Graphics Cards - Page 7

Ever since the GeForce RTX 40 Series first debuted on the scene back in 2022, led by the 450W flagship GeForce RTX 4090, there's been a steady stream of reports related to cables overheating, melting, and even damaging the GPUs they're connected to. The latest case actually dates back to this era, to the original 12VHPWR connector, which the safer 12V-2x6 connector has since replaced. Of course, there are still issues with the newer connector, but this is a gentle reminder that even a GPU that has been using the older 12VHPWR connector for more than a year is still at risk.

According to a new post on Reddit from 'Dry-Relationship5158,' they were using a GeForce RTX 4090 Founders Edition paired with the original Corsair Power Bridge adapter for 12VHPWR GPU connections. The 180-degree design of the adapter was developed for users seeking a cleaner look and to avoid bending their GPU power cables near the pins. Corsair's product page for the Power Bridge adapter says it's rated for up to 55A and can withstand temperatures up to 105 degrees Celsius, making it the right choice for a "power-hungry GPU" like the GeForce RTX 4090.

For this one Redditor, however, after their PC began freezing and randomly restarting every time they tried to fire up a game, unplugging the Power Bridge revealed a "total nightmare."

Microsoft's Advanced Shader Delivery has officially left the Xbox Insider program and is now available through the Xbox PC app. The update also brings a significant expansion of AMD GPU support, with coverage now extending all the way back to RDNA 1, meaning every Radeon GPU since the RX 5000 series in 2019 is supported.

Advanced Shader Delivery delivers precompiled shaders with the game download, rather than forcing the CPU to compile them locally at launch or during gameplay. This eliminates the long initial load times and traversal stutters that have plagued PC gaming for years. Microsoft's showcase example is Forza Horizon 6, which now loads in around 4 seconds compared to nearly 1.5 minutes without the feature, a reduction of roughly 95%.

To use Advanced Shader Delivery, users need Windows 11 24H2 or newer, Xbox Gaming Services version 37.113.11003.0 or newer, and AMD Adrenalin driver 26.6.1 or newer. The feature currently supports a substantial list of titles, including:

ASUS launched the ROG Equalizer cable this year as a premium solution for 16-pin connector-burning issues that have plagued high-end NVIDIA GPUs since the RTX 4090 era. Priced at $50, it promised balanced power delivery, lower cable temperatures, and more load capacity than standard 12V-2x6 cables. Now, it appears the cable that was supposed to prevent your components from burning has burned itself.

Posted on Chiphell Forums and picked up by UNIKO's Hardware on X, a photo shows a chat log with a user holding a burned ROG Equalizer cable with its distinctive purple connector tint and ROG-branded cable comb. At least three of the 12 primary pins show clear burn marks, with the top-right pin in the worst condition, with the plastic housing fully melted around the connector.

The ROG Equalizer's design had already drawn criticism before this, with hardware tester Der8auer concluding that it actually performed worse than a standard 12V-2x6 connector. Specifically, his tests showed significant current variance between pins, up to 4A on an RTX 5080, meaning some pins were being asked to carry disproportionately more current than others. That kind of imbalance is exactly the mechanism that causes connectors to burn in the first place.

An engineering sample of a reported GeForce RTX 2080 "Ti Super" has emerged, confirming that NVIDIA was toying with the idea of refreshing its entire RTX 20 series lineup with a Super model (not just the RTX 2080, 2070 and 2060). Discovered by Videocardz, a Redditor bought a working engineering sample of an RTX 2080 Ti Super on eBay and shared images of the card to prove its legitimacy.

The card features the same Founders Edition design as all other RTX 20 series (Founders Edition) models, but the color scheme on the Ti Super variant is noticeably different from all other RTX 20 series Super variants, featuring a black color scheme, with a mirror-like finish in between the card's twin fans.

One oddity with the card's exterior is the badging, which only says "RTX 2080 Super" and is missing the Ti nomenclature. If you want to dive into the weeds, this could mean NVIDIA was thinking about giving the RTX 2080 Super RTX Titan-like specs, but likely NVIDIA opted to use an RTX 2080 Super's middle piece for its engineering samples to save time.

