs, removing their three-fan coolers and heatsinks, and stacking the PCBs for conversion to dual-slot cooler configurations suitable for AI server racks. The scale of the operation looks truly vast, with what looks to be hundreds, possibly thousands, of GPUs getting the chop shop treatment.

According to HKEPC, the GPU and VRAM are desoldered and transplanted to a separate PCB ready for cooler transplant. Once converted into an AI server-suitable form factor, the GPUs are then tested on vast arrays of benchmarking rigs, ready to begin their new lives as AI workhorses rather than consumer graphics crunchers.

The horror, the horror, etc. These "grey market" facilities are said to be scattered throughout the country, despite the RTX 5090 not officially being available for sale in China. Nevertheless, this facility looks to be stuffed full of the highly prized GPUs, with skilled technicians pulling them apart for repurposing.

Reports of this sort of activity are nothing new, but this footage showing the scale of a single operation is still unsettling. If facilities like this are hoovering up GPUs and converting them at this sort of rate, it's perhaps no surprise that RTX 5090 consumer cards have become so difficult to find elsewhere.

We've stopped tracking the GPU in recent months on our GPU price watch page, as the only US listings we can find are from third-party sellers for vastly inflated prices. A gaming GPU for $8,000+ simply makes no sense at all, even if it happens to be Nvidia's top-level card for this generation.

And to see them being converted for, shock horror, AI crunching instead of good old gaming purposes? Well, chalk up another way AI server demand is making 2026 a year to forget for consumer PC hardware.

Whether you dismiss it as 'fake frames' or embrace the tech as a real game changer, frame generation has fast become a headlining act of this GPU generation. We've since discovered that not all frame gen (FG) tech is created equal, but AMD has promised to address a key criticism of its own FSR FG.

Earlier this week, AMD's Jack Huynh announced that FSR 4.1 would be shipping early to RX 6000 graphics cards, though that wasn't all. Huynh also announced that improvements to FSR's frame pacing were on the way, too.

Personally, I'm more familiar with Nvidia's approach to FG—especially after unlocking Multi Frame Gen on my RTX 4070 Super earlier this year. AMD's FSR has often been criticised for how it struggles with frame pacing in comparison. Ie, you may have lots of frames , but not every one of them is displayed for a consistent amount of time, making for gameplay that's a different flavour of choppy.

The issue can affect frames both traditionally rendered and generated, and it's been a stumbling block since at least FSR 3. For example, CD Projekt noted gamers may experience micro stutter back when Cyberpunk 2077 got official support for the tech in 2024. As recently as August this year, Hardware Unboxed's testing demonstrated that pacing is still an issue in games that support both FSR 3.1 and FSR 4 (via Videocardz).

It's good to hear that AMD is focusing on this issue, but without a solid timeline or a list of hardware set to enjoy these frame pacing improvements, this week's announcement amounts to little more than, 'We know, and we're working on it!'

Also conspicuously absent was any mention of Multi Frame Generation on AMD GPUs. Currently, AMD's Frame Generation generates one intermediary frame, though third-party developers have managed to crank that far higher. We already know MFG is on the way for AMD hardware, the question is more 'how soon?'

Here's hoping AMD's officially supported frame pacing improvements for FSR arrive before an enterprising third-party can show them how it's done.

According to hardware outlet ChannelGate, the wholesale spot prices of several RTX 50-series graphics cards are on the rise once again in China, which is an ominous indicator for consumer pricing worldwide.

The 16 GB RTX 5080 is said to have increased by around 2000 CNY, which equates to around $300, while the RTX 5070 Ti has seen an increase of around 600 CNY ($90). In the RTX 5080's case, that puts the current wholesale spot price at around 14,500-15,000 yuan, which equates to a truly stunning $2,171-$2,246 range when converted to US dollars.

Retail graphics card pricing has already gone bananas in recent months, with US prices surging to hitherto unthinkable levels. The cheapest RTX 5080 we can currently find is a wallet-aching $1,850, while a mere $70 separates the RTX 5050, RTX 5060, and RTX 5060 Ti 8 GB on our GPU deals page.

Yes, that means the cheapest RTX 5050 we can currently find is $400. That's a silly price for Nvidia's slowest gaming GPU of the current generation.

While the 16 GB RTX 5080 has seen the largest jump in price so far (exacerbated by its reported discontinuation in favour of a 24 GB version), ChannelGate also says that Asus has just updated its pricing policy across the stack.

This amounts to a 1000 yuan ($150) increase for the RTX 5080 16 GB, a 200 yuan ($30) increase for the RTX 5070 Ti, RTX 5070, and RTX 5060 Ti 16 GB, and a 100-300 yuan increase for all AMD cards.

While some of those increases may seem small, they're an indication that we might be looking at an even grimmer winter for anyone who wants to upgrade their graphics card for reasonable cash. And without a major change, likely a grim 2027, too.

