Graphics Cards - Page 27

Intel's XeSS 2.1 joins AMD's FSR 3.1 as a platform-agnostic solution for Super Resolution and Frame Generation, available to all GeForce RTX and Radeon RX gamers as long as they've got at least a GeForce GTX 10 Series or Radeon RX 5000 Series card.

Intel's XeSS 2, which is the company's answer to NVIDIA DLSS and AMD FSR, has just been updated with the XeSS 2.1 SDK for game developers, adding support for both NVIDIA GeForce RTX and AMD Radeon RX graphics cards. Intel's XeSS 2.1 includes the AI-powered XeSS Super Resolution, XeSS Frame Generation, and Xe Low Latency technologies to enhance in-game performance and responsiveness.

Like the original XeSS, which included support for non-Intel Arc graphics cards, it's expected that image quality for upscaling and frame generation on NVIDIA and AMD GPUs will not be as good as what you'd get on a dedicated Intel GPU like the Intel Arc B580. This comes down to XeSS technologies being optimized and developed primarily for Intel architecture and AI hardware.

The Steam Hardware & Software Survey results for July 2025 are in, and when it comes to the GPUs, the story remains mostly the same: NVIDIA's GeForce graphics cards continue to dominate the desktop market. With this year seeing the arrival of the new GeForce RTX 50 Series from NVIDIA and the Radeon RX 9000 Series from AMD, the RTX 50 Series is quickly gaining momentum, while GPUs like the Radeon RX 9070 XT are still nowhere to be seen.

Looking at the new GeForce RTX 50 Series, the most popular model among PC gamers so far is the GeForce RTX 5070, followed by the GeForce RTX 5080 and RTX 5070 Ti. When you take away integrated graphics, the RTX 5070 has cracked the Top 20 most-used graphics cards among PC gamers. The GeForce RTX 5070 launched on March 4, 2025, so it's only taken the mid-range RTX Blackwell offering five months to achieve this milestone.

The top GPUs are still comprised of NVIDIA's various x60-class cards, with the GeForce RTX 3060 still holding the top spot, followed by the GeForce RTX 4060 and the RTX 4060 Ti and RTX 3060 Ti variants. NVIDIA's new GeForce RTX 5060 and RTX 5060 Ti are slowly climbing up the rankings, but adoption looks to be slower than the GeForce RTX 5070.

GeForce Game Ready Driver 580.88 - WHQL is here, and in addition to adding day one support for Mafia: The Old Country, which is out on August 8, it fixes several bugs alongside adding 62 new G-SYNC Compatible gaming displays. Mafia: The Old Country, from Hangar 13 and 2K, is the latest entry in the long-running crime-saga series, with the action taking place in 1900s Sicily.

The Unreal-powered title will require a decently specced rig to run, with the minimum GPU for 1080p with upscaling listed as the GeForce RTX 2070. And that's the minimum, with the developers recommending PC players jump in with a GeForce RTX 3080 Ti or equivalent graphics card. On the plus side, GeForce gamers have access to DLSS 4 with Multi Frame Generation, DLSS Frame Generation, DLSS Super Resolution, and NVIDIA Reflex to boost performance, fidelity, and responsiveness.

The latest driver also adds support for Clair Obscur: Expedition 33's DLSS 4 update, which brings the technology to one of the most talked about and celebrated games of the year. Outside of game support, GeForce Game Ready Driver 580.88 also adds a whopping 62 new G-SYNC Compatible displays, which is a seal of approval that you're going to get a fantastic baseline Variable Refresh Rate (VRR) experience.

NVIDIA has announced that it's ending GeForce Game Ready Driver support for its Maxwell, Pascal, and Volta architectures after a final driver release in October 2025. This includes popular GeForce GTX 10 Series graphics cards like the GTX 1060, 1070, and 1080. According to the latest Steam Hardware Survey results, around 6.5% of all PC gamers are still rocking a Pascal-powered GeForce GTX 10 Series GPU.

There are more GeForce GTX 10 Series GPUs than there are Radeon RX 7000 or RX 9000 Series cards. Although Game Ready Driver support is ending for these older NVIDIA GeForce architectures, which covers day one support and optimization for new games and fixes, NVIDIA will still be releasing security updates for Maxwell, Pascal, and Volta GPUs for the next three years.

"GeForce GPUs based on Maxwell, Pascal, and Volta architectures will transition to receiving quarterly security updates for the next three years (through October 2028)," NVIDIA writes. "Our support lifetime for these GPUs reaches up to 11 years, well beyond industry norms."