A new report from Jon Peddie Research, covering the first quarter of 2026's GPU shipments from AIBs (add-in board partners for NVIDIA, AMD, and Intel), paints a relatively strong picture of the current GPU market. With 11.8 million units shipped in the first three months of the year, it's a 0.6% decrease from the previous quarter; however, this figure is much lower than the historical Q1 average drop of 12%.

Q1 is a quieter month for hardware shipments, so the minimal 0.6% decrease is notable. This covers GPUs in the GeForce RTX, Radeon RX, and Intel Arc lineups, and what makes this figure even more interesting is that it arrives when desktop PC shipments dropped 25% year-over-year and 24% quarter-to-quarter for Q1 2026. According to the data, the AIB attach rate, that is, GPU per desktop PC shipped, increased to 76%, which is up an impressive 33.2% from the previous quarter.

And with that, the new report highlights desktop GPU market share by vendor, with NVIDIA holding strong at 90%. This is a 1% drop from the previous quarter; however, most of that went to Intel rather than AMD.

After a long silence, we've recently seen a spike of activity in terms of rumors for NVIDIA's RTX 5000 Super refresh, and these GPUs have again popped up, this time courtesy of a well-known YouTube leaker.

Moore's Law is Dead's latest video focuses on NVIDIA's GPUs and mainly the rumored Super variants, but also touches on RTX 6000 graphics cards.

And what Moore's Law is Dead (MLID) tells us is pretty much along the lines of the same info he furnished us with last year (in September).

NVIDIA has issued GeForce Hotfix Display Driver version 610.52 in response to its latest official Game Ready Driver release (610.47), which added day-one support for 007 First Light and over 40 new G-SYNC-compatible displays. This new Hotfix Display Driver doesn't seem to be in response to a single major issue, but rather to a number of smaller ones.

It includes fixing frame pacing issues on GeForce RTX 40 Series graphics cards when G-SYNC is enabled, improving gaming stability in multi-monitor setups with DLSS Frame Generation, and fixing an issue where some monitors would not wake when exiting sleep mode. There's even a game-specific fix for World of Warcraft, where it seems that the game's performance and stability were affected by the most recent driver release. So, if you're a WoW player, you might want to install it.

These Hotfix drivers are considered Beta, so they're only recommended for those experiencing any of the issues outlined in the release notes, as the QA process is "abbreviated" compared to regular driver updates and releases. Here are the full GeForce Hotfix Display Driver version 610.52 release notes.

With NVIDIA's new RTX Spark chip for laptops pairing an Arm-based 'Grace' processor with RTX Blackwell graphics, the first wave of ultra-thin, premium laptops sporting the chip is expected to debut later this year. At Computex 2026, we got to go hands-on with a number of them and even see games like Alan Wake 2, Fortnite, Indiana Jones and the Great Circle, Pragmata, and more running and looking good.

And based on our discussions with both NVIDIA and Microsoft representatives at the show, these new Windows on Arm devices powered by RTX Spark are set to usher in a new era for the specialized version of Windows, with more native app support and better-optimized emulation. And with that, we'll probably start to hear a lot more about Windows on Arm updates and support in the lead-up to the RTX Spark launch.

Case in point, today we've got word from UL Solutions, the makers of the popular 3DMark benchmark suite for measuring PC gaming performance. The company has confirmed that it's updating more of its synthetic benchmarks with native Windows on Arm versions.

With no GeForce RTX 50 SUPER Series refresh this year, and the next-gen GeForce RTX 60 Series coming at some point in 2027 or even 2028, Computex 2026 was a little light when it comes to brand-new GPUs built for PC gaming. The good news, however, is that, as we all know, it's simply a matter of time; MSI showcased some of the brand-new cooling technology it will introduce with its next-gen graphics cards.

Basically, the company is set to level up the thermal performance on its more premium GPU offerings with a number of innovations, new technologies, and new features. Starting with the fans, MSI notes that it's going to transition the fan blades on its cards to 0.8mm ultra-thin, durable metal. This will not only improve airflow but also efficiency when spinning at high speeds.