But hey, ChannelGate does give us some info on the expected wholesale price of the mooted RTX 5080 24 GB. The outlet says it could potentially reach 19,999 yuan. That's, err, just under $3,000 at the current exchange rate.

Well, this is all very depressing news. The global memory crisis continues to bite, working alongside GPU production shortages to create a graphics card market that looks increasingly ridiculous as the months go on. At least gaming monitors are still pretty cheap? Silver linings, and all that.

I reported on a missing shipment of RTX 5090s earlier this week, with two pallets worth of the highly-prized graphics cards presumably stolen from a Brazilian airport. Now new details of what looks to be a heist have emerged, and one of the GPUs has reportedly ended up on a second-hand marketplace.

GMC Online reports that Brazilian news channel Jornal da Band spotted the graphics card for sale on OLX, an online marketplace for used wares. The serial number of the card is said to match one of the missing units, with the price set at R$32,000, equivalent to around $6,400 (via Videocardz).

The broadcaster has revealed new information on the missing cards, said to be obtained from a Brazilian police report. According to the outlet, two outsourced forklift operators removed the pallets from an airport warehouse on September 21, while a third employee allegedly helped to move them through electronic security controls.

The shipment of GPUs belongs to Brazilian electronics distributer Oderço. The CEO of the Oderço Group, Oderço Matos, has previously spoken on the company's Instagram account about the missing GPUs, calling for more clarification from the authorities as to what exactly happened to the company's cargo.

GMC Online reports that Guarulhos International Airport has responded to the outlet, confirming that it had notified the authorities of the incident and is now in contact with the importer's representative.

Representatives from the airport say the investigation is now the responsibility of the appropriate authorities, and that it has provided guidance for initiating the compensation process.

The Oderço Group CEO has stated that the missing GPUs were worth R$2.5 million ($495,000) at cost, so we're not talking about pocket change here. RTX 5090s have become extremely difficult to find in recent months, although Nvidia GeForce chief Jeff Fisher has insisted that the ultra high-end consumer card is still in production.

Regardless, we've stopped tracking the card on our graphics card price watch page, as the market is full of ludicrously priced listings from third party sellers. Even a single card is worth a huge amount of cash these days—but by the looks of things, shifting two pallets worth of ill-gotten GPUs without someone picking up on it may prove to be an impossible task.

If you care about retro PC gaming hardware, there's a good chance you've heard of the now-defunct 3dfx. And if you know 3dfx, there's an even better chance you've heard of the 3dfx Voodoo graphics card. Well, according to one of its founders, its success is owed at least in part to the legendary shooter Quake.

Ross Smith recently spoke to our friends at Club386 in celebration of the first 3dfx Voodoo turning 30 years old, and he talks about the 'Quake effect'.

Effectively, though Quake launched in June 1996, it didn't get GLQuake, which allowed for 3D hardware acceleration, until the start of 1997. John Carmack reportedly reached out to 3dfx, which then worked on an API called MiniGL that could run the game. In its implementation, 3dfx cards could then run the game at a better resolution in greater detail with a wider range of colours.

Smith says, "I would certainly give id, John, and Quake a great deal of credit for our overall success—we probably sold millions of boards because of that game, and it was magnificent. That’s the last game I’ve gone all the way to the end on, and I was exhausted after doing that. It was well crafted, beautiful, dark."

It makes sense why developers would want to work with the likes of 3dfx because better compatibility makes their games look better. And a developer is more likely to sell a game if it can wow an audience.

This symbiotic relationship was key to 3dfx development at the start. 3dfx reportedly reached out to developers to show games pre-rendered on 3dfx Voodoo cards, and "they loved it".

Smith argues this relationship to developers was quite unique at the time. "Most other companies, like S3 or others, saw game developers kind of as a necessary evil, not as the lifeblood of a company that’s focused on gaming. So we flipped the whole thing around.”

This is a relationship that might seem familiar nowadays, with the likes of Nvidia, Intel, and AMD working directly with developers to improve software. Nvidia and 3dfx famously took each other to court for various patent infringement complaints in the late '90s and early '00s, and Nvidia acquired many of 3dfx's assets after it filed for bankruptcy in the early 2000s. So in a way, we have 3dfx to thank for parts of Nvidia's successes in the following years, too.

And 3dfx was unique in a few ways. The Voodoo cards had great 3D acceleration because it was designed specifically for 3D acceleration as a co-processor alongside a 2D card, letting the CPU do other background work at the same time.

It was just one of many steps towards hardware that was specifically designed for gaming rigs. "We did something revolutionary there," says Smith, "we didn’t have a VGA. That was partly down to our lack of design experience, and partly down to my concern about profitability and compatibility. Being compatible with a million apps is really hard—being compatible with 200 games is much easier.”

We prattle on about GPU prices so much nowadays that even we're getting sick of it. Yet for as long as the memory crisis rages on and prices keep rising, we'll keep bringing it up. We'll also keep trying to spot the best prices for graphics cards, which is exactly how I spotted these specific pricing horrors.