Australian YouTube creator TrashBench wanted to create a video showcasing that the eight-year-old GeForce GTX 1080 Ti still had some life left in it. The goal was to overclock the 1080 Ti (with a complete liquid cooling mod) to the point where it would outperform the current entry-level GeForce RTX 5050 GPU, which is built on the four-generation newer RTX Blackwell architecture.

Using the 3DMark Time Spy benchmark, tests were also conducted with Assassin's Creed Valhalla, Red Dead Redemption 2, Cyberpunk 2077, and Shadow of the Tomb Raider running in 1440p. The initial results indicate that the GeForce RTX 1080 Ti lags behind the RTX 5050 by approximately 5-15% in gaming performance. And this is without DLSS and RTX technologies like Frame Generation, which would further increase the 5050's lead.

It turns out that, after trying a few different GeForce RTX 1080 Ti GPUs, he was unable to see a significant performance difference, with only Shadow of the Tomb Raider running notably 6% faster. The initial plan didn't include overclocking the GeForce RTX 5050; however, once he applied an extreme OC, boosting the clock speed to 3.3 GHz, the RTX 5050 saw a massive 17% increase in performance.

With the popularity of white-colored components, cases, and cooling, it only makes sense to pair an all-white build with an all-white GPU. And with that, Palit is introducing new GeForce RTX 5060, RTX 5060 Ti, and RTX 5070 WHITE Series graphics cards, with all models listed as OC variants shipping with GPU Boost Clock speeds that are overclocked out of the box.

The line-up includes both GeForce RTX 5060 Ti 8GB and 16GB models, with Palit noting that the 8GB variant is "well-suited for iCafes" and competitive title performance where the VRAM limitation won't be a bottleneck. All models feature compact dual-fan designs, except for the Palit GeForce RTX 5070 WHITE OC, which has three fans with the same two-slot thickness.

The Palit GeForce RTX 5070 WHITE OC is the flagship model of the line-up and is positioned as a 1440p 120FPS GPU for AAA ray-traced gaming thanks to DLSS 4 and the arrival of Multi Frame Generation. The mainstream GeForce RTX 5060 WHITE OC 8GB model is described as an entry-level 1080p gaming GPU and an "ideal entry" point for next-gen PC gaming.

ASUS has been celebrating 30 years in the graphics card business, and will be unveiling a new ROG Astral GeForce RTX 5080 graphics card at Gamescom 2025 on August 19.

The company has been teasing new product launches and a bunch of giveaways ahead of the event, with ASUS preparing to unveil a new graphics card teased as a "30th Anniversary Edition" release. The folks at VideoCardz looked into the page's source code on the ASUS website and found some placeholder data, commented out, with references to the ROG Astral RTX 5080 Hatsune Miku, but that card was unveiled just over a week ago now.

We were expecting to see a new RTX 5090 design from ASUS, but it looks like the company will be unveiling a new RTX 5080 design as the AI TOPS performance numbers match the 5080, and that the Hatsune Miku variant is probably just a temporary placeholder.

Inno3D has launched its new GeForce RTX 5090 and RTX 5080 iCHILL Frostbite Pro graphics cards, featuring a sleek single-slot design and pre-installed water block. Check them out below:

Inno3D partnered with Alphacool, which developed the single-slot water block for the new cards, with the big difference here being that the new RTX 5090 and RTX 5080 iCHILL Frostbite Pro graphics cards sport a single-slot display bracket, making this an actual, true single-slot graphics card solution.

Inno3D explains: "Today we're launching the INNO3D GeForce RTX 5090 & 5080 iCHILL Frostbite Pro! Both GPUs are equipped with full water block coverage, with reinforced carbon fiber backplate to ensure maximum thermal efficiency with minimal footprint in the latest NVIDIA Blackwell architecture. Making them perfect for compact, high-performance setups".

AMD CEO Dr. Lisa Su attended the recent 'Winning the AI Race' summit, where she talked about how artificial intelligence and AI will be leveraged to design the company's next-generation graphics cards. With AMD moving to a unified architecture for both data center and consumer products, this presumably includes Radeon GPU hardware for the PC and console market.

However, it's worth noting that Lisa Su's remarks refer to GPU hardware and architecture designed with the help of AI, not GPUs designed solely by AI. This means that human engineers at AMD are still in charge. "We think about how AI will design future chips, and it will design pieces of it," Lisa Su said. "But there's still a creativity of bringing it all together that I think humans are still absolutely at the centre of."

"I don't necessarily see the AI designing our next-generation GPU," Lisa Su adds. "But I do see it helping us design the next-generation GPU much faster and more reliably." This isn't a new development, as NVIDIA has already stated that it uses AI to help speed up chip design, which appears to be what AMD is aiming for.

