Graphics Cards - Page 6

It wasn't too long ago that we began to hear rumblings that NVIDIA was planning to re-release the Ampere-era GeForce RTX 3060 as a response to the current component and memory crisis. The idea is that it would be a more cost-effective mainstream solution due to its use of 12GB of GDDR6 memory compared to the more advanced GDDR7 memory found on the GeForce RTX 50 Series lineup, including the GeForce RTX 5060.

According to a new report from the German outlet hardwareLUXX, the GeForce RTX 3060's return with GDDR6 memory has arrived, with a brand-new ASUS Dual GeForce RTX 3060 OC V2 popping up in various retail outlets. The report notes that the lowest price of 333.64 EUR (about $380 USD) is notably higher than the RTX 3060's historic price of 245 EUR.

Although the big selling point compared to 8GB cards like the GeForce RTX 5050, RTX 5060, and RTX 5060 Ti 8GB is the increased VRAM capacity, the value proposition compared to the more affordable GeForce RTX 5060 is questionable based on the improved DLSS 4.5 support that includes Dynamic Frame Generation not being available. Even when it comes to raw 1080p and even 1440p gaming performance in titles that aren't VRAM capacity constrained, the GeForce RTX 5060 is still decidedly faster.

Just over a week ago, AMD Software: Adrenalin Edition 26.6.2 Driver was released for Radeon GPUs, and it was something of a surprise drop as it introduced FSR 4.1 Upscaling support to RDNA 3 graphics cards earlier than expected. For many Radeon RX 7000 Series owners, including those with the previous-gen Radeon RX 7900 XTX flagship, it was akin to getting a free double-digit performance upgrade.

However, not long after the driver was released, users began reporting errors when trying to get the new Adrenalin software to fire up on Windows 10. AMD, to its credit, responded in a timely fashion, advising users to roll back as it worked on a solution. And it arrived a few days later as the AMD Software: Adrenalin Edition 26.6.3 Hotfix Preview Driver, a non-WHQL release to resolve the problem and get RDNA 3 gamers on board with the FSR 4.1 hype train.

Cut to this week, and we've now got the AMD Software: Adrenalin Edition 26.6.4 Driver, an official WHQL release for Radeon RX 7000 Series owners that resolved the Windows 10 issue alongside fixing crashes in "some games" when enabling FSR 4.1 Upscaling.

AMD is actively working on a trimmed-down version of FSR 4 for RDNA 3/3.5 iGPUs, but there is no confirmation of when it will be arriving. In the meantime, a YouTuber has discovered a method of injecting FSR 4 at the driver level that works with RDNA 3/3.5 iGPUs. Channel AncientGameplays published a video showing how to do this using a third-party tool known as RadeonTuner.

RadeonTuner is designed to serve as a lightweight alternative to AMD's Adrenalin software. The application is functionally very similar to Adrenalin, with support for GPU overclocking and fan tuning. The application also allows users to change driver-related graphics settings and display settings. The main feature AncientGameplays was most interested in is the app's built-in overrides for FSR upscaling, frame generation and ray regeneration.

In testing with Spider-Man 2, the YouTuber discovered that RadeonTuner uses the same FSR override method as AMD's Adrenalin software to inject newer iterations of FSR into games. With the FSR override disabled, Spider-Man 2 showed upscaling support for FSR 3.1.5, but turning on the override changed the FSR version in-game to FSR 4.1.1. Spider-Man 2 also showed an asterisk next to the FSR 4.1.1 name, confirming the FSR version was being injected from the GPU driver. The YouTuber also tested several games using this FSR override method using a Strix Halo APU with an RDNA 3.5 iGPU and found it worked almost flawlessly (some issues with shadows and reflections reportedly occur at aggressive upscaling settings).

Power-related issues have been a running theme with NVIDIA's RTX 4090 and RTX 5090 cards. Despite newer 16-pin standards that were meant to be safer, repair shops and media outlets have become familiar with these cards ending up on their benches. The latest case out of Vietnam shows some of the worst power-related damage we have seen yet, severe enough to completely destroy a GPU and its VRAM.