And that's just the beginning, as MSI is set to update its heat pipes to a new, advanced spiral-groove heat pipe. Under a microscope, this shifts the pattern toward a more diagonal orientation rather than horizontal, increasing the contact area and improving heat transfer. And on that note, MSI is also switching to a diamond-composite thermal pad for memory, and a diamond-copper composite baseplate for the GPU and heatsink. MSI notes that this new approach will significantly improve conductivity, and yes, it includes diamond particles.

If we didn't have the current memory crisis and storage crunch brought on by the AI boom and the age of data centers becoming AI factories, the mid-generation GeForce RTX 50 SUPER Series would have released earlier this year. Based on credible leaks we've received about what the SUPER Series would bring to the table, the big thing would be more VRAM, thanks to newer 3GB GDDR7 modules.

Now, it's worth noting that the GeForce RTX 50 SUPER Series was never formally announced, so the delay and scrapped launch are still technically rumor. And with that, there's a new report over on Benchlife (via VideoCardz) that says the GeForce RTX 50 SUPER Series is back on the menu, so to speak, with NVIDIA planning on announcing the lineup at CES 2027, with the launch to happen soon after.

As for the lineup, previous reports have pointed to a GeForce RTX 5070 SUPER with 18GB of memory, alongside a GeForce RTX 5070 Ti SUPER and a GeForce RTX 5080 SUPER with 24GB of memory. The good news is that with 3GB chips the memory interface won't have to change, so it should be a fairly easy upgrade for NVIDIA's partners. Interestingly, with this new report, there's also a fresh bit of information/rumor with the lineup set to include a GeForce RTX 5060 SUPER 12GB.

Without a GeForce RTX 50 SUPER Series refresh this year (thanks to the current memory crisis and AI boom), or any new models on the horizon, we're basically in that period of waiting for the next-generation to arrive. However, this doesn't mean we won't see anything new, as at Computex 2026 we got to see a couple of prototype designs from INNO3D for a pair of new premium and flagship GeForce RTX 50 Series cards that could come to market sooner or later.

Now, it's worth reiterating that these designs are prototypes and not finalized, but we were told they would be paired with higher-end models in the GeForce RTX 50 Series lineup, such as the RTX 5070 Ti, RTX 5080, and RTX 5090. The first design immediately caught our attention, as it features a dual-fan push-and-pull cooling design reminiscent of NVIDIA's Founders Edition models that debuted with the RTX 40 Series.

Basically, you've got one large fan on either side of the card, with INNO3D noting that the minimal all-metal design took inspiration from the Founders Edition lineup. Albeit, here you've got a cooler designed for overclocked performance, with a few interesting touches like a fin stack featuring an almost 3D checkerboard pattern, and the 12V-2x6 connector located on the rear of the GPU rather than the side. Available in black or white, it's definitely stylish and something we'd love to see paired with a GeForce RTX 5080.

Six years after the heyday of the Ampere GPU lineup, a desktop GeForce RTX 3050 Ti has shown up online, and it is a card NVIDIA never actually released. Leaked by X/Twitter user @GOKForFree, the mysterious sample confirms that a desktop version did exist during the Ampere era, it just never made it to retail shelves.

The card is reportedly an engineering sample, and it carries the GA106-200-A1 die, the same silicon used in the RTX 3060, but with 4 of its streaming multiprocessors disabled. That leaves 3,328 CUDA cores active, along with 104 TMUs and 48 ROPs, on a 192-bit memory bus. The specimen is labeled by an unfamiliar brand, "Robiny," with stickers identifying its specs. It runs a dual-fan, dual-slot design with a single 8-pin connector.

The unreleased card also shows up in GPU-Z, but the software labels it simply as "NVIDIA Graphics Device." Screenshots show a 1,410 MHz base clock and a 1,665 MHz boost clock, with GDDR6 memory running at 1,750 MHz for 336 GB/s of bandwidth. The power limits sit at 120W default, with a 100W minimum. The sample was apparently tested on old Ampere-era drivers from mid-2022, which tells you roughly how long this prototype has been sitting around.

Intel's discrete desktop GPU ambitions are alive, at least according to one of its top executives. At a media Q&A during Computex 2026 in Taiwan, Intel EVP and GM of Client Computing Alex Katouzian was asked directly by Tweakers about the future of Arc desktop graphics cards.

