A short video has resurfaced in which Bill Gates promotes Windows 95 with DirectX as a gaming platform set to eclipse DOS. The video is part of footage from an internal Microsoft ‘Judgment Day’ developer/gaming event. So far, so ordinary, but the typically meek Gates appears dressed in a trench coat and holding a shotgun, standing inside a Doom-style environment littered with demon carcasses. You can watch the video by expanding the tweet below.

The Doom franchise began in 1993, well before Windows was widely used for games beyond the likes of Minesweeper and Solitaire desktop distractions. Throughout much of the 1990s, PC users would run games through DOS, with Windows seen as something to use for multitasking productivity tasks. With Windows 95, Microsoft wanted to change that; hence, this promotional video to developers, starring Gates himself.

Though there were (and are) DOS gaming devotees, Windows 95 was promoted as delivering improved multimedia, Plug and Play hardware support, and, with DirectX, even better gaming performance.

Gates begins his in-Doom spiel by saying that “these games are getting really realistic” before boasting that “Windows 95 is THE BEST game platform, whether it’s the best performance, the best setup, the best integration.” He then admitted that DOS can be hard work for end users, developers, and tech support. But the key improvement being dangled in front of developers was the new DirectX API.

During this brief presentation intro, Gates touted 75 new games coming for Windows 95 in the coming year. He reckoned with this push and collaboration between Microsoft and game developers, “we’ll be able to clean up this DOS mess, and get everything focused on Windows.”

This unusual video isn’t just Bill Gates standing and spouting about Windows 95 gaming, thankfully. Halfway through this clip, he is interrupted by an enemy grunt, and quickly turns to give it both barrels, “Don’t interrupt me.” But when delivering this punchline, Bill sounds more like Kermit the Frog than Arnie.

The video closes with Microsoft’s logo, some sinister laughing soundtrack, and the tagline “Who do you want to execute today?”

Key to the background collaboration between Microsoft and a host of game developers was the work of Gabe Newell. The future Valve co-founder initiated and led the Microsoft team responsible for porting id Software’s Doom to Windows 95, which would result in the release of Doom95 in 1996. Some sources say that Doom was installed on more computers worldwide than Windows 95 was at this time (late 1995).

The UK’s Imagination Technologies (IMG) is teasing its roadmap to “bring high‑performance, scalable, PowerVR graphics to desktop, workstation, and cloud environments.” We’ve seen and heard of a few false starts from IMG since the start of the 2020s, but this is the firm’s clearest messaging yet, with videos showcasing its work with Microsoft's DirectX, and stating that this work is “essential for mainstream PC gaming.”

IMG will be best known in PC circles, among folk of a certain vintage, for its pioneering PowerVR 3D graphics cards over three decades ago. It was also the graphical force behind the Sega Dreamcast (which launched in 1999). Since then IMG has spent focused on the mobile market, being a key Apple partner between 2007 and 2017.

In 2020 we noted IMG launching its B-Series GPU IP. That was a significant change of tack, after the disastrous mobile upset suddenly delivered by Apple. This IP would become the foundation for Chinese desktop GPU makers like Moore Threads and Innosilicon, despite their boasts of ‘domestic’ graphics tech.

In 2022, as it celebrated its 30^th anniversary, IMG would confirm that its tech was behind the aforementioned Chinese desktop cards. By this time there had been new generations from the Far East, with models like the 'double power' Fantasy 1 Type B card launching.

A year later Imagination’s IMG DXD, a DirectX‑capable GPU IP line explicitly aimed at desktop, laptop, and cloud gaming markets, were announced. Analysts at JPR thought that this marked IMG’s full hearted re-entry into the high-end PC market. DXD was its first real PC-ready architecture effort since Kryo in 2002.

Handling DirectX workloads with confidence

Today, IMG wanted to let graphics enthusiasts know it has reached a milestone in modern PC graphics support. From the few reviews and hands-on tests we’ve seen of Chinese PowerVR architecture cards, they have often been pretty poor at supporting PC gaming titles, especially newer games.

“We started adding hardware‑based support for DirectX to our GPUs with our last generation, D-Series,” IMG reminds viewers. “Below you can see it in action, in real silicon, handling complex DirectX 11 workloads with confidence.”

This is the same video referenced in the embedded LinkedIn post.

IMG indicates this demo evidences the beginning of its commitment to its long‑term roadmap embracing higher power devices like PCs. It asks that we stay tuned. Of course we will. But even if the demos get better and better, moving up through DX12, and adding other modern frills – it will probably be dismissed by the Crimson Desert devs…

In 2022, Nvidia introduced hardware-level Shader Execution Reordering (SER) with its RTX 40-series GPUs in order to make ray tracing less taxing. Now, it's officially part of DXR 1.2, which is included in the new DirectX Agility SDK (version 1.619). The announcement blog isn't a casual read because of all the technical jargon, so let's break down what this actually means and how it improves performance.

SER basically reduces the per-pixel rendering time in a very intense ray-traced or path-traced scene. Unpredictability is a GPU's worst nightmare, so when rays start to bounce off of surfaces in a very uncontrolled manner, sure, the scene looks good, but it's crippling the silicon. Here, SER slots in and dynamically categorizes all the reflections and light bounces to form cohesion.

It allows the GPU to find patterns across rays, grouping them together to enable better parallel execution. SER works hand-in-hand with Opacity Micromaps (OMMs), the other highlight feature included in DXR 1.2, which saves processing power by telling the GPU not to run a shader when hitting a transparent or translucent surface.

Your graphics card will only shade the visible pixels as the Opacity Micromaps will give it precise hints on what part of the scene needs to be opaque (and what doesn't). So, SER begins by grouping similar ray-traced shaders together, and then the OMMs let it skip the "invisible" ones entirely. Reducing unnecessary shader work simply allows you to maintain more FPS in games, especially in complex scenes.

In a branching blog, Microsoft shows its own demo for SER, where a scene is rendered with and without it. Using SER, Nvidia GPUs saw a 40% boost in performance while some Intel Arc B-series GPUs got up to 90% more FPS. This feature, now being standardized, means we can potentially see Intel and AMD implement their own hardware-level SER in next-gen GPUs.

The last noteworthy inclusion in this SDK update was Shader Model 6.9, which is what actually enables developers to interface with both OMMs and SER. This will make game developers very happy, but it's ultimately up to them to implement these features before a player-facing upgrade is ever seen. To be clear, these features were announced last year but just came out of preview today.

There are a lot more details in the blog that we didn't go over, such as support for Long Vector, 16-bit float operations, and general changes to streamline hardware overhead. Some of them target the poorly optimized games we see today, struggling with anything less than 12 GB of VRAM. It's all early, programmer-focused patchwork for now, but it can translate to real-world improvements soon.

It's difficult to imagine how games would look today without the quiet yet transformative change that Microsoft brought about in the year 2000. Twenty-five years ago, to this day, the company introduced DirectX 8. The release was accompanied by little fanfare, no generation-defining tech demo, but it carried one major breakthrough with it — programmable shaders — which would forever revolutionize the way GPUs render graphics.

Before DirectX 8, graphics cards worked on a fixed-function pipeline, meaning that almost everything was predefined, baked into the silicon itself. Lighting equations, texture blending, transformations; it was all at the mercy of the GPU's support. For instance, in lieu of real-time reflections, you'd have environment maps because the GPU itself wasn't able to calculate them dynamically. You were bound by the logic of the hardware, which wasn't very flexible.

Think of it as adjusting knobs on a console. You can tweak the parameters slightly, but what if you wanted to change the very knobs themselves? Enter, DirectX 8.

DirectX 8's programmable shaders

Microsoft added Shader Model 1.0 to DirectX, which came with Vertex Shader 1.0 that would allow devs to manipulate each vertex, and Pixel Shader 1.0 that would allow devs to control the final color of each pixel. Previously, none of this was really accessible, but DirectX 8 gave control to the people making the games, enabling them to write code to program the GPU to render in a certain way.

Devs could define the very math behind lighting, for example, or control material behavior, specular highlights, (basic) tessellation, and so on for the first time ever. It was no longer about just accepting what the GPU provided by default, but rather what the GPU could do, full stop. It turned the silicon from a locked-down machine to an actual, programmable processor, true to its name.

Piggybacking off the possibilities of DirectX 8, games like Half-Life 2 would arrive with groundbreaking graphics for the era. They were built on the ideas of authorship, giving autonomy (even if minor at the time) to devs to build real-time shadows, refraction and water shaders, post-processing effects and so much more. All because they could now write custom code, telling the GPU how to calculate something.

It was up to the creatives to shape how light interacted with each object, something that may seem primitive by today's standards. Even in 2025, some things are pre-baked, like global illumination, for efficiency's sake, while others operate in real-time. And it's shaders that decide how both interact with each other, merging them seamlessly to give you "better graphics."

DirectX 8's release coincided with the launch of Nvidia's GeForce 3 , which added hardware blocks, such as shader-execution units, to run DX8's Shader Model. In our original coverage of the GPUs from over two decades ago, we listed the inclusion of a Pixel Shader and Vertex Shader as the two biggest improvements, carrying endless possibilities — truer than ever in hindsight.

A year later, ATI (AMD) caught up with more powerful DX9-class hardware, building upon the legacy of DX8. Even the original Xbox shipped with a DirectX 8 GPU at the time, and there was a lot of excitement around true per-pixel lighting among the developer community. Games like Morrowind and Splinter Cell were early adopters of this technology, eventually made common by Unreal Engine 2 later on.

All in all, DirectX 8 wasn't an explosive update, but it completely altered the way modern rendering works. Every single device we use right now, from phones to consoles to computers, they wouldn't work the way they do without employing DX8's fundamental principle of control. Letting the people, who make the software, program the hardware however they want is what led subsequent graphics breakthroughs we consider normal today.