The report comes from Vietnamese GPU repair outlet quyle.gpufix, which received two RTX 5090 cards for repair. The first had already been tampered with before arrival, arriving with no power, no GPU detection, and a missing A1 VRAM package. That card was repaired and returned the same day. The second was a different story entirely.

The second RTX 5090 had its 16-pin connector blown out completely, leaving no physical trace. The burn damage was severe enough to expose the copper layers of the PCB, and the destruction extended to both the GPU die and the VRAM, rendering them dead. Out of two cards, only one made it out.

AMD has issued a hotfix driver for Windows 10 users to fix a problem preventing Radeon gamers from running Adrenalin driver 26.6.2 - AMD's first driver with FSR 4.1 enablement for RDNA 3 GPUs, on Windows 10. AMD announced the new hotfix driver on X, version 26.6.3, and further elaborated that the hotfix will only be needed for a week, as the team gets ready to push the Windows 10 installation fix to an upcoming WHQL update next week.

Adrenalin driver 26.6.2 was one of the most important driver updates AMD has released this year, featuring FSR 4.1 upscaling support for previous-generation RX 7000 series GPUs and RDNA 3.x integrated graphics chips. After the driver's release, gamers quickly discovered that the update was broken on Windows 10, forcing Windows 10 users to switch back to an older revision. The issue specifically relates to a conflict with the driver and AMD's control panel software, which spits out an error message telling the user that the control panel version is incompatible with the driver. The issue is widespread, affecting a broad range of hardware, including Ryzen 9000 series CPUs and RX 9000 series GPUs. Windows 11 users are apparently not affected for unknown reasons.

AMD could have waited a week to get this fix out to everyone, but demand for AMD to extend FSR 4 support to older RDNA 3 and RDNA 2 GPUs has been significant over the past year on social media and comments on various AMD YouTube videos. Likely incentivizing AMD to push out a hotfix driver to fill in the gap. It's worth mentioning that hotfix drivers do not represent the most stable releases, so if you value driver stability, you might want to skip this update. But, for those who want FSR 4.1 support right now and are running an RX 7000 GPU and Windows 10, AMD's latest hotfix driver is the best option.

The infamous 12V-2x6 power connector has claimed another victim. This time it is the media outlet, Club386, whose GeForce RTX 5090 Founders Edition and their be quiet! Dark Power 13 PSU were both destroyed after a properly seated, single-cable 16-pin connection melted at both ends. The cable was not a third-party product, and the connector was fully seated. The team has been testing GPUs for decades, so it is unlikely to be user error, but none of that mattered.

NVIDIA has since replaced the GPU, but the damage was done, and the point stands regardless. The melting connector problem goes all the way back to the RTX 4090 launch in late 2022, when the original 12VHPWR standard started showing up in repair shops at an alarming rate. The root cause was uneven power distribution across the connector's pins.

Thermal camera testing by overclocker "der8auer" revealed one wire hitting 150°C on an RTX 5090 while other pins carried almost no load at all. When one pin absorbs the bulk of a 575W card's power draw, something burns.

After teasing the design earlier this year at CES, and then showcasing a more final version at Computex, GIGABYTE's new AORUS GeForce RTX 5090 INFINITY graphics card has debuted with a unique physical design alongside rare double flow-through cooling. And with that, it looks like GIGABYTE is expanding its INFINITY lineup with a stylish white-colored AORUS GeForce RTX 5080 INFINITY WOOD graphics card, and two additional RTX 5080 and RTX 5070 Ti models.

AORUS GeForce RTX 5080 INFINITY WOOD is certainly a looker, as it takes the darker INFINITY design and gives it a bright white makeover with added wood panels. Based on the specs found on the official product page for the card, it shares the same physical quad-slot dimensions as the GeForce RTX 5090 model and the same premium WINDFORCE Hyperburst cooling system.

As a premium GPU, the AORUS GeForce RTX 5080 INFINITY WOOD also arrives as an OC model with an out-of-the-box Boost Clock of 2,805 MHz, a sizeable increase over the 2,617 MHz reference spec and NVIDIA's Founders Edition model, which sports a similar double-flow-through cooling design.