Over the past couple of months, we've seen specs and details emerge for NVIDIA's upcoming GeForce RTX 50 SUPER Series refresh. With an RTX 5070 SUPER, RTX 5070 Ti SUPER, and RTX 5080 SUPER on the way, the big takeaway is that there will be a massive 50% increase in VRAM capacity for all RTX 50 SUPER GPUs. Only the RTX 5070 SUPER is expected to receive a minor increase in CUDA Core count.

With VRAM capacity being a hot topic for both high-end PC gaming and AI workloads, a GeForce RTX 5070 SUPER with 18GB of fast GDDR7 memory, and a GeForce RTX 5070 Ti SUPER and RTX 5080 SUPER with 24GB sound pretty good. The increased memory will also be accompanied by an increase in power draw, indicating that the SUPER series will ship with higher boost clock speeds than their non-SUPER counterparts.

There has been some speculation and rumor that NVIDIA was planning on releasing them in early 2026 with a big CES 2026 reveal. However, based on information obtained from sources, the RTX 50 SUPER Series refresh is actually on track for a holiday 2025 or Q4 2025 release.

NVIDIA's new GeForce RTX 5090 D V2 graphics card launches in China on August 12, with reports that the new made-for-China variant will be $830 cheaper than the RTX 5090 D.

The $380 price reduction is a good thing to see, but gamers in China will be losing 12GB of VRAM as the RTX 5090 D V2 comes with 24GB of GDDR7 compared to the RTX 5090 D which packs the full 32GB of GDDR7 memory. The RTX 5090 D was already cut down from the RTX 5090, but the new RTX 5090 D V2 is cut down again.

When it comes to availability, it seems that the RTX 5090 D is mostly sold out and it is the last batch on Chinese shores. According to a post on Chiphell, "sthuasheng" writes: "now the flagship stores of various platforms are on sale, basically 19999 (yuan), the same as H20, the 5090D sold this time are the batch of goods produced before the new regulations came out, because the ban on sales was forced to become inventory after the new regulations came out".

GIGABYTE's custom GeForce RTX 5090 AORUS Master graphics card has had its BIOS flashed with the leaked XOC BIOS for the ASUS ROG Astral RTX 5090 D, pushing its power limit to 1600W and providing around 10% more performance over the stock BIOS.

YouTuber JayzTwoCents used the leaked XOC BIOS for the ASUS ROG Astral RTX 5090 D on his GIGABYTE RTX 5090 (non-D) AORUS Master, which on the ROG Astral RTX 5090 enables a huge 2001W power limit, but on the RTX 5090 non-D it's limited to 1600W. None of the custom RTX 5090 cards have dual 16-pin power connectors, which is why we won't see them pushed up to 2001W TDPs.

Once the XOC BIOS was flashed to the RTX 5090 AORUS Master, JayzTwoCents noticed that one of the fans on graphics card stopped spinning, which is because the RTX 50 series GPUs support up to three fan controllers, and the ROG Astral RTX 5090 has four fans but uses two controllers.

SK hynix has confirmed that it's cooking up higher-capacity GDDR7 memory modules that will roll out later this year, confirming it is increasing the maximum capacity of its GDDR7 from 16GB (2GB per module) to 24GB (3GB per module).

The South Korean memory giant commented on its larger 3GB GDDR7 memory modules for AI GPUs, but we know that we'll see 3GB GDDR7 memory modules arrive on desktop gaming GPUs like NVIDIA's upcoming GeForce RTX 50 SUPER series GPUs that have been in the rumor mill for months now.

We should see NVIDIA release a new GeForce RTX 5090 SUPER with a possible 48GB or even 64GB of GDDR7 memory, the RTX 5080 SUPER has been rumored with the full GB203 GPU and a larger 24GB of GDDR7 at 32Gbps and 420W+ of power. The RTX 5070 SUPER is rumored with 18GB of GDDR7 in the future.

A new investigation has been published into the smuggling of banned NVIDIA GPUs into China during the early stages of the Trump administration's tightening of export restrictions. The report states that at least $1 billion of high-powered NVIDIA AI hardware was smuggled to China, circumventing the newly implemented US trade restrictions.

The report comes from the Financial Times, which states that a billion dollars worth of NVIDIA GPUs were shipped to China during the three months before Trump eased the trade restrictions with the nation, with the report citing "dozens of sales contracts, company filings, and multiple people with direct knowledge of the deals." More specifically, Chinese distributors from the Guangdong, Zhejiang, and Anhui provinces are alleged to have sold various NVIDIA AI GPUs in large quantities, including the B200, H100, and H200.

These GPUs were sold and moved in racks that were approximately the size of a "large suitcase" and contained all of the necessary components to integrate them directly into an existing data center. Each of these "racks" was valued at approximately $489,000, which means the price was jacked up by at least 50% when compared to the average retail value in the US.