He said that desktop GPUs remain a "super important" part of the company's PC lineup, pointing to gaming as a major revenue driver across both mobile and desktop. He also said Intel is seeing good traction with its GPU cores, and that gamers and game engine developers are actively working with the company.

Despite Alex's optimism, the comments come at an awkward time. The Battlemage generation only produced two gaming desktop cards, the Arc B580 and B570, and the long-rumored "Big Battlemage" Arc B770 has still not materialized. The B770 was expected to drop this year, but Intel chose to limit that SKU to the professional and workstation market.

AMD's Radeon RX 9060 XT has reached another extreme-overclocking milestone, with overclocker Bill "Sampson" Alverson pushing an engineering sample of the GPU to 4,643 MHz, about 48% above AMD's listed boost clock. The result is reportedly the second-highest GPU frequency ever recorded, just behind the current world record of 4.769GHz.

That world record belongs to AMD and overclocker Splave, set on the same card at 4.769GHz. Sampson did not beat that record this time, but the result still puts him in rare company. Only two discrete GPU overclocking runs have ever exceeded 4.0GHz, and both involve the RX 9060 XT.

The card used for this run was an engineering sample, and the screenshots show custom AMD internal software being used for liquid-nitrogen overclocking. One visible command references "L1N2control.exe" with settings targeting 4,650 MHz and 1,610 mV. These appear to be non-public AMD utilities, firmware, or BIOS modifications that allow far greater control over voltage and frequency than standard tools like MSI Afterburner would ever permit. The motherboard used was reportedly an MSI board, though the exact model has not been confirmed.

ASUS is taking a page from EV vehicle designers and testing a higher operating voltage with the 12V-2x6 power connector. ASUS demoed a modified RTX 5090 at Computex 2026 featuring a 12V-2x6 power connector running at 48 volts with a 1200-watt power limit.

Images from ComputerBase show the modified GPU running 3DMark while consuming over 1000 watts of power. The GPU is able to do this thanks to the higher-voltage architecture, allowing the RTX 5090's 16-pin power cable to operate at significant wattages without overheating. ASUS also had to modify the power supply to run at 48V, noting (in a press release) that it has developed a unit that can adapt between 12V and 48V automatically when the PSU detects a compatible (48V) GPU.

In scientific terms, higher voltages translate into lower amperage. ComputerBase did the math, noting that at 48 volts, the 12V-2x6 power connector's individual wires only need to operate at 3.5 amps to achieve a blisteringly high 1000 watts of power delivery. By contrast, at 12V (the power connector's official operating voltage), each wire needs over 12 amps to achieve just 600W of power delivery.

ASRock is one of AMD's key Radeon gaming partners, so much so that our flagship review of the new Radeon RX 9070 GRE was released this week, focused on the company's impressive all-new dark Steel Legend design. A rare 2026 gaming GPU release, as the focus gradually shifts to the next generation of Radeon graphics powered by RDNA 5.

Although the current market is pretty bad for consumers, with seemingly all capacity going to building AI factories and data centers, RDNA 5 is still on the horizon and should debut. Current rumors point to a mid-2027 release, with the design for AMD's next-gen graphics architecture reportedly in its final stages. Of course, the 2027 window is due to the current DRAM crisis, with previous reports indicating that RDNA 5 could have been ready by the end of this year.

With that in mind, it's not outside the realm of possibility that companies like ASRock are already developing and finalizing GPU designs, cooling, and specs for the next generation of Radeon graphics cards. And with that, relatively hidden in ASRock's booth at Computex 2026, behind a highly reflective tempered glass panel, we spotted a next-generation Radeon GPU design for its Phantom Gaming series.

Intel used its Computex 2026 stage time to pull back the curtain further on Crescent Island, its upcoming data center GPU built on the Xe3P architecture. The chip is aimed squarely at AI inference workloads, and the headline detail from the presentation is that board partners will be able to build configurations with up to 480GB of LPDDR5X memory onboard.

The 160GB figure that Intel had already confirmed represents the reference configuration, while the 480GB ceiling opens the door for AIB partners to build much denser variants, similar to how add-in board partners approach gaming GPU memory configurations.