Earlier today, AMD released AMD Software: Adrenalin Edition 26.6.2 for Radeon GPUs, which added FSR 4.1 Upscaling support to the RDNA 3-powered Radeon RX 7000 Series graphics cards. This earlier-than-expected release of FSR 4.1 for RDNA 3 is definitely welcome, as it will help breathe new life into previous-gen GPUs like the mid-range Radeon RX 7800 XT and flagship Radeon RX 7900 XTX.

To make FSR 4.1 compatible with RDNA 3, AMD had to develop a new version of the technology that uses the INT8 format, since the RDNA 4 version was built specifically for the new architectures' advanced AI hardware and the FP8 format. The good news is that FSR 4.1 for RDNA 3 delivers no quality loss compared to running on RDNA 4 GPUs. However, as per the Adrenalin Edition 26.6.2 release notes and word from AMD, this new version is limited to RDNA 3 desktop GPUs. At least, officially.

Turns out the new FSR 4.1 DLL leaked a little early, via Valve and the Proton Experimental channel for SteamOS/Linux, of all places. And, like the previous version, when paired with the OptiScaler tool to force it, it can run on systems and configurations that aren't desktop RDNA 3 GPUs. In fact, according to this thread over on Reddit, some users are getting it to run just fine on RDNA 3.5 GPUs.

AMD has officially pushed FSR 4.1 support out the door ahead of schedule to previous-generation RX 7000 series GPU owners through a new driver update. Adrenalin version 26.6.2 has been released with FSR 4.1 support for RDNA 3 GPUs, along with game support for Assassin's Creed Black Flag Resynced, Doom: The Dark Ages | Revelations, and two bug fixes.

Adrenalin 26.6.2 marks the halfway point in AMD's endeavor to get FSR 4.1 working on older RX 7000 and RX 6000 series GPUs (RDNA 2 support is coming in 2027). Last month, AMD made the announcement to officially support FSR 4.1 on the aforementioned hardware, but confirmed a deadline of July for RX 7000 support, making AMD's latest Adrenalin driver arrival at least eight days early. AMD gamers won't have to worry about quality compromises compared to FSR 4.1 on RDNA 3 GPUs; AMD confirmed that FSR 4.1 on RDNA 3 will feature the same image quality as on RDNA 4 GPUs. The only caveat with AMD's current FSR 4.1 adaptation for RDNA 3 is that only the upscaling portion was ported over, leaving FSR 4.1 frame generation as an RDNA 4 exclusive feature.

Pressure to push FSR 4 to older hardware began almost a year ago when AMD accidentally leaked its own FSR4 source code featuring two versions of the upscaler: the out-of-the-box version with FP8 support and a second build using INT8 instructions. The second version with FP8 support was what grabbed everyone's attention, as it enabled FSR 4 functionality on non-RDNA 4 hardware. Afterward, modders were able to grab the INT8 version (before AMD took down the source code), and implement it in apps such as Optiscaler, enabling users to upgrade supported games from FSR 3 to FSR 4 with the INT8 build.

AMD has been in the news lately with its RDNA 4 lineup, but the focus is already shifting to what comes next. A new leak from Moore's Law is Dead (MLID), sourced from an unnamed major OEM partner, points to a mid-2027 launch window for RDNA 5-based Radeon desktop GPUs, with that source adding they would "be shocked" if the cards didn't arrive in 2027.

Importantly, this lines up with reports from Computex 2026, where OEM partners were already expecting an RDNA 5 release in the mid-to-late 2027 timeframe, adding credibility to the claim.

The timing also makes a lot of sense given the broader AMD roadmap. Both the next-gen Xbox "Project Helix" and the PlayStation 6 are reportedly powered by RDNA 5 semi-custom APUs, and both consoles are targeting a 2027 launch window as well. If the silicon is ready for console production, desktop Radeon cards are a natural follow-on.