The RTX 5050, RTX 5060, and RTX 5060 Ti 8 GB are now only separated by a mere $70. That's not between each card—the jump from an RTX 5050 to an RTX 5060 Ti 8 GB is just $70, down from nearly double that at their intended MSRPs.

'What's the big deal, then?' I hear you ask. 'Surely that means less cash for a better card?'

It is a big deal, and no, it doesn't.

This is due to dreadful circumstances: the cheapest RTX 5050 that I could find is this one currently $400 at Best Buy. It's not only far in excess of the MSRP—which should be a mere $249—but it's actually in excess of the MSRP of the RTX 5060 Ti 8 GB at $379.

The cheapest RTX 5060 Ti 8 GB I could find is $470, and the cheapest RTX 5060 is $420.

Moreover, if you look beyond the 8 GB options at the bottom of the stack, to the RTX 5060 Ti 16 GB, that card jumps up to $789. That should be around the $429 mark MSRP-wise—should.

What's even worse is the RTX 5060 Ti 16 GB is currently the same price as the RTX 5070. That card is also $789.

It's pretty terrible anywhere you look. Our own in-house tracking shows prices are currently spiking once again, not the least bit the high-end market.

Lowest price per GPU

We've been measuring the lowest available price of each current-gen GPU since August 14, 2025.

Lowest pricing (Price, $)

Week

AMD Radeon RX 9070 GREAMD RX 9060 XT 16 GBAMD RX 9060 XT 8 GBAMD RX 9070AMD RX 9070 XTIntel Arc B570Intel Arc B580Nvidia RTX 5050Nvidia RTX 5060Nvidia RTX 5060 Ti 16 GBNvidia RTX 5060 Ti 8 GBNvidia RTX 5070Nvidia RTX 5070 TiNvidia RTX 5080

Lowest pricing Data

Timestamp (Week)	AMD Radeon RX 9070 GRE (Price, $)	AMD RX 9060 XT 16 GB (Price, $)	AMD RX 9060 XT 8 GB (Price, $)	AMD RX 9070 (Price, $)	AMD RX 9070 XT (Price, $)	Intel Arc B570 (Price, $)	Intel Arc B580 (Price, $)	Nvidia RTX 5050 (Price, $)	Nvidia RTX 5060 (Price, $)	Nvidia RTX 5060 Ti 16 GB (Price, $)	Nvidia RTX 5060 Ti 8 GB (Price, $)	Nvidia RTX 5070 (Price, $)	Nvidia RTX 5070 Ti (Price, $)	Nvidia RTX 5080 (Price, $)
1	N/A	370	280	600	700	230	260	250	300	430	368	550	780	1100
2	N/A	370	270	600	700	230	260	250	300	430	340	550	750	1099
3	N/A	370	280	550	700	230	250	250	291	430	370	524	750	1000
4	N/A	370	280	550	695	230	250	250	296	430	345	546	750	1000
5	N/A	350	270	600	670	230	250	250	300	430	340	481	749	1000
6	N/A	350	270	600	670	230	250	240	289	379	340	499	749	929
7	N/A	350	270	550	650	230	250	240	299	430	350	520	750	999
8	N/A	350	270	550	650	200	250	250	299	430	342	500	749	980
9	N/A	350	270	560	650	220	250	250	299	430	323	500	750	999
10	N/A	360	270	550	600	210	250	250	299	430	340	480	730	989
11	N/A	340	250	550	600	200	250	230	280	400	320	489	699	990
12	N/A	350	275	530	600	200	250	220	279	400	320	500	730	989
13	N/A	350	280	530	600	200	250	220	279	420	342	527	750	1000
14	N/A	370	280	530	600	202	250	220	279	420	313	540	750	1000
15	N/A	380	288	530	600	202	240	200	279	430	350	499	749	1000
16	N/A	380	288	530	620	202	240	200	270	430	350	499	750	1100
17	N/A	390	310	580	660	200	250	235	260	430	330	549	750	1230
18	N/A	390	310	580	700	200	250	240	290	450	330	540	829	1268
19	N/A	400	310	580	720	200	260	240	299	450	330	549	899	1380
20	N/A	440	310	600	720	200	290	240	299	520	330	620	1000	1350
21	N/A	440	350	590	730	250	290	260	299	520	351	620	950	1365
22	N/A	400	346	630	730	250	300	260	299	520	360	620	950	1365
23	N/A	440	346	630	730	250	300	260	340	520	360	630	950	1290
24	N/A	440	346	620	730	250	300	260	340	520	360	625	950	1290
25	N/A	440	346	610	730	250	300	260	349	550	360	630	950	1250
26	N/A	440	340	610	730	250	300	290	349	550	360	604	950	1260
27	N/A	440	340	630	730	250	300	290	349	570	360	640	950	1290
28	N/A	440	346	620	720	250	290	290	349	515	390	630	900	1250
29	N/A	449	350	620	710	250	290	290	349	500	390	636	990	1290
30	N/A	440	350	620	710	260	290	290	330	500	390	630	990	1290
31	N/A	440	350	620	670	260	290	290	330	500	380	599	990	1290
32	N/A	430	360	620	710	260	290	290	310	515	380	599	990	1290
33	N/A	430	360	600	710	260	290	290	349	490	400	599	980	1290
34	N/A	440	360	620	710	260	290	290	349	550	380	599	980	1270
35	N/A	430	360	600	700	260	304	290	350	550	380	599	900	1290
36	550	440	360	600	700	260	300	290	340	550	370	599	980	1250
37	550	440	350	600	700	250	304	290	340	570	370	599	980	1250
38	550	440	350	600	700	250	304	290	340	570	370	599	980	1250
39	500	400	360	600	700	250	298	290	340	570	330	599	920	1250
40	500	430	360	600	700	260	291	290	340	565	330	599	955	1250
41	500	430	360	600	700	250	290	290	339	560	330	560	900	1250
42	550	450	360	635	690	260	290	290	350	570	380	550	950	1250
43	500	470	370	650	720	260	310	299	359	650	420	755	1030	1290
44	500	470	400	650	720	260	310	299	N/A	780	420	763	1088	1479
45	530	460	360	630	730	260	310	299	359	773	389	800	1150	1550
46	530	470	400	650	740	260	310	299	359	570	389	759	1150	1550
47	520	470	409	650	740	260	310	299	359	570	389	725	1150	1450
48	500	470	430	650	760	260	330	N/A	359	780	389	779	1150	1450
49	570	500	450	667	760	260	330	310	380	780	488	779	1150	1653
50	550	520	400	700	800	290	330	310	439	789	470	779	1150	1654
51	570	520	400	700	830	290	330	400	N/A	789	470	789	1200	1800