Microsoft has announced a new feature that will reduce initial load times called advanced shader delivery. According to the DirectX Developer Blog, this will move shader compilation from the device to the cloud, allowing you to download the necessary files directly to your device for caching. Aside from allowing you to get into your game much faster, with the company reporting an observed 85% reduction in loading times for Avowed, it also claims to save your battery and processing power, as Microsoft handles the heavy lifting on its end.

Microsoft announced this week its DirectX Raytracing (DXR) 1.2 application programming interface, which introduces new features that significantly improve visual quality and rendering performance by up to 2.3 times. AMD, Intel, Nvidia, and Qualcomm, along with game developers like Remedy, are working to integrate DXR 1.2 technologies into future gaming hardware and software.

The DirectX Raytracing 1.2 update includes Opacity Micromaps (OMM) and Shader Execution Reordering (SER), two technologies that boost performance in raytraced games by 2 (SER) to 2.3 times (OMM). Both technologies must be implemented into actual games or game engines to experience the performance benefits.

2X – 2.3X performance boost

One of the main issues with alpha-tested geometry (foliage, fences, hair, etc.) in raytracing is extra calculations required for light to determine whether it hits a surface or passes through. Opacity Micromaps (OMM) improve how alpha-tested geometry processes by applying a texture with an alpha channel to a flat surface. It then removes pixels below a certain transparency threshold. OMM reduces the number of times shaders need to be used, leading to higher efficiency and performance.

In the best-case scenario, Microsoft claims an improvement of 2.3 times. However, keep in mind that not all games and scenes contain a lot of elements like foliage and fences. For example, while S.T.A.L.K.E.R. 2 has loads of grass, leaves, and fences in practically all scenes, Cyberpunk 2077 barely has any foliage.

Shader Execution Reordering (SER) seems to be a more universal feature as it reorders how shaders are executed to avoid shader divergence. Shader divergence occurs when nearby pixels require shaders to do different tasks, a common situation in scenes with heavy raytracing effects, such as complex lighting, realistic shadows, and detailed reflections.

GPUs process shaders in parallel threads organized into groups called warps or wavefronts. Ideally, all threads within a group execute identical instructions simultaneously, maximizing GPU efficiency. Shader divergence occurs when threads in the same warp or wavefront need to perform different instructions. In this case, simultaneous execution is impossible, forcing the GPU to handle each instruction path separately, leaving some threads idle and increasing latency.

According to Microsoft, SER sorts or batches similar shader workloads together, reducing divergence and maximizing GPU utilization and speeding up rendering by up to two times.

Hardware support

Regarding hardware support, the situation is a mixed bag, which is common with new API features.

All Nvidia GPUs dating back to Turing (GeForce RTX 20-series) support Opacity Micromaps (OMM), so these graphics cards can potentially experience a performance boost once game developers implement them into their titles. Intel said its next-generation Celestial (Xe3) GPUs will also support OMM.

Nvidia's GPUs have supported Shader Execution Reordering (SER), starting with the GeForce RTX 40-series Ada Lovelace family. Intel said it looks forward to supporting SER "when it is available in a future Agility SDK.' However, whether it will be supported on Intel's Arc 'Alchemist' or Arc 'Battlemage' GPUs (or both) is unclear.

AMD does not seem to support OMM or SER on its RDNA 2/3/4 GPUs, though Microsoft said that the red company is working with it on the widespread adoption of these technologies. Also, AMD has certain scheduling optimizations that may mimic how SER works, so if game developers take time to optimize for Radeon GPUs, the latter may get some speed improvements.

Qualcomm also does not support OMM or SER, but it said it would on its next-generation integrated GPUs.

The preview version of DXR 1.2 will launch in April 2025.

Nvidia and Microsoft announced on Thursday that they would be adding neural shading support to the Microsoft DirectX preview this April. Neural shading will use cooperative vectors and Nvidia's Tensor cores (matrix operations units) to speed up graphics rendering in games that support the technology. It will better allow for the generic use, via HLSL (high level shading language) of traditional rendering techniques alongside AI enhancements.

While real-time computer graphics and graphics processing units (GPUs) have come a long way, the graphics rendering pipeline itself has evolved slower than hardware. In particular, while Nvidia's GPUs have featured Tensor cores (primarily aimed at AI compute) for over seven years now, they have only been used so far for things like upscaling (Nvidia's DLSS), ray reconstruction (DLSS 3.5) and denoising, and frame generation (at least for DLSS 4).

This is going to change with the so-called neural rendering — a broad term that describes a real-time graphics rendering pipeline enhanced with new methods and capabilities enabled by AI.

A specific subset of neural rendering focused on enhancing the shading process in graphics is called neural shading. Its main purpose is to improve the appearance of materials, lighting, shadows, and textures by integrating AI into the shading stage of the graphics pipeline. The addition of cooperative vectors — which let small neural networks run in different shader stages, like within a pixel shader, without monopolizing the GPU — is a key enabler for neural shading.

Cooperative vectors rely on matrix-vector multiplication, so they need specialized hardware, such as Nvidia's Tensor cores, to operate. To that end, they can work on Intel's XMX hardware as well as Nvidia's tensor cores. Intel also released a statement saying cooperative vector support will be provided on Arc A- and B-series dedicated GPUs as well as the built-in Arc GPUs found in Core Ultra Processors (Series 2) — basically, every GPU from Intel that includes XMX support.

It seems as though cooperative vectors may also work on AMD's RDNA 4 AI accelerators, though RDNA 3 seems more doubtful (as it lacks AI compute throughput compared to the competition). Still, Microsoft is working with AMD, Intel, Nvidia, and Qualcomm to ensure cross-vendor support for cooperative vectors over time. 

"Microsoft is adding cooperative vector support to DirectX and HLSL, starting with a preview this April," said Shawn Hargreaves, Direct3D development manager at Microsoft. "This will advance the future of graphics programming by enabling neural rendering across the gaming industry. Unlocking Tensor Cores on Nvidia RTX GPUs will allow developers to fully leverage RTX Neural Shaders for richer, more immersive experiences on Windows."

Nvidia's phrasing makes it sound as though the upcoming DirectX preview with cooperative vectors is as an Nvidia exclusive. However, Intel will co-present a session Cooperative Vectors with Microsoft, so the only unknown right now appears to be AMD GPU support. If driver support is available from AMD, it should work on its GPUs as well.

Ultimately, we'll need to wait to find out not only whether it works, but how well it works — both in terms of image fidelity as well as performance. Differing levels of compute among GPUs will likely affect the end user experience.

Microsoft is advancing its DirectX API to support neural rendering, signaling a transformative shift in graphics rendering by incorporating AI and machine learning. This development, highlighted in a recent blog post, is designed to enhance visual quality and efficiency in gaming and other graphics-intensive applications.

Neural rendering makes use of machine learning models to generate or enhance visual elements such as textures, lighting, and image upscaling. By offloading complex rendering tasks to AI, this approach improves both performance and visual fidelity while reducing the computational burden on traditional rendering pipelines. Technologies like Nvidia’s DLSS and AMD’s FSR have already demonstrated the potential of AI-enhanced rendering. Microsoft’s initiative seeks to provide a standardized, open framework for such capabilities within the widely used DirectX API.

A key feature of the forthcoming DirectX update is Cooperative Vector Support. This feature will enhance AI workloads for real-time rendering by optimizing matrix-vector operations crucial for AI tasks like training, fine-tuning, and inferencing. This feature allows AI tasks to run in different shader stages, enabling efficient execution of neural networks, such as in a pixel shader, without monopolizing the GPU. By integrating neural graphics into DirectX applications, it provides access to AI-accelerator hardware across platforms, empowering developers to create more immersive experiences.

Microsoft has confirmed that Cooperative vectors will leverage Tensor Cores in Nvidia's new RTX 50-series GPUs to enable neural shaders, enhancing game asset visualization, optimizing geometry for improved path tracing, and supporting tools for creating photorealistic game characters.

Microsoft’s High-Level Shading Language (HLSL) team is said to be working closely with major GPU manufacturers, including AMD, Intel, Nvidia, and Qualcomm, to ensure these new capabilities are optimized for a wide range of hardware architectures.

By embedding neural rendering capabilities into DirectX, Microsoft could broaden the adoption of AI-driven graphics across multiple platforms. Potential applications range from enhanced real-time ray tracing to adaptive resolution scaling for high-definition displays. While proprietary AI rendering technologies have been limited to specific ecosystems, Microsoft’s open approach could democratize access, fostering greater innovation and competition.

Though the updates are still in development and lack a definitive release date, they highlight the increasing role of AI in shaping the future of graphics technology.

As of earlier today, a recent Vulkan spec update— version 1.3.291, to be specific— went live on GitHub, and besides miscellaneous fixes and the such, a new "VK_AMD_anti_lag" extension has been added to Vulkan, bringing full AMD Anti-Lag support outside the boundaries of the recent DirectX graphics APIs. Nvidia Reflex is also already supported with Vulkan in select titles, though some users have reported issues with that functionality under Vulkan. This update for adding Anti-Lag to Vulkan also addresses an existing "VK_NV_low_latency2" extension, as well.

For those unfamiliar with AMD Anti-Lag or Nvidia Reflex, the best way to think of these options when they are available in your game is "automated FPS capping." Traditionally, users may set FPS caps in games with unlocked FPS to stabilize visuals, performance, or GPU usage at a certain level without buffering frames (increasing latency) using V-Sync. In-engine FPS caps are almost always the best way to reduce input lag in a game, so long as that capped FPS target can be consistently met.

However, there are many caveats to FPS capping—namely, it's a very manual process, and it usually can't be easily adjusted during gameplay. For many gamers, doing it at all may be too difficult—or they're worried about maximum FPS losses. However, it's worth noting that a slightly lower, more consistent FPS target will nearly always "feel" smoother than a higher average with regular dips.