A user on a Taiwanese hardware forum reported that their ASUS ROG Astral GeForce RTX 5090 left visible discoloration on the PCH heatsink of an ASUS ProArt X870E-Creator WiFi motherboard after about 6 months of use. The system was installed entirely in ASUS hardware, including an ASUS ProArt PA602 case, making the incident particularly awkward for the company.

The discoloration appeared in the area of the motherboard directly below the graphics card. The user said the mark could be partially removed with a wet cloth and some pressure, but the heatsink could not be restored to its original condition. The graphics card itself showed no visible issues and continued working normally. The risk becomes even greater in systems where airflow around the card is not optimized.

The RTX 5090 is known to push extreme power figures, with typical load consumption around 700W and peak figures reportedly approaching 1,000W in certain scenarios. While motherboard heatsinks are usually coated with heat-resistant paint, heat radiating downward onto nearby components can still become a genuine concern at those power levels.

An enthusiastic RTX 5090 owner attempted to install a second 12V-2x6 power connector on their flagship Blackwell graphics card in a desperate attempt to reduce the chance of the connector melting. Unfortunately, the owner ended up bricking the graphics card as a result of the mod and was forced to send the card to a repair shop to get the GPU fixed. YouTube channel northwestrepair covered the story and showed how the 5090 owner was able to install a second power connector on their card. Luckily for the owner, the card was saved after the technician transplanted the GPU and memory to a new PCB.

The RTX 5090 owner had an air-cooled AORUS variant with a second 12V-2x6 slot on the PCB, making it possible to solder on a second power connector to the card. At first glance, the existence of this slot makes it seem as if the card can support two connectors, but unfortunately, northwestrepair revealed that the card is not wired from the factory to support two connectors simultaneously. Specifically, the card reportedly lacks any shunt resistors wired up to the second power connector slot, making it impossible for the PCB to monitor and regulate power coming from the second connector. As a result, when the owner went to power the card with the secondary plug connected, the card overheated and burned a hole through the PCB.

The unfortunate RTX 5090 owner was likely inspired by existing RTX 5090 graphics cards with dual 16-pin connectors and modders who have successfully installed a second power connector on RTX 5090s. Earlier this year, reports surfaced that a modder had successfully added a second 12V-2x6 power connector to a GIGABYTE AORUS Xtreme Waterforce RTX 5090, which coincidentally has the same PCB layout as the AORUS RTX 5090 repaired in this story. MSI's RTX 5090 Lighting Z and Galax's RTX 5090 HOF are also cards that come equipped with two 16-pin power connectors from the factory.

It looks like GPU prices just will not catch a break in 2026. NAND and DRAM prices have already gone through the roof thanks to the AI boom, and now GPUs are not far behind. According to Gazlog, which picked up the report from Chinese board channel sources, AMD is gearing up for another Radeon RX 9000 price increase of around 10-15%, expected to kick in during Q3 2026, possibly as early as July.

The reason behind the price hike is thought to be the rising GDDR6 VRAM costs. Since AMD bundles GPU dies with VRAM before shipping kits to its AIB partners, any jump in memory procurement costs gets passed straight down the chain. The spot price of GDDR6 has ballooned since late 2025, and AI data center demand shows no sign of letting up. As memory gets more expensive, graphics cards get more expensive too.

We've been covering this trend for a while. TweakTown reported the first confirmed RX 9000 price bump back in December 2025, followed shortly by AMD's broader price-hike announcement. Even AMD's own VP David McAfee, when we interviewed him at CES 2026, wouldn't rule out further increases. So the pattern here is pretty well established.

AMD has worked with the HandBrake development team to fix two threading bottlenecks that were affecting transcoding performance on high-core-count Threadripper CPUs. The fixes are included in HandBrake 1.11.0 and later, and the results are significant enough that owners of existing Threadripper hardware can expect a substantial free performance boost simply by updating the software.

According to AMD, video transcoding workloads generally benefit from higher core counts. However, that was not always the case with its Threadripper lineup, where performance sometimes scaled in the opposite direction, particularly at lower resolutions, resulting in longer transcode times. AMD says it identified two key causes behind the issue.