If you were in need of a graphics card right now, I'd probably buy an RX 9070 GRE for $570—that's only $20 over MSRP.

That's about as good as it gets.

In the era of speedy solid state drives, I barely remember the last time I saw a loading screen. I say 'barely' because it was, in fact, last night as I waited for Star Wars Zero Company to compile its shaders for a second time following a crash I encountered during the tutorial mission. Some of us are just blessed (read: cursed) like that. Thankfully, more and more games are moving towards pre-compiling shaders.

Case in point, Nvidia just released a fresh GeForce Game Ready Driver that adds support for Advanced Shader Delivery in both Call of Duty: Modern Warfare 4 and Gears of War: E-Day. I've not had a chance to test the Nvidia GeForce 621.01 WHQL driver yet, but journalist Sebastian Castellanos notes "considerably lower system RAM usage" with it installed.

He wrote on X, "In Gears of War: E-Day, [RAM usage is] about 7-8 GB lower in my case. CPU usage is still relatively high though."

To briefly recap, Advanced Shader Delivery is Microsoft tech that compiles a game's shaders before you ever hit 'play', distributing them during the download process and banishing that pesky in-game load screen. This update brings support to all RTX GPUs, from the 20-series to the 50-series, but the tech is only available for certain games.

Back in May, Microsoft worked with AMD to get Advanced Shader Delivery support in a number of games—but only if you download them via the Xbox app. It's the same story here, with Nvidia writing, "When you download or update a supported title, the exact Precompiled Shader Database (PSDB) tailored to your rig is bundled with your download via the Xbox PC app, meaning ASD-supported games purchased from the Microsoft Store will offer a seamless experience."

Nvidia also breaks down why one cannot simply add Advanced Shader Delivery to every game, and why waiting around for a game to compile shaders is even a thing to begin with. "Modern games rely on tens of thousands of complex shaders," Nvidia explains, "However, generic shader files must be compiled locally into instructions specific to your GPU and driver version, which is a CPU-heavy process that forces players to wait up to 30 minutes every time a game launches or updates.

"Pre-compiling every possible combination on the developer’s side would mean users download terabytes of data with each new game and update, leaving local compilation as the long-standing norm." Nvidia goes on, "Microsoft’s new Advanced Shader Delivery solves this by offloading the heavy lifting to the cloud."

If you are rocking an AMD graphics card or AMD laptop/handheld, there's good news. Not only will AMD be adding FSR 4.1 support to APUs and Xbox products in the future, but as of this week, it's also been added to Radeon RX 6000-series graphics cards.

FSR is effectively AMD's answer to Nvidia's DLSS upscaling solution, and the latest generation (which has been supported on RX 7000-series cards since earlier this year) offers performance and quality increases. RX 6000-series support was originally planned for 2027, but this week's announcement also declared that FSR 4.1 will be coming to all Gorgon Halo and Strix Halo chips for mini desktops and mobile workstations, all Gorgon and Strix Halo Ryzen 9 laptop chips, and all Z2 Extreme and Z2A handheld chips.