So, the greatest benefit of AMD Anti-Lag and Nvidia Reflex technologies is that they entirely automate the FPS capping process according to end-user latency and GPU utilization. Since fully maximizing GPU utilization can increase latency and the severity of FPS dips, both technologies tend to cap GPU utilization around ~95% while dynamically adjusting the framerate cap to ensure the smoothest image and lowest latency.

As a side note, the Steam Deck's built-in Refresh Rate/FPS Limiter toggle is one of the few ways to dynamically adjust an FPS cap without restarting the game, using external applications, relying on Anti-Lag, etc.

The most exciting implication of Vulkan adding support for AMD Anti-Lag is that this may lead to Anti-Lag being natively supported on Linux. As-is, there are already some efforts from Nvidia to get Reflex working on Linux through Proton extensions, though users have given mixed reports on how well this works. There's also an open-source, vendor-neutral LatencyFleX solution, but since it works akin to the canned AMD Anti-Lag+ (game engine hooking), it can trigger bans in games with anti-cheat, so it's far from perfect.

For now, only time will tell if these Vulkan updates lead to native AMD Anti-Lag or AMD Anti-Lag 2 support being added to native Linux games or at least AMD Linux systems running the Proton compatibility later, like Steam Deck. Since AMD FSR 3 Frame Generation can be made to work through Proton and there are already efforts to get Reflex working, we're cautiously optimistic.

While AMD formally introduced its FidelityFX Super Resolution 3 technology almost a year ago, so far, no actual games have gained support for this version of the frame generation upscaling technology. But the wait seems to be over as the company said that the first two titles to support FSR3 would show up today! The new technology from the FidelityFX package promises to bring substantial improvements to AMD's Radeon RX 6000 and RX 7000-series boards, which are among the best graphics cards money can buy today.

"We are working with our partners to bring you the first two FSR3-enabled games tomorrow," Frank Azor, Chief Architect of Gaming Solutions and Marketing at AMD wrote in an X post last night. "More titles will begin ramping-up thereafter, as we saw with prior versions of FSR, which are now supported in over 300 announced games." 

The first two games to support FSR3 are Forspoken by Ascendant Studios and Immortals of Aveum by Electronic Arts.

Frame Generation

The primary improvement of AMD's FSR 3 compared to FSR 2 is the introduction of frame generation capability. This feature leverages an advanced form of AMD's Fluid Motion Frames (AFMF) optical flow interpolation technology and temporal game data such as motion vectors to insert one extra frame between existing ones. AMD says that by generating that extra frame, it can double performance in supported games. 

When paired with Super Resolution upscaling, AMD FSR 3's frame generation enables previously unachievable framerates in supported games. For example, AMD claims that using FSR3 with frame generation and performance super resolution upscaling can increase performance in Forspoken from 53 FPS to 175 FPS. This boost in performance can be especially advantageous in challenging scenarios, like playing games at 4K resolution with the highest ray tracing configurations or when the CPU performance restricts framerates. 

It is crucial to mention that for the best and most responsive experience, AMD advises maintaining at least 60 FPS before applying the frame generation feature.

Latency Reduction

Image generation by definition increases latencies, so FSR 3 also includes its own latency reduction technology for smoother gameplay with frame generation enabled (it is still recommended to maintain 60+ FPS before applying frame generation). This latency reduction method promises to work on all graphics cards albeit in supported games. 

Gamers with AMD Radeon RX 7000-series graphics cards can also use the upcoming AMD Radeon Anti-Lag+ technology to decrease latency further in games like Forspoken, but those with Radeon RX 6000 or earlier GPUs can opt for Anti-Lag. 

Native AA for Maximum Quality

Sophisticated antialiasing methods are very taxing on GPU performance and while techniques like FSR, DLSS, and XeSS are designed to wed quality and performance, their main focus is the latter. Meanwhile, there are games that already run great and can use some additional eye candy, which is AMD set to offer with its Native AA mode. 

Native AA allows users to benefit from AMD FSR 3's sharpening enhancements without the upscaling. So, if a game's built-in anti-aliasing is not up to par, the 'Native AA' setting can enhance the visual quality. 

While there are performance trade-offs with this mode, pairing it with AMD FSR 3's frame generation can lead to significant performance gains, as illustrated by AMD in the graph. In essence, the 'Native AA' setting combined with frame generation promises to deliver both enhanced quality and performance. 

Intel Arc Alchemist GPUs landed about a year ago, and now compete among the best graphics cards. Drivers have been a recurring theme with Arc, both good and bad, but Intel has made strides to improve things and close the performance gap. To that end, Intel has released new drivers specifically aiming to improve its DirectX 11 performance.

We spoke with Intel earlier this week, and the gains for now are limited to specific games. To provide some background detail, after the initial Arc launch, Intel found that many of the base driver elements for their GPUs — integrated and discrete — were suboptimal, particularly for Arc. With integrated graphics, you can basically always assume that the GPU is the bottleneck. Anything that could be done to move work back to the CPU to alleviate that bottleneck was a "good thing." Obviously, that same logic doesn't necessarily apply any longer when the GPU becomes ten times faster, as in the case of the Arc A770.

So, since the initial launch, Intel spent a lot of time and effort reworking the drivers. We saw this first in early 2023 when Intel touted gains in DirectX 9 performance. The initial drivers for Arc leveraged Microsoft's DX9 on DX12 tools, which provided compatibility and got things up and running. There was a lot of performance left on the table, however. We're told Intel's driver team worked with the open source DXVK project (DirectX on Vulkan) to optimize the various elements better — something it couldn't do with the closed-source Microsoft tool. Eventually, things reached the state that all the DX9 support (or most of it?) was transitioned to using the new DXVK-optimized path.

We're now getting a similar treatment for DirectX 11 games — not in the DXVK sense, but Intel has worked to build a from the ground-up different code path for DX11 support into its drivers. Since the existing DX11 support already works, even if performance isn't ideal, the focus has been on testing specific games and then "whitelisting" them in the drivers to use the new code path. At present, Intel has whitelisted ten popular DX11 games: Apex Legends, Counter-Strike 2, Destiny 2, DOTA 2, Genshin Impact, GTA Online / GTA V, League of Legends, Middle-Earth: Shadow of War, Overwatch 2, and Valorant.

In short, the latest drivers optimize how the graphics hardware interacts with the DX11 API. By fine-tuning various parameters and implementing more efficient rendering techniques (it didn't go into any low-level details), Intel improved overall performance on DirectX 11 games. Here are the results of its internal performance testing, comparing its latest drivers (internal versions 4571 and 4642, though version 4644 drivers are now out) with the original launch drivers (3490).

Across the ten games (eleven if you want to count GTA Online and GTA V as separate entries), the new DX11 code path boosted performance by anywhere from 5% to 33%. But what's particularly interesting is that Intel specifically targeted the driver optimizations at "mainstream" CPUs — it tested with a Core i5-13400F and an Arc A750.

Later in the slides, Intel notes that the gains with a top-end Core i9-13900K weren't as pronounced. The performance increased by only 0% to 28% across the selected games, with an overall average improvement of 12%, compared to the 19% seen with the Core i5 CPU.

We applaud Intel's efforts and testing, as this makes a lot of sense for potential buyers of the Arc A750. With Arc A750 cards now starting at $199, they're a great bargain, providing performance roughly equivalent to the RTX 3060 and RX 6700 10GB. Both of those still tend to sell in the $270 range (while supplies last on the 3060), meaning the A750 provides great bang for the buck. At the same time, while we might standardize all of our testing on a Core i9-13900K CPU to eliminate other bottlenecks as much as possible, people building a PC with an A750 aren't likely to go much above $200 for the CPU, and that's where the i5-13400F sits.

Intel had some other interesting news and bullet points to discuss. The major one is that PresentMon — the foundation of Nvidia FrameView, AMD OCAT, and CapFrameX — is getting an update. PresentMon was created by Intel as an open-source performance monitoring tool to capture frametimes and other metrics related to graphics performance. It has seen many updates over the years, and now Intel is joining AMD and Nvidia in offering a more robust solution.

The new PresentMon beta adds a GUI, making it far more user-friendly than the existing version. Intel has also added a bunch of new features, including a robust overlay, histograms, and more. There's also a new "GPU Busy" metric, which shows how much time was spent waiting on the GPU versus waiting on the rest of the system. Minimizing this difference has been a key focus of the new drivers.

Intel continues to work on validating additional games with the new DX11 code path, and they'll roll those into future drivers. Perhaps at some point, the new code will be deemed robust enough that Intel will simply flip a switch, and all games will default to it — and maybe instead have a blacklist where games known to have issues can stick to the older code.

The new drivers and the PresentMon Beta are available for anyone to try. In fact, the DX11 performance improvements discussed here have been present in the past three Intel WHQL driver releases. The full slide deck for Intel's Q3'23 quarterly driver update is below for your reference.

Right now at Newegg, you can find the ASRock Challenger Arc A750 graphics card for its lowest price yet. It has a recommended price of $199 but using promo code VGAEXCAA784 at checkout will take the price down to $179. It’s not clear how long the code will be active nor how long the discount to $199 will be made available.

We've been excited to see this graphics card since it was first announced at the Tokyo Game Show late last year. Since its launch, the price has steadily lowered to the notably low price we have today.

The ASRock Challenger Arc A750 graphics card has an ATX form factor, spanning 271mm across and taking up two slots on the rear I/O panel. It has a core clock speed of 2200 MHz and a game clock speed of 2050 MHz. This edition comes with 8GB of GDDR6.

According to ASRock, this card uses DirectX 12 Ultimate as well as OpenGL 4.6. It’s compatible with PCIe 4.0 x16 interfaces and provides three video outputs to take advantage of. The ASRock Challenger Arc A750 has two DisplayPort outputs and one HDMI 2.1 port.