First, HandBrake wasn't built to manage systems with more than 64 logical processors. On chips like Threadripper, which have far more than 64 threads, the application left compute resources idle rather than distributing work across all available cores.

RTX 4090 scams on Chinese second-hand markets are nothing new, but a repair channel on Bilibili has uncovered one that is a little too "plastic." Chinese repair channel Brother Zhang purchased a broken ASUS GeForce RTX 4090 for 1,500 RMB and tore it down to discover, unlike some previous scams using older GPU dies, this one used a piece of plastic made to resemble the GPU core.

The graphics card carried markings meant to resemble an NVIDIA AD102-300-A1 GPU, and it looks real for the most part, unless you start tearing it apart. UNIKO's Hardware on X noted that the markings did not follow NVIDIA's actual nomenclature, and the date code on the die suggested the chip was manufactured in 2030, which is obviously impossible for a retail RTX 4090. The substrate also lacked the usual QR code in the bottom-left corner, and the surrounding capacitors were in the wrong positions compared to a real AD102 layout.

When Brother Zhang physically touched the die and cleaned off the thermal paste, the difference in texture was immediately apparent. What looked like a GPU was a plastic piece with fake markings etched onto the surface. There was no silicon underneath at all, meaning the card had zero functional compute capability and could not even output a display signal.

Modders have figured out how to double the memory capacity of NVIDIA's six-year-old entry-level GTX 1650 from 4GB to 8GB. YouTube channel Paulo Gomes published a video showing off their work modding the entry-level Turing GPU with 8GB of GDDR6, and showed off impressive performance results from the card due to the extra memory capacity. For instance, in Unigine Superposition, the GTX 1650's benchmarking score doubled from 624 points to 1245 points moving from 4GB to 8GB of VRAM.

Upgrading the memory on the GTX 1650 was straightforward and required no extra firmware modifications. The modders swapped out the GTX 1650's existing 1GB GDDR6 modules with 2GB Samsung HC16 GDDR6 modules by desoldering the old modules and soldering the new modules on the PCB. Their first attempt at booting the card failed as a result of installing two faulty HC16 modules, but after swapping out the faulty modules with working counterparts, their second attempt was successful.

One major caveat the modding team noted was that their 8GB mod requires a very specific variant of the GTX 1650 to work - the TU106 version of the GTX 1650 with GDDR6 memory. For the uninitiated, NVIDIA has four different variants of the GTX 1650 (not even counting the GTX 1650 Super), featuring TU117, TU116, and TU106 dies paired to either GDDR5 or GDDR6 memory. The TU106 version was one of the later trim models NVIDIA launched to use up defective TU106 silicon. TU106 was one of NVIDIA's most popular Turing dies and was used to make nearly all of NVIDIA's mid-range RTX 20 series GPUs, such as the RTX 2060, RTX 2060 Super, and RTX 2070. The modders did not fully explain why only the TU106 version is compatible, but likely memory controller and/or firmware limitations are the culprit on other variants of the GTX 1650.

GIGABYTE's new AORUS GeForce RTX 5090 INFINITY graphics cards, created in part to celebrate the company's 40th anniversary, not only feature a premium quad-slot design befitting of a flagship GPU, but a unique one as well. Even though it sports a double flow-through cooling design reminiscent of NVIDIA's own Founders Edition model, the look reminds us of a futuristic spaceship's twin engines and a retro stereo from the 1980s.

The AORUS GeForce RTX 5090 INFINITY's WINDFORCE Hyperburst cooling system includes two large Hawk fans alongside a hidden, smaller fan underneath a mesh panel. With double flow-through cooling, air is pushed through both sides of the reinforced backplate, with composite metal grease, superconducting heat pipes, and a direct-touch vapor chamber found underneath the hood.

It's as premium as it is stylish, and as an OC model it ships with a Boost Clock speed of 2730 MHz, which is notably higher than the 2407 MHz found on the Founders Edition model. Unfortunately, its debut comes amid the current memory and component crisis, which has seen GeForce RTX 5090 prices rise dramatically over the past year. And with that, a MicroCenter listing for the AORUS GeForce RTX 5090 INFINITY puts the price for this new flagship at $5,299.99 USD.