AMD also announced FSR 4.1 would be coming to Xbox, and will be optimised for Asus ROG Xbox handhelds (though one might assume that based on Z2A and Z2E support).

Jack Huynh, the SVP & GM of Computing & Graphics at AMD, took to X to talk a little bit more about it, announcing that AMD powers "over a billion gaming devices around the world" and that "we're still building".

Huynh says FSR Redstone is now in over 400 games on Radeon 9000-series graphics cards, and that Ray Regeneration (a denoiser like DLSS's Ray Reconstruction) is coming to more games. FSR 4 will be coming to Vulkan, too.

Think of Vulkan like a non-proprietary version of Microsoft's API, DirectX 12. This is what Linux machines use (including the Steam Deck), and a lack of Vulkan support holds back non-Microsoft devices. Greater platform support is great news for Linux AMD gamers.

We unfortunately don't have a timeline on future support, but here's hoping it, too, is ahead of schedule. This means even older rigs should be able to get better performance going forward, and with the next Xbox and PlayStation being presumably AMD-based, it's a great time to lay the groundwork to get more out of AMD's hardware.

Nvidia's RTX 5090 GPU is 'unavailable' from many reputable retailers, approaching an asking price of $9,000 if you happen to track one down on Newegg's marketplace of resellers of dubious virtue. The AI boom is at least in part to blame for this, encouraging team green to shift production resources away from consumer hardware. However, business customers like Amazon are running into logistical issues that just add insult to injury for PC gamers left out in the cold.

Amazon Web Services took delivery of approximately $5 billion of Nvidia GB300 racks in August alone, according to an email seen by Business Insider. Reportedly, the cloud company has also accepted $3 billion worth of hardware each month since. The kicker is that, despite having all of this kit lined up to install in its data centres, AWS does not yet have enough energy to power everything.

This is one reason AWS's free cash flow dropped from $53 billion in Q2 of 2024, to $7.6 billion during Q2 2026. Free cash flow is now in the red according to a leaked recording of an internal staff meeting last month; AWS CFO John Felton joked to the executive leading the company's AI infrastructure, Prasad Kalyanaraman, "You've taken us free cash flow negative as a company, so congratulations."

This friendly joke can only hide so much pain though. Obviously, AWS is no stranger to building data centres, but at this speed and scale, the company is having to juggle multiple different infrastructure timelines amid a whole lot of near-term volatility when it comes to hardware pricing.

Typically, AWS would foot the bill for Nvidia's GPU racks only once they were installed. However, with no clear idea of when that will be, Amazon has had to implement an "Early Rack Receipt" process in the meantime. With no way to power this paid for hardware, that leaves a lot of cash tied up in tech AWS can't actually use—hence the reportedly negative free cash flow.

But AWS is not inclined to stop scooping up GPU racks or pause its AI infrastructure build out any time soon. Free cash flow problems aside, AWS's second quarter revenue grew by 37% year-over-year This was the fastest growth AWS has seen in 18 quarters, making $42.2 billion by the end of July. That's a big win that in turn encourages more big spending.

Indeed, Amazon CEO Andy Jassy recently described AI as "a once-in-a-lifetime opportunity" to investors, with the company intending to spend almost $200 billion on its infrastructure buildout. On a recent earnings call, Jassy even said he was sure the company has a "clear line of sight to strong financial returns" on that sizable investment. But in the near-term, while AWS figures out its power problems, a lot of that investment may well be sat collecting dust somewhere.

YouTube channel, 修电脑的张哥, roughly translated to Zhang the Computer Repairman, has posted a video this morning because of the sheer volume of Nvidia GPUs they have had sent in for repair in just the past two days alone. And they're all because of burnt interfaces between the 12+4-pin power cable and the card itself.

Seven cards in total have come in for repair (via Uniko's Hardware), mostly RTX 5090 cards, along with Asus and MSI RTX 4090. They all come from various manufacturers; as well as Asus and MSI there is also a Gainward card in the mix as well as a Founders Edition RTX 5090, too.

You can see in the video that either the socket interface is burned, sometimes blackening the PCB, or the 12+4-pin power connector itself has melted into the socket and the user has had to snip the cable off in order to be able to send the card in for testing.

Zhang makes the necessary repairs to each of the cards—removing the interface, cleaning the contacts, and soldering on a new 12+4-pin socket—and then goes ahead and tests each one of them. With the new sockets in place, the repaired cards are back in action and while the GPUs run through some 3DMark benchmarks he checks the temperatures on the cards and the cables.

Some still run a little hot, with the cables at the PSU also running pretty toasty, but nothing particularly worrying at this point.

But still the fact remains that this is a really high volume of cards coming into a single repairshop, all presenting the same symptoms and all from different manufacturers, too. He does also note the users are not necessarily running weak PSUs, and some are even from lab computers (though these might also get hammered more than a card someone is using to game on at home).