To get a closer look at this deal, visit the ASRock Challenger Arc A750 8GB GPU product page at Newegg for more details and purchase options.

A community-based project recently went live, providing access to the long-lost Windows Update pages, which work with classic versions of Windows to enable updating the OS. The Windows Update Restored site enables updating some classic versions of Windows, and the team members are busy getting their update locations ready. The team eventually hopes to provide service to those installing and updating systems with Windows 95, NT 4.0, 98, Me, 2000, and XP. The site's compatibility pages also have Windows Server 2003, plus Windows "Vista and Newer" entries, indicating they could eventually be part of the roadmap. 

Right now, the update site that is working and available is limited to a clone of the Windows Update v3.1 website (1997), which covers Windows 95, NT 4.0, and Windows 98 (and SE). The folks behind the project flag to site visitors that isn't an official Microsoft-backed project, and neither are the update pages it links to. Use these pages at your own risk, they warn, and they also stress that the update pages are for "archival purposes only." However, the below video shows that this site is functional for updating an OS.

The above video demo walkthrough shows likely scenarios that ancient Windows version users will encounter. At the start of the video, updating Windows 95 looks extremely quick and easy. After firing up the default Internet Explorer (works in v5 and v5.5), the user hits the 'Product Updates' button on the left column of the site. A 'Security Warning' pops up, where you must accept the authenticity of the Windows Update Control Package. Then a 'Please Wait' prompt appears to be replaced by a page full of checkboxes allowing you to select the updates that have been detected to be useful for your Windows system.

Watching this first Windows 95 update example unfold, it is startling to see download sizes so small relative to what we're used to today. The critical updates package offered to this OS was just 3.1 MB, a security update weighed in at just 124 KB, and the biggest download we saw was DirectX 8a (recommended, optional) at 11.2 MB. Multiple update page visits and PC restarts might be needed to complete the updating process.

Elsewhere in the video, you will see similar demonstrations with other aged OSes. Specifically, the video host also uses the Windows Update Restored v3.1 site to update Windows NT4 to Service Pack 6a and Windows 98 SE with many critical and optional packages. All critical updates for these OSes should be available now, but some optional updates aren't linked at the time of writing. For example, DX8.1 currently won't be downloaded and installed in Windows 95, even if you check the box on the update page.

In the above table, you can see the web addresses you need to point Internet Explorer at, in the respective OSes, to access the cloned Windows Update pages. For clarity, we reiterate that Windows 95, NT 4.0, and Windows 98 (and SE) are updatable at the v3.windowsupdaterestored.com address.

Windows Update pages which work their similar updating magic on Windows XP, Windows Server 2003, and support for Windows Vista or Newer are still being prepared. No timescale has been given for these pages appearing, so you will have to wait and see or check on the project's MSFN Forum posts. Windows Me and Windows 2000 are also not yet covered by the update service, but the video mentions a "near future" launch target for support for these OSes.

Windows Update Restored is both a welcome and interesting project, but it isn't officially recommended by the team for maintaining the use of older versions of Windows. We agree and would like to steer PC users to more secure modern OSes like Windows 10 and 11 (and the latest versions of Linux) for your browsing, connected tasks, gaming, and productivity.

Nvidia's GeForce X060-tier cards are usually among its most popular SKUs, so we are on tenterhooks for the launch of the RTX 4060, which poses to be one of the best graphics cards. Today, leak-centric video cards site VideoCardz says it managed to obtain some 3DMark scores from reviews in progress, being prepared for "the official embargo lift on June 28th." It has shared extensive comparative scores featuring several near-neighbor SKUs, purportedly taken from 3DMark Speed Way, Port Royal, Time Spy, and Fire Strike synthetic tests.

Before sharing and discussing some of the scores, it is essential to point out that our headline highlights the apparent 23% gen-on-gen uplift delivered by the upcoming RTX 4060 8GB against the established RTX 3060 12GB. The relatively recent addition of the RTX 3060 8GB is left much further behind by the new Ada Lovelace architecture entrant. According to the figures gathered by VideoCardz, the new RTX 4060 is reportedly 47.9% faster than the RTX 3060 8GB version. The source also highlights that both 8GB cards mentioned have a 128-bit memory bus and PCIe Express 4.0 8x interface making them likely unsuitable for modern 1440p (2560x1440) gaming.

We have reproduced an abridged set of results from VideoCardz sources. Remember to add a pinch of salt to these numbers, but also that they are synthetics. UL designs its 3DMark benchmarks to resemble games people play across generations and APIs. Still, with the variety of 3D engines, the correlation between these scores and gameplay FPS isn't always powerful.

Table data via VideoCardz

We would have liked to see an RTX 3070 and RX 6700 comparing the results. For bang-for-buck reasons, PC enthusiasts and DIYers might look sideways at old-stock and used GPUs instead of products like the RTX 4060 / Ti 8GB. Our upcoming review will be much broader than these results and feature many of the most popular games and display resolutions in 2023.

Remember that Nvidia is also preparing GeForce RTX 4060 Ti models with 16 GB of VRAM on board starting in July. Models with doubled VRAM will still be hindered by the 128-bit memory bus, so we will have to try and see if and when they are worth considering.

Japanese tech site PC Watch has managed to get its hands on a Moore Threads MTT S80 graphics card. This card uses a GPU that, despite some obfuscation deployed by Moore Threads management, appears to use the Imagination Technologies PowerVR architecture. This isn't going to make the list of the best graphics cards, not even if we roll back the clock to 2015, but it's good to start to see cards like this tested outside of China in benchmarks and games we're familiar with.

Though we're pretty sure that the PowerVR architecture is behind the card, PC Watch reported from Moore Threads verbatim regarding the specs of the MTT S80. So, the card uses the Chunxaio architecture, employing 4096 MUSA cores. Other specs include the GPU's clock speed of 1.8 GHz, and its peak performance of 14.2 TFLOPS. There's a generous 16GB of GDDR6 14 Gbps memory onboard the sample tested by PC Watch, and that connects to the GPU via a 256-bit bus for 448 GB/s bandwidth.

The MTT S80 is relatively power hungry, with a TGP (total graphics power) of 255W. That's probably why it includes a triple fan design. Also interesting is that the card uses a PCI Express Gen5 x16 connector. These raw specs don't tell the full story, of course, and driver support could still be a major factor in performance. But let's see what PC Watch found in its testing.

Before we look at the benchmarks and gaming tests, please note that PC Watch found there were lots of current games that wouldn't run on the MTT S80 - even using a supported motherboard, OS and CPU. DX12 and Vulkan games were insurmountable hurdles for this card, but some DX11 titles could run with varying degrees of success. Modern benchmarks faced a similar issue, with the most current version of 3DMark stable and usable being 3DMark 06.

Above you can see that the MTT S80 fares very badly when put up against even modest competition like Nvidia's GeForce GTX 1050 Ti, a budget GPU that debuted in 2016. On paper, the GTX 1050 Ti is woefully outmatched — it offers up 2.1 teraflops of compute, has 4GB of GDDR5 memory, and 112 GB/s of bandwidth with a 75W TGP. Even with such weak competition, the Moore Threads graphics card falls flat.

The MTT S80 fared best in DX9 graphics comparisons. It was still significantly behind the GTX 1050 Ti but not completely trounced. Or... well, it was completely trounced, but the average performance lead in the DX9 tests was 'only' 86% — so not quite double the performance.

Things were much worse for the MTT S80 when PC Watch looked at a selection of DX11 games. Besides some rendering errors, performance was very poor and some games effectively failed to work at all with single digit fps. Even if we discount Asetto Corsa where the Nvidia GPU was 90 times faster, the average lead in the DX11 games was still 188%, nearly triple the performance. One percent lows were also frightful on the MTT S80.

Poor performance and compatibility isn't the end of the sorry tale of the Moore Threads graphics card, unfortunately. As it stands, the card sucks up a lot of watts for very little. The MTT S80 on average consumed 142W, while the GTX 1050 Ti averaged just 60W.

In terms of performance per watt, even discounting Asetto Corsa (again), the MTT S80 managed just 0.33 fps/W while the GTX 1050 Ti averaged 1.86 fps/W. That makes the old Pascal GPU over five times as efficient.

There's a clear disconnect between the raw specs of the MTT S80 and its real-world results. On paper, the MTT S80 has four times as much memory, four times the memory bandwidth, and nearly seven times the raw FP32 compute. It's nowhere near reaching that theoretical level of performance.

PC Watch seems to think that the Moore Threads graphics card's major issue is with drivers, so it has some hope that things will continue to improve over the coming months. For now, the MTT S80 is not for gamers, curious developers, or graphics card collectors.

Back in February we reported on Korean TechTuber BullsLab Jay's video featuring the same MTT S80 graphics card. At that time, gaming tests were restricted to DX9 titles due to platform / driver immaturity. The fact that the MTT S80 can now at least try to run some DX11 games shows progress with the drivers, but there's still a long way to go. DirectX 12 and Vulkan games are also not currently supported.

Benchmarks of the Apple M2 Ultra's GPU appear to have leaked online. Today, we spotted various social media posts sharing Geekbench GPU compute scores and GFXBench Aztec Ruins scores. On the surface, these results seem to show that the GPU in the new M2 Ultra is highly formidable, offering GPU compute performance somewhere between a GeForce RTX 4070 Ti and RTX 4080. But we need to step back a bit and explain precisely what it is we're looking at.

We reported upon the new Apple M2 Ultra SoC reveal during the firm's annual WWDC earlier in the week. The chip will arrive in updated Mac Studio and Mac Pro designs, which are scheduled to ship next week. However, someone appears to be testing these new Mac computers/processors before consumers can get hold of them, as a raft of benchmarks has appeared online recently.

Yesterday, we picked through a selection of Apple M2 Ultra benchmarks, focusing on the SoC's CPU core performance. Today, we will look more closely at the potential compute capabilities of the M2 Ultra's GPU, thanks to the Geekbench and GFXbench leaks mentioned in the intro.