NVIDIA has published a new game-ready driver, version 610.62, that fixes a plethora of issues that were introduced with the first driver in the 600-series branch, 610.47. The most severe of these bugs include V-Sync-related bugs and frame pacing issues when running G-Sync monitors on RTX 40 series GPUs.

Critically, this driver takes all of the bug fixes from NVIDIA's recent 610.52 hotfix patch and applies them to 610.62, while adding two additional bug fixes. An Apex Legends glitch was fixed that would cause visual corruption after extended time playing, and a DLSS-related bug was fixed where DLSS settings were greyed out in certain games after updating (not clean installing) the GPU driver to 610.47 (typed as "615.47" in the patch notes by mistake).

The most significant improvement with this driver update is the G-Sync fix. Starting with 610.47, many GeForce gamers went on various forums to complain about severe frame pacing issues with systems paired with G-Sync monitors. I experienced this issue personally on my 5800X3D/RTX 4070 rig paired to an Alienware AW3225QF, where the monitor's refresh rate was fluctuating constantly to 240Hz and back down to whatever frame rate the game was running at multiple times per second. The issue was so severe it made games nearly unplayable with G-Sync turned on. NVIDIA's patch notes reveal that this specific issue is related to Ada Lovelace RTX 40 series GPUs and only certain G-Sync displays.

AMD's FSR 4.1 is coming to more Radeon gamers, starting with desktop RDNA 3 support arriving next month, which brings the vastly superior AI-powered version of FSR to the Radeon RX 7000 Series of graphics cards. Up until now the technology has been exclusive to the latest RDNA 4 generation and cards like the Radeon RX 9070 XT, but that's about to change.

Sitting down with AMD's Chief Software Officer, Andrej Zdravkovic, and AMD's Senior Director of Software, Terry Makedon, TechPowerUp has learned that the version of FSR 4.1 that will debut for RDNA 3 GPUs will arrive with no quality loss when it comes to the final image displayed on screen. This is good news because AMD has had to rebuild and optimize FSR 4.1 to run on RDNA 3's first-gen AI hardware, which uses INT8 data, rather than the more efficient FP8 used on RDNA 4, which served as the basis for FSR 4.

According to the report, AMD has had to modify the underlying model to support RDNA 3, but the end result for image quality will be the same. For gamers with cards like the Radeon RX 7800 XT and previous-gen flagship Radeon RX 7900 XTX, FSR 4.1 support will unlock a new level of performance and image fidelity, especially when gaming at higher resolutions. As for bringing FSR 4.1 to RDNA 2 and GPUs from the Radeon RX 6000 Series, it's a lot more complicated, hence the 2027 release window.

Yes, after a year or so since its release, the flagship RDNA 4 graphics card from AMD has finally made its way into Valve's Steam Hardware & Software Survey results for May 2026. And it's a strong debut, with the Radeon RX 9070 XT's 1.33% market share making it one of the Top 25 most-used discrete gaming GPUs on the platform, and also making it the most popular Radeon GPU amongst PC gamers.

The reason for the Radeon RX 9070 XT's sudden appearance in the latest Steam Hardware & Software Survey could be chalked up to recent updates Valve has made to Steam's GPU reporting - read about that here. And the 9070 XT's immediate rise to become the most popular discrete Radeon GPU on PC is fantastic news for RDNA 4, even if it trails its direct competitor, the GeForce RTX 5070 Ti. Basically, it's the best showing for a Radeon GPU since the RDNA 2-era.

And with that, the mainstream Radeon RX 9060 XT has also made the cut, now holding more market share than the Radeon RX 7800 XT, the most popular card from the previous RDNA 3 generation. As for the Top 25 PC Gaming GPUs for May 2026, not much has changed aside from RDNA 4 finally getting its due. The GeForce RTX 3060 remains the most popular card among gamers, while the GeForce RTX 5070 continues to cement its position as the card of the current generation.