It might be that it's a very popular repair channel on BiliBili and YouTube, so potentially he might get sent more than other shops, but it's obvsiously still an issue. It feels like, on a long enough timeline, a whole lot of RTX graphics cards using a 12+4-pin connector are going to suffer from this problem.

One thing Zhang does note is that during use, if you touch the back of your finger against the power cable, maybe an inch or so above the GPU connector, you should feel it being a little warm. But if it's notably too hot to touch then there's likely an issue.

Likewise he suggests keeping an eye on the power draw via GPU-Z, and particularly the 12+4-pin voltage readout. With failing cards, he notes that if the 12+4-pin voltage drops well below 12 V—to 11.3 or 11.4—then there is likely an issue and you should check the temperature by hand at that point.

What's clear though is this issue isn't going away, isn't purely down to dodgy vendor adaptors or old PSUs, or even miss-aligned cables. The amount of power being drawn by these beefy cards is melting cables and connectors, and if you have one of these GPUs it's worth monitoring it on the regular.

That said, if you want a little more peace of mind, there are new gadgets that can offer some. If not complete protection, then some forewarning of an issue before it becomes a melty mess. Such as this new OCP dongle from Super Flower.

Power supply manufacturer Super Flower has launched a new accessory that sits between a graphics card and its power cable to ensure neither ends up a hot, expensive mess as too many Nvidia cards seem to be.

It's called the VGA 12+4 pin OCP Dongle. It's a big lump of plastic and ARGB LEDs that measures for high current across the power pins that make up the 12V-2x6 (or 12VHPWR) connector. This connector is capable of delivering up to 600 watts—enough for an RTX 5090—but it's overcurrent (a current that exceeds the cable/connectors ratings) that could cause a catastrophic failure.

There are really six 'hot' +12V pins on the connector. That means each of those pins needs to carry around 8.33333 amps or thereabouts. That power load doesn't stay perfectly balanced between the pins, due to resistance and imperfections, which means some pins get less amps, while some get more amps. It's when they get a lot more amps that things start to get melty.

Super Flower's OCP Dongle keeps a close watch on those amps, and will indicate when any pin exceeds 9.5 A.

If any one pins is between 9.5 - 12 A, the dongle will light up orange and emit an intermittent buzzing sound.

If any one pin exceeds 12 A, the dongle will light up red and emit a continuous buzzing sound. If this continues for 195 seconds, VGA Black Screen Protection is triggered, which tries to buy you some time to check your connections.

It's a pretty neat solution and one that's not entirely new to me, as I've got something very similar in my own PC: Cooler Master's GPU Shield Adapter Cable. It operates on more or less the same principle, with lights and buzzers to let you know if it's all going to hell in a handbasket.

That might be a more reliable choice right now, as Uniko's Hardware, who spotted this Super Flower dongle, suggests it is only available in Taiwan right now, included in a bundle with a power supply. There are power supplies that offer similar features, too, such as those with MSI's 'GPU Safeguard+'.

Though why I have to deal with this gnawing worry about my expensive graphics card is another thing entirely. Surely it doesn't have to be this way?

DLSS Swapper just became even more useful, as the upscaler-switching software has rolled out a bulk DLL swapping feature in the latest update.

Version 1.2.7.0 adds a Bulk Swap button to the top right hand menu. Click on it, and you can select multiple games (or all of them) from your libraries at once, before individually selecting your upscaling version preferences in one handy menu (via Techpowerup). Hit apply, and they all get updated at the same time.

There are many different iterations of DLSS, FSR, and XeSS to choose from, including separate frame generation options and individual upscaler presets. And should a change cause issues, your games can be restored to their original DLLs and preset settings en masse, too.

It's the little things in life, isn't it? This update also adds DLSS Frame Generation Preset C, which is the new beta version that is supposed to significantly boost performance in exchange for a slight drop in visual fidelity. Up to 35% on an RTX 5060, supposedly.

I've had some odd visual results when messing around with different upscaler versions in certain games (Crimson Desert, I'm looking at you), so the option to easily roll back changes either individually or all at once is a welcome one. I might have a mess around with Preset C later, though, just to see if I can gain a few frames.

A word of warning. DLSS Swapper points out on first boot that messing around with game files could cause some anti-cheat software to flag the changes, so it's probably best to avoid anything with an online component. Lest the banhammer strike for thee.

Nvidia's RTX 5090 GPU has become particularly elusive lately. In light of swirling rumours, Nvidia promises the 5090 isn't dead but constrained supply and runaway pricing has made it a tough card to track down. So, amidst all of this, what does Palit-owned hardware manufacturer Galax do? Tease a gem-studded edition of the RTX 5090.

The GeForce RTX 5090 Hall of Fame Edition was technically already teased back during Computex, though folks thought this showfloor piece was alluding to a whole new GPU generation at the time. Now, we know this special edition is actually current gen and will feature a GPU shroud encrusted with Swarovski crystals (via Techpowerup).