M2 Ultra Geekbench 6 Compute Benchmarks

It looks like several pre-release M2 Ultra Apple Mac system users have run Geekbench 6's Metal and OpenCL GPU benchmarks. Apple's Metal API is a proprietary graphics API developed by the firm for fast 'direct to the metal' hardware addressing, like an Apple optimized parallel to Microsoft's DirectX.

While these are very interesting synthetic compute benchmarks, we look forward to some thorough real-world testing of the new Apple Macs with M2 Ultra processors, shortly, particularly in gaming, which has been a pain point for Apple's chips.

Note that GPU compute tends to scale far better with multi-chip approaches than GPU graphics — think about Ethereum mining back in the day, where you could connect eight (or more) GPUs to a single modest CPU via PCIe x1 connections and they would all mine at basically 100% of their maximum performance. Not all compute workloads scale that well, but it's still very different than the scaling traditionally seen with real-time graphics used in games.

The new M2 Ultra chip's GPU, featuring up to 76 integrated graphics cores, looks impressive compared to the previous gen M1 Ultra with its maximum 64 graphics cores. In a direct comparison using Geekbench 6, the new-gen M2 Ultra scores about 220,000 or thereabouts, compared to about 155,000 on average for the M1 Ultra.

By this metric, M2 Ultra is over 40% faster than its predecessor in GPU compute tasks. It only has around 10% more graphics cores, so Apple has done well to achieve this scale of uplift.

Turning to a different comparison, the new Apple M2 Ultra's 220,000 Geekbench 6 Compute scores (Metal) sit between the GeForce RTX 4070 Ti (208,340 OpenCL) and RTX 4080 (245,706 OpenCL). For a direct Geekbench 6 OpenCL comparison, the Apple M2 Ultra Open CL scores of about 155,000 are much closer to PC GPUs like the Nvidia RTX A5000 and AMD Radeon RX 6800 XT.

Again, we need to note that Geekbench 6 OpenCL compute performance does not usually translate well to actual gaming workloads. Synthetic Geekbench 6 performance is very different from real-world graphics.

M2 Ultra GFXBench Benchmarks

Another user has put their M2 Ultra-powered Apple Mac through the GFXBench test suite. In the 4K Aztec Ruins offscreen tests, the new Apple SoC is over 55% faster than its predecessor. This seems to be an even better result than the Geekbench 6 GPU compute comparison. Theoretically, it should also give a better indication of 3D accelerations / gaming performance uplifts, since this is a graphics workload rather than compute.

However, GFXBench is not an ideal desktop GPU comparison tool as it doesn't usually scale well with the highest-end graphics hardware. The Aztec Ruins benchmark is more appropriate to smartphone SoCs than to GPUs like RTX 4070 Ti. Furthermore, we need to focus on the GFXBench "offscreen" tests, which assesses the rendering power of the GPU without worrying about the display resolution in use. Considering the entire benchmark suite weighs in at less than 1GB, it's a safe bet that it's not very representative of modern games that can total more than 100GB.

The screenshot above shows that the Apple M2 Ultra can achieve nearly 315 FPS in the 4K Aztec Ruins high-tier offscreen test (Metal API). Above, we mentioned that this score is over 55% faster than the M1 Ultra. A comparable score on the PC platform would again be something between the GeForce RTX 4070 Ti and RTX 4080 (DX12 API).

Big spoonfuls of salt are required here. Ideally, we need to see performance in games and other graphics and compute applications rather than these synthetic workloads that may not be representative of anything truly useful. We look forward to seeing some thorough real-world testing of the new Apple Macs with M2 Ultra processors, which should be available next week.

The best graphics cards aren't just for gaming, especially not when AI-based algorithms are all the rage. Besides ChatGPT, Bard, and Bing Chat (aka Sydney), which all run on data center hardware, you can run your own local version of Stable Diffusion, Text Generation, and various other tools... like OpenAI's Whisper. The last one is our subject today, and it can provide substantially faster than real-time transcription of audio via your GPU, with the entire process running locally for free. You can also run it on your CPU, though the speed drops precipitously.

Note also that Whisper can be used in real-time to do speech recognition, similar to what you can get through Windows or Dragon NaturallySpeaking. We did not attempt to use it in that fashion, as we were more interesting in checking performance. Real-time speech recognition only needs to keep up with maybe 100–150 words per minute (maybe a bit more if someone is a fast talker). We wanted to let the various GPUs stretch their legs a bit and show just how fast they can go.

There are a few options for running Whisper, on Windows or otherwise. Of course there's the OpenAI GitHub (instructions and details below). There's also this Const-Me project, WhisperDesktop, which is a Windows executable written in C++. That uses DirectCompute rather than PyTorch, which means it will run on any DirectX 11 compatible GPU — yes, including things like Intel integrated graphics. It also means that it's not using special hardware like Nvidia's Tensor cores or Intel's XMX cores.

Getting WhisperDesktop running proved very easy, assuming you're willing to download and run someone's unsigned executable. (I was, though you can also try to compile the code yourself if you want.) Just grab WhisperDesktop.zip and extract it somewhere. Besides the EXE and DLL, you'll need one or more of the OpenAI models, which you can grab via the links from the application window. You'll need the GGML versions — we used ggml-medium.en.bin (1.42GiB) and ggml-large.bin (2.88GiB) for our testing.

You can do live speech recognition (there's about a 5–10 second delay, so it's not as nice as some of the commercial applications), or you can transcribe an audio file. We opted for the latter for our benchmarks. The transcription isn't perfect, even with the large model, but it's reasonably accurate and can finish way faster than any of us can type.

Actually, that's underselling it, as even the slowest GPU we tested (Arc A380) managed over 700 words per minute. That's substantially faster than even the fastest typist in the world, over twice as fast. That's also using the medium language model, which will run on cards with 3GB or more VRAM — the large model requires maybe 5GB or more VRAM, at least with WhisperDesktop. Also, the large model is roughly half as fast.

If you're planning to use the OpenAI version, note that the requirements are quite a bit higher, 5GB for the medium model and 10GB for the large model. Plan accordingly.

Whisper Test Setup

Our test PC is our standard GPU testing system, which comes with basically the highest possible performance parts (within reason). We did run a few tests on a slightly slower Core i9-12900K and found performance was only slightly lower, at least for WhisperDesktop, but we're not sure how far down the CPU stack you can go before it will really start to affect performance.

For our test input audio, we've grabbed an MP3 from this Asus RTX 4090 Unboxing that we posted last year. It's a 13 minute video, which is long enough to give the faster GPUs a chance to flex their muscle.

As noted above, we've run two different versions of the Whisper models, medium.en and large. We tested each card with multiple runs, discarding the first and using the highest of the remaining three runs. Then we converted the resulting time into words per minute — the medium.en model transcribed 1,570 words while the large model resulted in 1,580 words.

We also collected data on GPU power use while running the transcription, using an Nvidia PCAT v2 device. We start logging power use right before starting the transcription, and stop it right after the transcription is finished. The GPUs generally don't end up anywhere near 100% load during the workload, so power ends up being quite a bit below the GPUs' rated TGPs. Here are the results.

WhisperDesktop Medium Model, GPU Performance

Our base performance testing uses the medium model, which tends to be a bit less accurate overall. There are a few cases where it got the transcription "right" while the large model was incorrect, but there were far more cases where the reverse was true. We're probably getting close to CPU limits as well, or at least the scaling doesn't exactly match up with what we'd expect.

Nvidia's RTX 4090 and RTX 4080 take the top two spots, with just over/under 3,000 words per minute transcribed. AMD's RX 7900 XTX and 7900 XT come next, followed by the RTX 4070 Ti and 4070, then the RTX 3090 and 3080. Then the RX 6950 XT comes in just ahead of the RTX 3070 — definitely not the expected result.

If you check our GPU benchmarks hierarchy, looking just at rasterization performance, we'd expect the 6950 XT to place closer to the 3090 and 4070 Ti, with the 6800 XT close to the 3080 and 4070. RX 6750 XT should also be ahead of the 3070, and the same goes for the 6650 XT and the 3060.

Meanwhile, Intel's Arc GPUs place at the bottom of the charts. Again, normally we'd expect the A770 and A750 to be a lot closer to the RTX 3060 in performance. Except, here we're looking at a different sort of workload than gaming, and potentially the Arc Alchemist architecture just doesn't handle this as well. It's also possible that there are simply driver inefficiencies at play.

Let's see what happens with the more complex large model.

WhisperDesktop Large Model, GPU Performance

The large model definitely hits the GPU harder. Where the RTX 4090 was 2.36X faster than the RTX 3050 with the medium.en model, it's 2.7X faster using the large model. However, scaling is still nothing at all like we see in gaming benchmarks, where (at 1440p) the RTX 4090 is 3.7X faster than the RTX 3050.

Some of the rankings change a bit as well, with the 7900 XTX placing just ahead of the RTX 4080 this time. The 6950 XT also edges past the 3080, 6800 XT moves ahead of the 3070, and the 6650 XT moves ahead of the 3060. Intel's Arc GPUs still fall below even the RTX 3050, however.

It looks like the large L2 caches on the RTX 40-series GPUs help more here than in games. The gains aren't quite as big with the large model, but the 3080 for example usually beats the 4070 in gaming performance by a small amount, where here the 4070 takes the lead.

WhisperDesktop GPU Power Use and Efficiency

Finally, we've got power use while running the medium and large models. The large model requires more power on every GPU, though it's not a huge jump. The RTX 4090 for example uses just 6% more power. The biggest change we measured was with the RX 6650 XT, which used 14% more power. The RTX 4090 and RX 6950 XT increased power use by 10%, but some of the other GPUs only show a 1–3 precent delta.