AMD's next-gen RDNA 5-based flagship Radeon RX 10900 XT graphics card will reportedly sport a huge 154 CUs, 36GB of GDDR7 memory with 1.7TB/sec memory bandwidth, and will have performance that will rival NVIDIA's next-gen GeForce RTX 6090 graphics card.

In his latest video, Tom from Moore's Law is Dead reports on his latest leaks that would see the Radeon RX 10900 XT (or whatever it is called, but the RX 10900 XT sounds fantastic). We are looking at the "AT0" die that will feature a huge 154 Compute Units of RDNA 5 GPU power, 40MB of L2 cache, a 380W TDP, and 36GB of GDDR7 memory clocked at 36Gbps on a 384-bit memory bus that would pump out 1.728TB/sec of memory bandwidth.

This would result in AMD's next-gen Radeon RX 10900 XT graphics card competing with NVIDIA's next-gen GeForce RTX 6090, meaning the company is aiming for its new RDNA 5 flagship GPU to be faster than the current RTX 5090, which is something all gamers want to hear. These are just estimates in performance so things can change obviously.

Not so long ago, GPUs made in China were something of a spectacle, boasting performance levels that meant they weren't taken seriously - but some serious progress has seemingly been made with the Lisuan G100.

As you may recall, just a month ago, the Lisuan G100's performance couldn't match the NVIDIA GTX 660 Ti, but now the board is apparently faster than the RTX 4060, and not too far off the RTX 5060.

That's based on a reported Geekbench OpenCL run (as flagged by Benchleaks, add your own salt) where the still-unreleased G100 scored 111,290 points (being run in conjunction with a Ryzen 5 7600 processor).

NVIDIA may be about to release a new spin on its GeForce RTX 3050 graphics card, going by the latest from the GPU grapevine.

This theoretical launch is the NVIDIA RTX 3050 A, and the hint that it's inbound is the fact that it has just been listed in GPU-Z, which has a new release out (version 2.67.0).

This isn't the first time we've heard about the RTX 3050 A, as rumors were circulating last year (when the GPU was spotted in the PCI ID repository). As Tom's Hardware, which picked up on this, observes, back at the time, NVIDIA confirmed the graphics card and said it was built with an AD106 chip rather than GA107.

NVIDIA is ramping to the launch of its new GeForce RTX 5090 D V2 graphics card in China, with the new flagship Blackwell gaming GPU hitting China on August 12.

In new reports from Expreview and Benchlife, we're hearing that the NVIDIA GeForce RTX 5090 D V2 is coming to China and that AIB partners have been notified about the launch date as August 12. The new RTX 5090 D V2 will replace the RTX 5090 D as the latest gaming GPU from NVIDIA that is compliant under US export restrictions to China, and the third gaming graphics card tweaked for after the RTX 4090 D.

NVIDIA's new tweaked-for-China RTX 5090 D V2 uses the GB202-240 die versus the GB202-250 die on the RTX 5090 D, and the GB202-300 die used on the standard RTX 5090. The new RTX 5090 D V2 uses a newer PCB design -- the PG145 SKU 40 -- with changes made to the VRAM and GPU layout.

The ASUS ROG Astral GeForce RTX 5090 D graphics card has a 600W TDP, but with the XOC BIOS, it pumps that up for professional overclockers and enthusiasts to an insane 2001W, which has been leaked.

The leaked XOC BIOS for the ROG Astral RTX 5090 D is available on TechPowerUp but remember it's not official or verified, but if you use it on the card you've got a huge 2001W power limit at your hands. The new XOC BIOS is compatible with the China-aimed ASUS ROG Astral RTX 5090 D Gaming OC, supplying 2001W of power through its single 16-pin power connector.

GALAX has its own high-end RTX 5090 D HOF graphics card which has its own leaked XOC BIOS that unlocks the same 2001W power limit, with the card and its 2100W of power flow breaking the world record on 3DMark Port Royal recently. GALAX's custom RTX 5090 D HOF card has a 2610MHz boost clock out of the box, but in the hands of overclocker Team OGS, pushed a huge 3.6G5Hz boost clock (nearly 1000MHz more than stock) under LN2 cooling.

NVIDIA's laptop version of the RTX 5090 has been a disappointment for some gamers - at least in some respects - but what if you could make it 20% faster, or more?

That's what an intrepid YouTuber, GizmoSlipTech, has managed to do, but it involves a complex and risky form of overclocking known as shunt modding.

This involves replacing a resistor on the graphics card (or adding one in parallel) to mess with the voltage controls, and essentially pipe through more power than would otherwise be allowed.