No price has yet been announced ahead of this exceptionally extra card's slated October 22 release date. With how much a standard RTX 5090 costs, odds are that those planning to scoop this shiny GPU will need very deep pockets indeed.

For context, we've stopped including the RTX 5090 in the graphics card price watch, as our Dave notes the card is 'unavailable' from many reputable retailers. Our Jacob notes that the card can be found in places like Newegg's Marketplace for upwards of $9,000.

Those stones are far from the only Extra feature about this special edition GPU. Like many of Galax's other offerings, it'll be available in a striking white. This card will also enjoy dual 12V-2×6 power connectors, much like other Hall of Fame editions. That'll offer a beefy amount of power headroom.

Anyway, I enjoy aesthetic hardware as much as the next person, but even I can see that Swarovski crystals are a level of excess the RTX 5090 didn't really need. Our Andy also argued last year that the Nvidia RTX 5090 Founders Edition is the best looking graphics card he's ever seen. Mind you, I do kind of want to see how those crystals will refract the light from my rig's RGB now. Talk about shining bright like a diamond…

Good news: I know of the existence of two entire pallets worth of RTX 5090 graphics cards—a rare thing in this day and age, as stock appears to have dwindled to virtually zero. Bad news, they've 'gone missing' from a Brazilian airport.

Brazilian electronics distributor Oderço has published an Instagram post drawing attention to the missing Zotac GPUs, which supposedly arrived at Guarulhos International Airport in São Paulo, but then promptly vanished.

"In our more than 38-year history, we have never encountered a situation like this—and we will not remain silent," says CEO Oderço Matos (via Tom's Hardware, machine translation). "Two pallets of Zotac RTX 5090 graphics cards have gone missing at Guarulhos International Airport. The merchandise is worth approximately R$2.5 million (at cost), and to date, we have received no answers."

"We are seeking clarification and want to understand what happened," Matos continues. "How can a shipment of such value disappear from within an international airport without any explanation being provided to us?"

The company has not provided the details of the shipment's arrival date or tracking numbers, but the Oderço Group CEO seems more than a little miffed about the lack of official response towards the missing wares. "We are bringing this matter to the public because we need answers," says Matos. "We expect those responsible for the case to come forward, and we hope that the shipment is located."

R$2,500,000 equates to roughly $495,000, which represents the acquisition cost for the company. I have no idea how many graphics cards that would equate to once they were broken down into individual units, but two pallets worth? Lots, I think I can say that for certain.

I'd also hazard a guess that "gone missing" may be a euphemism for "probably stolen," as I think the wayward GPUs are unlikely to turn up down the back of an airport lounge sofa. I suppose they could always have become mixed up with other freight and shipped off to goodness-knows-where, but one way or the other, the company claims they haven't made it to their final destination.

Has anyone checked in on the whereabouts of Michael De Santa, Franklin Clinton, or Trevor Phillips recently? It seems like their sort of caper. Regardless, the GPUs now appear to be at large—and given retail prices for brand new RTX 5090s are truly jaw-dropping right now, anyone who ends up with a couple of pallets' worth may soon be rolling in cash.

Rumours have been circulating as of late that Nvidia has halted production of its flagship RTX 5090 graphics card, in favour of moving the GB202 silicon towards RTX Pro data center GPUs instead. However, Jeff Fisher, the senior vice president of Nvidia's GeForce business unit, has denied its demise at an event in Taipei.

"The supply chain (capacity) is indeed somewhat constrained, but Nvidia is doing everything it can to get the product on the market," says Fisher (via 4Gamers). "Although there are other variants of the RTX 5090 (chip), production has not been discontinued, and Nvidia will continue to ship, which is very important."

However, the card has disappeared from Nvidia's official US marketplace, and has reached near-unobtanium levels elsewhere. We've stopped tracking the card on our best GPU deals page in recent months, as the only listings we can find have ludicrous prices from third-party sellers.

It appears we have another Schrödinger's graphics card situation to contend with. On the one hand, a (very) senior Nvidia official reportedly says the card is still a-go. Taiwanese outlet CNA quotes Fisher as saying "it's not true" in regards to the ceased production rumours, although this is via machine translation.

On the other hand, while GB202 silicon for consumer RTX 5090 cards may still be under production in some form or fashion, the finished GPUs themselves seem to have gone AWOL. If even Nvidia itself isn't listing them, it's difficult to imagine where meaningful stock might exist elsewhere—although two pallets have gone missing from a Brazilian airport, so there are at least some of the beasties out there somewhere.

Redditors have even taken to pointing out the odd single example sitting on a retail shelf, which has become worthy of note in its own right. The fact that some are now highlighting individual cards, like bird watchers with binoculars around their necks speculating on the existence of a rare warbler, really does make it seem like the Nvidia card is no more. Or, at the very least, so supply constrained as to be functionally unobtainable.