One thing that's very clear is that the new Nvidia RTX 40-series GPUs are generally far more power efficient than their AMD counterparts. Conversely, the RTX 30-series (particularly 3080 and 3090) aren't quite as power friendly. AMD's new RX 7900 cards however end up being some of the biggest power users.

Converting to efficiency in words per minute per watt provides what is arguably the better view of the power and performance equation. The Ada Lovelace RTX 40-series cards end up being far more efficient than the competition. The lower tier Ampere RTX 30-series GPUs like the 3070, 3060, and 3050 come next, followed by a mix of RX 6000/7000 cards and the lone Arc A380.

At the bottom of the charts, the Arc A750 and A770 are the least efficient overall, at least for this workload. RTX 3090, RX 6950 XT, RX 6750 XT, RTX 3080, and RX 7900 XTX are all relatively similar at 10.6–11.2 WPM/watt.

Whisper OpenAI GitHub Testing

For what will become obvious reasons, WhisperDesktop is the preferred way of using Whisper on PCs right now. Not only does it work with any reasonably modern GPU, but performance tends to be much better than the PyTorch version from OpenAI, at least right now. If you want to try the official repository, however, we've got some additional testing results.

Getting Whisper running on an Nvidia GPU is relatively straightforward. If you're using Linux, you should check these instructions for how to get the CUDA branch running via ROCm on AMD GPUs. For now, I have opted to skip jumping through hoops and am just sticking to Nvidia. (Probably just use WhisperDesktop for AMD and Intel.)

Here are our steps for Nvidia GPUs, should you want to give it a shot.

1. Download and install Miniconda 64-bit for Windows. (We used the top link on that page, though others should suffice.)

2. Open a Miniconda prompt (from the Start Menu).

3. Create a new conda environment for Whisper.

conda create -n Whisper

4. Activate the environment.

conda activate Whisper

5. Install a whole bunch of prerequisites.

conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia ffmpeg

Note: This will take at least several minutes, more depending on your CPU speed and internet bandwidth. It has to download and then compile a bunch of Python and other packages.

6. Download OpenAI's Whisper.

pip install -U openai-whisper

7. Transcribe an audio file, alternatively specifying language, model, and device. We're using an Nvidia GPU with CUDA support, so our command is: 

whisper.exe [audiofile] --model medium.en --device cuda

If you want a potentially better transcription using bigger model, or if you want to transcribe other languages:

whisper.exe [audiofile] --model large --device cuda --language en

The first time you run a model, it will need to be downloaded. The medium.en model is 1.42GiB, while the large model is 2.87GiB. (The models are stored in %UserProfile%\.cache\Whisper, if you're wondering.)

There's typically a several second delay (up to maybe 10–15 seconds, depending on your GPU) as Python gets ready to process the specified file, after which you should start seeing timestamps and transcriptions appear. When the task is finished, you'll also find plain text, json, SRT, TSV, and VTT versions of the source audio file, which can be used as subtitles if needed.

Whisper PyTorch Testing on Nvidia GPUs

Our test PC is the same as above, but this time the CPU appears to be a bigger factor. We ran a quick test on an older Core i9-9900K with an RTX 4070 Ti and it took over twice as long to finish the same transcription as with the Core i9-13900K, so CPU bottlenecks are very much a reality with the faster Nvidia GPUs on the PyTorch version of Whisper.

We should note here that we don't know precisely how all the calculations are being done in PyTorch. Are the models leveraging the tensor core hardware? Most likely not. Calculations could be using FP32 as well, which would be a pretty big hit to performance compared to FP16. In other words, there's probably a lot of room for additional optimizations.

As before, we're running two different versions of the Whisper models, medium.en and large (large-v2). The 8GB cards in our test suite were unable to run the large model using PyTorch, and the larger model puts more of a load on the GPU, which means the CPU becomes a bit less of a factor.

We created a script to measure the amount of time it took to transcribe the audio file, including the "startup" time. We also collected data on GPU power use while running the transcription, using an Nvidia PCAT v2 device. Here are the results.

Whisper PyTorch Medium Model, GPU Performance

Like we said, performance with the PyTorch version (at least using the installation instructions above) ends up being far lower than with WhisperDesktop. The fastest GPUs are about half the speed, dropping from over 3,000 WPM to a bit under 1,600 WPM. Similarly, the RTX 3050 went from 1,328 WPM to 565 WPM, so in some cases PyTorch is less than half as fast.

We're hitting CPU limits as you can tell by how the RTX 4090 and RTX 4080 deliver essentially identical results. Yes, the 4090 is technically 0.5% faster, but that's margin of error.

Moving down the chart, the RTX 3090 and RTX 4070 Ti are effectively tied, and the same goes for the RTX 4070 and RTX 3080. Each step is about 10–15 percent slower than the tier above it, until we get to the bottom two cards. The RTX 3060 is 24% slower than the RTX 3070, and then the RTX 3050 is 27% slower than the 3060.

Let's see what happens with the more complex large model.

Whisper PyTorch Large Model, GPU Performance

The large model increases the VRAM requirements to around 10GB with PyTorch, which means the RTX 3070 and RTX 3050 can't even try to run the transcription. Performance drops by about 40% on most of the GPUs, though the 4090 and 4080 see less of a drop due to the CPU limits.

It's pretty wild to see how much faster the Const-Me C++ version runs. Maybe it all just comes down to having FP16 calculations (which are required for DX11 certification). Not only does it run faster, but we could even run the large model with a 6GB card, while the PyTorch code needs at least a 10GB card.

Whisper PyTorch, GPU Power and Efficiency

We only collected power use while running the medium model — mostly because we didn't think to collect power data until we had already started testing. Because we're hitting CPU limits on the fastest cards, it's no surprise that the 4090 and 4080 come in well below their rated TGP (Total Graphics Power).

The 4090 needs just under 200W while the 4080 is a bit more efficient at less than 180W. Both GPUs also use more power with PyTorch than they did using WhisperDesktop, however. The RTX 3080 and 3090 even get somewhat close to their TGPs.

Converting to efficiency in words per minute per watt, the Ada Lovelace RTX 40-series cards are anywhere from 43% to 103% more efficient than the Ampere RTX 30-series GPUs. The faster GPUs of each generation tend to be slightly less efficient, except for the RTX 4080 that takes the top spot.

Again, it's clear the OpenAI GitHub source code hasn't been fully tuned for maximum performance — not even close. We can't help but wonder what performance might look like with compiled C++ code that actually uses Nvidia's tensor cores. Maybe there's a project out there that has already done that, but I wasn't able to find anything in my searching.

Whisper AI Closing Thoughts

As we've noted on numerous occasions, support for Nvidia's GPUs via CUDA tends to be the de facto standard in many AI projects these days. The most popular projects often get AMD GPU support... eventually; sometimes there are even Intel forks. For Whisper, there are currently a couple of options: ROCm and CUDA translation running under Linux, or a DirectCompute implementation that's far more compact and performant. It's almost like Python tends to not be an optimal way of coding things.

Hardware requirements can also vary quite a bit, depending on which version of Whisper you want to use. WhisperDesktop basically only needs enough VRAM to hold the model, while the PyTorch version requires about double the VRAM.

That means, like Stable Diffusion that can run even on GPUs with 4GB of memory, WhisperDesktop isn't particularly demanding on VRAM. But the PyTorch version is a lot more like other large language models (LLMs) and definitely wants more memory, sometimes a lot more. OpenAI only released a large model for Whisper (based on GPT-2) that needs about 10GB of memory, but certainly it could work to improve the accuracy and quality even more while requiring additional VRAM.

Our testing of Whisper also shows that coding and optimizations can often trump hardware. In theory, Nvidia's tensor cores should allow a GPU like the RTX 4090 to be potentially 2.7X faster than an RX 7900 XTX, based on FP16 compute potential — double that to 5.4X faster if sparsity applies. But WhisperDesktop doesn't bother with tensor cores and so we're left with FP16 GPU shader compute performance. Or maybe not even that, as in theory the double performance FP16 of RDNA3 GPUs would then put them ahead of the Ada GPUs.

If you have any suggestions on other AI workloads you'd like to see us test, or a different version of Whisper that might better take advantage of Nvidia's tensor cores and/or Intel's XMX cores, let us know in the comments. We're always looking for potentially useful and interesting tests to quantify GPU performance.

Ultimately, Whisper isn't necessarily the sort of tool that tons of people will find useful, and there are other transcription services and speech recognition applications available. However, given the price — free, assuming you have any GPU made in the past decade or so — Whisper once again illustrates the potential of AI training algorithms. It's certainly a tool we'll keep in mind for the future.

Microsoft’s DirectX developers have released DirectStorage 1.2. The headlining introduction with this version is the option to enable buffered IO mode to speed transfers from slower storage media such as HDDs. Another tweak allows devs to check the decompression processing path currently in use. Last but not least, there is an intelligent switch to improve GPU decompression performance and a handful of bug fixes.

To provide some background to DirectStorage, Microsoft’s DirectX API is designed to bring speed, bandwidth, and latency optimizations to the Windows storage subsystem. In addition, it brings Xbox technology, where game assets stream directly from storage to GPU. On Microsoft’s Xbox, the technology was dubbed the Xbox Velocity Architecture and relied on the new fast NVMe storage of the Xbox Series X / S consoles. Initially, fast SSDs were a requirement of the PC’s DirectStorage implementation, but with v1.2, Microsoft accommodated gamers still hanging onto spinning rust.

New in DirectStorage 1.2 is the ability to use the same code path on both thrillingly fast SSDs and ye olde HDDs. Microsoft says before DirectStorage 1.2 files would be opened in unbuffered mode, getting data from storage to GPU as immediately as possible. However, it has now made it possible to run DirectStorage in buffered mode “to mask the long seek times” of HDD technology.