What makes Crescent Island stand out from the competition is its choice of memory. While conventional AI accelerators from NVIDIA and AMD rely on expensive high-bandwidth HBM memory, Intel's chip uses consumer-oriented LPDDR5X DRAM and is designed to run in air-cooled servers rather than liquid-cooled setups.

NVIDIA's five-year-old RTX 3060 has made a comeback sooner than expected. Earlier reports pointed to a July return, but the Ampere-based GPU has already appeared on shelves in China, presumably ahead of a wider global rollout. According to ChannelGate, the GeForce RTX 3060 12GB is now available in mainland China.

The specific model making its entry is the Colorful Battle-Ax GeForce RTX 3060 DUO 12GB V2. The bulk wholesale price is 2,199 yuan, or around $325, while VideoCardz highlights that Colorful's official store lists it at 2,349 RMB, which comes to approximately $347. For context, the card's original MSRP back in 2021 was $329. The listing also shows shipment starting this week, with delivery expected before June 5.

The initial allocation is reportedly very limited, with some regions receiving only a few dozen units. Colorful is reportedly planning weekly restocks to build toward a stable supply, but it will likely take a couple of months before the GPU is fully available again at scale. Apart from Colorful, previous reports also named ASUS, MSI, and GALAX among the board partners expected to receive allocation.

With the arrival of the GeForce RTX 50 Series last year, and the new flagship RTX 5090, ASUS introduced a brand-new premium Republic of Gamers GPU design with its new Astral series. In addition to the stylish and robust physical design with impressive cooling, the notable thing was its inclusion of four fans in a new quad-fan cooling system. And with ASUS celebrating 20 years of ROG at Computex 2026, the company is introducing the new ROG Astral GeForce RTX 5090 Edition 20.

The company notes that this new model, which sports a stylish black-and-gold look, "pushes both performance and visual expression to new extremes," and it's not hard to see why. And that's because it features a curved AMOLED display that wraps around the sides and rear of the card, and can be configured to display real-time hardware information, custom animations, or dynamic 3D visuals crafted by the ROG team. As the 4.7-slot thickness makes it one of the thickest GPUs on the market, adding a large screen does add a new and impressive aesthetic layer to the Astral design.

And it pairs perfectly with ASUS's new 20th-anniversary-themed AIO cooler, letting users synchronize the displays of both pieces of new hardware to create a visually striking build. As an Astral GPU, you've also got 20% improved airflow thanks to the fourth fan visible on the backplate, plus premium cooling for overclocked performance with its vapor chamber and liquid-metal GPU cooling.

AMD introduced the Radeon RX 9070 GRE GPU, or Golden Radeon Edition, last year as a China-exclusive. Utilizing the same Navi 48 GPU as the Radeon RX 9070 and Radeon RX 9070 XT, albeit in a cut-down form with 12GB of GDDR6 memory instead of 16GB, it was positioned as a more affordable entry point to 1440p gaming, leveraging the company's latest RDNA 4 architecture. Cut to June 1, 2026, and the Radeon RX 9070 GRE is getting a global release with an MSRP of $549 USD.

The announcement and launch arrive alongside AMD's Computex 2026 announcements, coming at a time when we weren't expecting any new GPU releases due to the current memory, chip, and storage crisis. In fact, the current market reality means that it replaces the baseline Radeon RX 9070 16GB GPU at the $549 price point, which is seeing its official MSRP increase to $619 USD. Based on current GPU pricing, the $549 Radeon RX 9070 GRE makes the GeForce RTX 5060 Ti 16GB its closest GeForce competitor, not the GeForce RTX 5070.

And with that context, AMD notes that the Radeon RX 9070 GRE, a GPU built for 1440p gaming, is 22% faster than the RTX 5060 Ti 16GB at this resolution across 40 titles, with and without ray tracing. This is something we've been able to confirm as we're publishing two Radeon RX 9070 GRE reviews today, providing a detailed look at its performance and features across two models from AMD partners ASRock and Sapphire. Plus, our reviews highlight how the new FSR 4 and FSR 4.1 add even more value to the GPU thanks to its fantastic image quality at 1440p and 'free' performance boost.