Rumours surrounding another Schrödinger's GPU, the lesser-spotted RTX 5080 24 GB/Super, continue to pop up now and then. But with the RTX 60-series still believed to be a while away yet, and the memory supply crisis still biting, the GPU market really is looking rather barren in the last half of 2026, RTX 5090 or otherwise.

The phrase 'RMA' is likely to send shivers down many a spine, especially when it concerns something like a graphics card. Returning a faulty component can often be a 'cross your fingers and hope for the best' kind of deal, so naturally I am pleasantly—very pleasantly—surprised to hear Asus has enough of a heart to not only give one lucky Redditor an upgrade but also get the colour right.

According to the Redditor's original post about the issue, they sent their white Asus ROG Strix RTX 4080 Super OC in for RMA and were sent back a black Asus TUF Gaming RTX 5080 when they discovered the former couldn't be repaired. That's a win, right? Well, not fully, because the Redditor wanted a white one, like their original RTX 4080 Super, to match their all-white build.

So, what's to be done about that? Naturally, asking for them to swap it: "After about 6 weeks of emails and escalating the case to higher-level management, Asus finally approved the exact card I asked for: the ROG Astral RTX 5080 White OC. I sent the TUF back, and they even upgraded the shipping to FedEx Standard Overnight at no extra cost. The card finally arrived!"

This is an RMA result, if ever I've seen one. It might seem a bit much, but to be honest if you have an all-white build and an expensive GPU that fails, I can see why it makes sense to at least attempt to get a replacement just right.

Funnily enough, I can speak to this very kind of issue with a much cheaper product. A few years ago, I sent my cream-coloured Sennheiser HD 599 headphones for RMA, and they sent back a new pair... except they were black, not the lovely cream colour I sent off. But being the non-confrontational and lazy weasel I am, I just accepted my fate, and those are the headphones I use to this day.

RTX 4080 vs RTX 5080 Super

Comparing the real-world gaming performance of the RTX 4080 and RTX 5080 Super at 1440p resolution, using upscaling and frame gen when possible.

Black Myth Wukong (1440p High + Upscaling (quality))Cyberpunk 2077 (1440p RT Ultra + upscaling (quality))F1 24 (1440p Ultra high +upscsaling (Quality), 2x FG)Homeworld 3 (1440p Epic + upscaling (quality))The Talos Principle 2 (1440p Ultra +upscaling (quality), 2x FG)

Avg FPS

1% Low FPS

RTX 4080 Super

121

94

RTX 5080

132

116

037.575112.5150

Black Myth Wukong (1440p High + Upscaling (quality)) Data

Product	Value
RTX 4080 Super	121 Avg FPS, 94 1% Low FPS
RTX 5080	132 Avg FPS, 116 1% Low FPS

Avg FPS

1% Low FPS

RTX 4080 Super

155

118

RTX 5080

189

123

050100150200

Cyberpunk 2077 (1440p RT Ultra + upscaling (quality)) Data

Product	Value
RTX 4080 Super	155 Avg FPS, 118 1% Low FPS
RTX 5080	189 Avg FPS, 123 1% Low FPS

Avg FPS

1% Low FPS

RTX 4080 Super

223

161

RTX 5080

259

129

075150225300

F1 24 (1440p Ultra high +upscsaling (Quality), 2x FG) Data

Product	Value
RTX 4080 Super	223 Avg FPS, 161 1% Low FPS
RTX 5080	259 Avg FPS, 129 1% Low FPS

Avg FPS

1% Low FPS

RTX 4080 Super

137

66

RTX 5080

136

62

037.575112.5150

Homeworld 3 (1440p Epic + upscaling (quality)) Data

Product	Value
RTX 4080 Super	137 Avg FPS, 66 1% Low FPS
RTX 5080	136 Avg FPS, 62 1% Low FPS

Avg FPS

1% Low FPS

RTX 4080 Super

172

99

RTX 5080

203

94

075150225300

The Talos Principle 2 (1440p Ultra +upscaling (quality), 2x FG) Data

Product	Value
RTX 4080 Super	172 Avg FPS, 99 1% Low FPS
RTX 5080	203 Avg FPS, 94 1% Low FPS

The RTX 5080 performs at least as well as the RTX 4080 Super, and often noticeably better at 1440p.

To give just a glimpse of how difficult and frustrating RMAs can be, consider the fact that we recently discovered that a faded serial number could scupper your chances of being able to RMA your Founders Edition card. So yes, Asus swapping an RTX 5080 out just to get the colour right is pretty sweet. I can only assume the Redditor must have been very persistent. And fair enough, because it worked.

This is all on top of the performance improvement they received, too, of course. As you can see from the chart above, the RTX 5080 performs better than the RTX 4080 Super, and it has access to more of the latest and greatest DLSS tech. Just don't go breaking your 40-series cards and trying the same. Who knows what the RMA gods will have in store for future requesters.