Developers using DirectStorage 1.2 must enable the HDD-compatible setting using a new switch, and do some further background configuration. Microsoft stresses that it is the game/application developer’s responsibility to use and apply this setting correctly, and that it should only be used for slower HDDs. It isn’t mentioned in Microsoft’s blog post, but buffering implies this feature will create additional system RAM overheads.

The second major addition in DirectStorage 1.2 is not a new feature for end-users, but designed to allow developers to better understand the decompression paths being used by the API. The aptly named new GetCompressionSupport API is important because “there are scenarios where a CPU-based fallback path is chosen,” explains Microsoft. With a better understanding of why and when the fallback mode comes into play, developers can optimize texture resolution setting better.

The only highlighted performance improvement in DirectStorage 1.2 also focuses on GPU decompression of textures. Microsoft has moved “the copy after GPU decompression onto the compute queue for GPUs on which this is faster,” for this performance gain. A trio of bug fixes for DirectStorage are also highlighted by Microsoft’s blog post.

We reported on the much more significant update to DirectStorage 1.1 back in October 2022, when Microsoft claimed to have sped game loading by as much as 40%. In January we saw the first DirectStorage 1.1 comparisons where the GPU was the variable, and a win for Intel’s Arc Alchemist architecture.

Microsoft has announced a new DirectX12 GPU optimization feature in conjunction with Resizable-BAR, called GPU Upload Heaps, that allows the CPU to have direct, simultaneous access to GPU memory. This can increase performance in DX12 titles and decrease system RAM utilization since the feature circumvents the need to copy data from the CPU to the GPU. The new feature is available now in the Agility SDK.

We don't know the actual implications of this feature, but the performance advantages could be significant. Graphics card memory sizes and video game VRAM consumption are getting larger and larger every year. As a result, the CPU needs to move more and more data between itself and the GPU.

With this feature, a game's RAM and CPU utilization could decrease noticeably due to a reduction in data transfers alone. This is because the CPU no longer needs to keep copies of data on both system RAM and GPU VRAM to interact with it. Another bonus is that GPU video memory is very fast these days, so there should be no latency penalties for leaving data on the GPU alone. In fact, there will probably be a latency improvement with CPU access times on high-end GPUs with high-speed video memory.

For gamers, the only requirement you'll need is Resizable-Bar or Smart Access Memory support on both your CPU and GPU. Resizable-bar is the foundation for GPU Upload Heaps since the feature enables Windows to manage GPU VRAM directly. 

For developers, the feature is already supported by Nvidia, Intel, and AMD drivers. For example, it's already included in Nvidia's latest Game Ready and Studio Drivers (version 531.41 or newer) and Intel A-series/Xe GPUs (with driver 31.0.101.4255 or newer). For AMD GPUs, developers must consult their AMD alliance manager to get a supported driver.

The feature just came out from Microsoft, so don't expect it to be implemented in games anytime soon. Instead, developers will need time to see whether or not it will provide a significant enough performance advantage to make it worthwhile.

AMD is set to give a sneak peek of its next-generation image improving FidelityFX technologies at one of its upcoming sessions at the Game Developers Conference 2023 in late March, the company revealed this week.  

AMD's FidelityFX package is mostly known for the company's Super Resolution image upscaling technology, as it is supported by a wide variety of games and can work on AMD-powered game consoles and with non-Radeon hardware, as well. But AMD's FidelityFX is actually a large collection of eye candy enhancing technologies designed to greatly improve visual quality of games or improve their framerates without visible degradation of quality.  

In addition to FSR, AMD's current version of FidelityFX includes things like Contrast Adaptive Sharpening (CAS), Combined Adaptive Compute Ambient Occlusion (CACAO), Variable Rate Shading (VRS), Stochastic Screen Space Reflections (SSSR) and ray-traced shadow Denoiser just to name a few. Many of AMD's FidelityFX technologies are used by games, but none of them has managed to become as popular as Super Resolution (we assume something similar is true about Nvidia's image enhancing technologies). 

It is not completely clear what other technologies AMD plans to add to its FidelityFX collection, but Super Resolution 3.0 is certainly one of the options. Other possibilities also include various ways to improve performance of ray tracing-enabled games, but it remains to be seen what AMD could possibly do here given hardware capabilities of RDNA 2 and RDNA 3-based GPUs.  

"In this session, FidelityFX SDK lead Jason Lacroix will walk you through the features and architecture of the SDK, discuss how to implement FidelityFX technology on top of your engine's own abstractions, and look at the effects released as part of the SDK," AMD's description of the session reads. "This session will also include an exciting sneak peek of new FidelityFX technologies that will be available soon. " 

On a side note, AMD this month released a plugin for Unreal Engine 5.1 that enables its FidelityFX Super Resolution 2.2 technology in games that use this engine. The plugin allows developers to easily add FSR 2.2 to their games and gives them some additional control to tackle ghosting and improve appearance of things like animated foliage that tends to suffer from ghosting artifacts.

Multi-display configurations are used by many people in the small office/home office space (SOHO) these days. Intel's moderately-priced Arc A770 graphics cards would be a good fit for these setups if they did not consume 40W of power in idle mode when driving two or more monitors. Apparently, the latest drivers fix the issue and cut down power consumption by four times, reports VideoCardz. But there is a catch. 

Driving several displays at once requires powering up the appropriate number of display pipelines that consume energy. Also, sending high-resolution images with deep colors to monitors requires transferring loads of data and this means power. But in idle mode GPUs don't need to send 60 or 120 high-resolution deep color images per second to displays, which is something that Intel's Arc-based graphics cards did (and competing products from AMD and Nvidia do not). 

Apparently, Intel's latest driver version 31.0.101.4146 has the fix for this issue for at least some dual-display setups, according to Intel Community Github. There is still one bump though: it only works for dual-monitor configurations and does not seem work for systems with three or more screens. 

Driver issues are arguably the main thing that has plagued Intel's Arc discrete graphics cards for desktops since their launch. The good news is that the latest drivers reduce power consumption of dual-display setups almost to the levels of configurations with one monitor. The company also recently released drivers that improve performance in DX9 games.

Intel’s Arc Alchemist GPUs launched toward the latter part of 2022, vying for a spot among the best graphics cards. You can see where they land on our GPU benchmarks hierarchy as well, but that's only part of the story. Intel has been busily updating drivers on a regular basis since launch, and rarely does a fortnight pass without at least one new driver to test.

And that's the crux of the story: How much better are Intel’s latest drivers compared to the original launch drivers? The company made a lot of noise about its improved DirectX 9 performance, claiming 43% higher performance on the A750 with recent drivers compared to the original 3490 launch drivers — and at the same time dropping the price of the A750 to $250 down from $290. What about newer games, though?

We set about testing (and retesting) every Arc graphics card, including the Arc A770 8GB using an ASRock model that we haven't quite got around to reviewing just yet, due to all the retesting that's been going on.

We began our testing a couple of weeks back, right after the 4123 Intel drivers became available. Last week, version 4125 drivers came out, but a quick check on one of the Arc cards indicates the only changes are related to some recent game launches and that the performance otherwise remains the same as the 4123 drivers for our test suite.

Update: Intel just released the 4146 drivers on February 28, 2023. These fix a problem that we noticed with Red Dead Redemption 2 crashing on the Arc A750, among other things. We have included the results for that one game from the newer drivers (as before we couldn't get a result).

Plugging the Holes in the Arc

Arc arrived with what were arguably the best GPU drivers Intel has ever created – but that’s not saying much. A few years ago, a lot of games simply refused to work at all on Intel’s integrated graphics solutions — and even those games that worked would often perform poorly. Intel’s DG1 helped pave the way for more frequent driver updates, but things were still iffy back in 2021.

Since the first dedicated Arc A380 cards started showing up in China — which meant they also started getting shipped to individuals around the world — Intel has been cranking out updated drivers on a regular basis. There are presently eleven different versions of Arc drivers available from Intel: five WHQL (Windows Hardware Quality Labs) certified and six beta drivers, but there were other “hotfix” or beta drivers that are no longer listed.

Many of the updates have been targeted at one or two specific games — there were two different drivers that addressed problems with Spider-Man Remastered performance on the Arc A380, for example. Other updates have had much further reaching ramifications, with the biggest change being DirectX 9 optimizations.

DirectX 9, 20 Years Later

Microsoft’s DirectX 9 API became publicly available back in 2002, but it continues to see quite a bit of use even today. Some of that goes back to the venerable Valve Source engine used for Half-Life 2, which saw widespread adoption for various mods and spinoff games like Counter-Strike: Global Offensive, Left 4 Dead 2 and Team Fortress 2; StarCraft 2 and League of Legends likewise continue to use DX9.

Considering how old the API is, it’s easy to understand Intel’s decision to “focus on modern [DX12 and Vulkan] APIs” for the launch of the Arc graphics cards. Most recently released games use DirectX 12 or DirectX 11, or sometimes Vulkan. Why put a bunch of effort into trying to optimize the drivers for older games? For DX9 compatibility, Intel opted for support via DX12 emulation, which probably sounded like a good idea at some point.

The problem is that a lot of older games remain immensely popular. Token support for DX9 via DX12 emulation might have sounded fine on paper, but in practice it was a glaring indication that Intel’s GPUs and drivers weren’t as good as AMD and Nvidia drivers. That’s even though most DX9 games aren't particularly demanding, especially when compared with recent DX11/12 releases, performance in DX9 is only as good as the emulation.

Simply put, at launch, the DX9 via DX12 emulation for Intel Arc wasn't very good. It was very prone to fluctuations in frametimes, leading to stuttering in games. People complained and Intel decided to rethink its DX9 strategy. The solution Intel came up with was to leverage more translation layers to turn DX9 API calls into either DX12 calls or Vulkan calls. The DX12 emulation still comes from Intel’s internal work on the drivers, or perhaps Microsoft’s D3D9On12 mapping layer, but now with a bit more tuning.

The bigger change appears to be the use of DXVK — DirectX to Vulkan — from Steam for “some cases.” Intel hasn’t detailed exactly which games use DXVK and which use DX12 emulation, but whatever the changes, frametimes and average performance have increased substantially. This was all rolled out with Arc’s December driver update.

This is great news if you want to buy an Arc GPU for a modern PC to run old games. Remember: PCIe Resizable BAR is basically required for Arc GPUs, which means you generally need a 10th gen or later Intel CPU, or Ryzen 5000 or later AMD CPU. Otherwise, you’ll lose about 20 percent of the potential performance, based on testing, at which point you should just buy an AMD or Nvidia GPU.

So Intel can't simply ignore older games, as weak performance in games that millions of people still play on a regular basis makes all Intel GPUs look bad. But how much has really changed?

Arc Performance Update: Test Setup

Given all the changes, we pulled out a collection of fifteen different games, including two DX9-based games, to check how the various cards perform. The fifteen games and their respective APIs consist of Borderlands 3 (DX12), Bright Memory Infinite Benchmark (DXR, or DirectX Raytracing), Control Ultimate Edition (DXR), Counter Strike: Global Offensive (DX9), Cyberpunk 2077 (DXR), Far Cry 6 (DX12), Flight Simulator (DX11), Forza Horizon 5 (DX12), Horizon Zero Dawn (DX12), Mass Effect 2 (DX9), Metro Exodus Enhanced (DXR), Minecraft (DXR), Red Dead Redemption 2 (Vulkan), Total War: Warhammer 3 (DX11), and Watch Dogs Legion (DX12).

All of the games were tested at "medium" and "ultra" settings, except for Mass Effect 2 — it only has three graphics options, and it's old enough that we simply enabled all three options for the "ultra" testing. The three A7-class cards were tested at 1080p, 1440p, and 4K resolutions, except for in ray tracing games where we dropped 4K testing (for what will become obvious reasons in a moment). The A380 dropped the 4K testing completely, and also dropped 1440p testing for ray tracing games.

Each game and setting were tested at least three times, dropping the high and low result. Where there was more variance (looking at you, CSGO), the tests were done five times and we took the median result. This is an important point, because sometimes the first run was very poor (again, CSGO, particularly on the 3490 drivers).

All these tests were done using the same Core i9-12900K PC that we used for the previously published Arc GPU launch reviews, so that we can look at how performance has changed. How would the old drivers differ from the launch performance and now? That's where things get a bit messy.

There have been multiple major updates to games in our test suite over the past few months, and we've been working to shift all of our testing to a new system with a Core i9-13900K, plus adjusting our test suite and settings as needed. Our older test system is now effectively (slightly) out of date, and while we've retested some cards with the updated games, it was best to just hit the reset button.

What we know for certain is that Cyberpunk 2077, Flight Simulator, Forza Horizon 5, and Total War: Warhammer 3 all had updates that changed performance and/or settings quite a bit — on some GPUs more than others. Cyberpunk now has DLSS 3 and FSR 2.1 support, Flight Simulator added DLSS 3 and DLAA, and Forza Horizon 5 added DLSS 2 and FSR 2.2 along with TAA. All three of those games also added support for Nvidia Reflex. As for Total War: Warhammer 3, a couple of months back there was a major update that improved performance by roughly 20% on virtually all GPUs.

And those are just the changes that were readily visible. Almost all of the other games in our test suite have also received various updates, and keeping track of what has changed and what remains consistent is difficult. Regardless, we'll have the original launch performance data in our charts, with an asterisk indicating it's from a different version of the game and may not reflect current performance. Just for good measure, we also applied all the latest Windows 11 updates and flashed our motherboard BIOS.

And somewhere along the way, we ran into a problem. Windows 11 defaults to Virtualization Based Security (VBS) being on, and we previously turned it off. Some time in the past few months, it got turned back on (likely when I was poking around at Stable Diffusion and tried using WSL2). You can still turn this off, but we didn't realize it was on until after testing was completed. So, if you saw this article earlier and we talked about how performance dropped about 7% on average since our launch testing last year, that's the explanation. We retested the A750 with VBS disabled and found that, in general, performance matched or exceeded our October 2022 results and was around 7% faster than our current VBS-enabled testing.

Intel Arc Graphics Performance: 3490 Launch vs. Current 4123 Drivers

We have two results for each of the test cards in the following chart: 3490 are the launch drivers for the A770 and A750 — the Arc A380 was available several months earlier with 3259 drivers, and we saw 3267, 3268, and 3276 drivers released before our review was finished. But we're sticking with the 3490 drivers for all of the Arc "launch" drivers to keep things consistent. 4123 are the "current" drivers (4125 drivers are "Game On" for Company of Heroes 3, The Settlers: New Allies, Atomic Heart, and Wild Hearts; more recently, 4146 are "Game On" for Destiny 2: Lightfall and Wo Long: Fallen Dynasty). Let's start with the 1080p medium test suite.

We should also discuss some of the anomalies we'll see before we get to the charts. First, Red Dead Redemption 2 using the Vulkan API consistently crashed our test PC with the A750 card — and it crashed a different test PC as well. It was mostly stable using the original 3490 launch drivers, enough so that we could complete a few test runs before the PC would restart, but every later driver we checked had stability problems. Intel was able to replicate the problem and it’s limited to the A750; a fix was made available in the latest 4146 drivers, and we've included those results here (even though they're still labeled "4123" for our charting purposes).

Next, Bright Memory Infinite Benchmark, a ray tracing title, has a rather severe VRAM memory leak problem — or at least, that's how it appears. We could complete one benchmark run at 1080p with the Normal preset on the 8GB cards at 1080p, but higher settings (and the A380) would go from being somewhat smooth to a stuttering mess partway through the benchmark, with minimum framerates of 1 fps. This is another bug that Intel has confirmed and is working to fix — it's not clear if this affects the game as well, or only the standalone benchmark (which is far more demanding and better looking than the actual game, if you were wondering).

Finally, Minecraft performs very poorly on all Arc GPUs right now. When the cards first launched, you couldn’t even turn on ray tracing in the game, as apparently it was coded with a “whitelist” of cards that support DXR. An update to the game in December finally allowed the Intel GPUs to run with ray tracing enabled, but the results remain much lower than expected. As a point of reference, in other demanding ray tracing games like Cyberpunk 2077, the A770 is only 10 percent slower than an RTX 3060; in Minecraft, the RTX 3060 is currently nearly triple the performance. In fact, even the RX 6600 outperforms the fastest Arc GPU by 25 percent, where the A770 is 60 percent faster in Cyberpunk. Intel is also looking into the Minecraft performance, so hopefully that improves with a future driver.

Disclaimers aside, here are the results of our testing.

Overall performance changes between the 3490 and 4123 drivers at 1080p medium are… not a lot. That's pretty much expected. The three A7-class cards are 1–2 percent faster, while the A380 is 5% faster.

We don't have a "medium" result from Mass Effect 2, and CSGO appears to be mostly CPU limited to around 360 fps (with fluctuations as it's a non-deterministic test sequence — I played against harmless bots on the Mirage map, running the same route each time, basically a loop to the right through bomb site A as a counter terrorist). That means two of the biggest potential gains are out of the picture, except on the A380 where CSGO performance was 16% higher on the updated drivers.

Flipping through the individual charts, the only major differences are the somewhat larger than expected drop in performance in Control on the two Intel Arc Limited Edition cards — the ASRock A770 8GB and A380 didn't show such behavior — with modest gains elsewhere. That's mostly the same pattern we'll see at the other test settings.

Moving the quality setting to "ultra" results in bigger gains overall for the latest drivers, but much of that comes from the two DirectX 9 games. Overall, we're looking at a 5–10 percent aggregate improvement in performance with the 4123 drivers compared to the 3490 launch drivers. However, if we drop the two DX9 games from the geometric mean calculation, then the net improvement is only 3–6 percent.

There are positive and negative changes once again, with some of the biggest deltas coming in DXR games where they're often meaningless — A380 benchmarked at 6.1 fps compared to 4.5 fps, is a 34% increase in performance… at completely unplayable framerates.

There are a few cases where we see up to a 10% improvement in games that aren't using DXR or DX9, like Total War: Warhammer 3, but overall the changes are mostly within the margin of error, or at least close to it. But look at the two DX9 games… or at least the one DX9 game.

Mass Effect 2 performance improved by anywhere from 40% on the A380 to as much as 90% on the A770 8GB. That's for average fps, but the 99th percentile fps shows even bigger gains. The low fps results are 60% faster with the 4123 drivers on the A380 and up to 135% higher on the A770 8GB.

What about CSGO? The A770 16GB card shows basically no change, which makes sense given the large amount of VRAM. The A770 8GB improved by 9%, and the A750 by a few percent, but the A380 performance was slightly slower. Perhaps Intel focused its DX9 optimizations more on the bigger chips than on the ACM-G11 used in the A380? There's also the question of what map to use and how to test, so perhaps other CSGO maps would show different performance results, or maybe playing online rather than against bots.

Moving to 1440p ultra testing, the A380 is now mostly not a factor to consider outside of playing older DX9 games. Overall, we again see a 5–10 percent improvement in performance for the three A7-class Arc cards, with the A750 showing the biggest overall increase. Of course, if we omit the DX9 games, it goes back to being almost margin of error, 2–3 percent, with the A750 still improving by 9% thanks to a big increase in the Cyberpunk result, from an unplayable 10 fps to a still-unplayable 14 fps.

The gains in CSGO are still pretty muted, 3% on the A770 16GB, 10% on the A770 8GB, and 17% on the A750 — and the A380 performance again dropped, this time by 6%. Mass Effect 2 on the other hand shows gains of 50% on the Intel Limited Edition cards, and 70% on the ASRock A770 8GB — but only 22% on the A380, which still manages to break 100 fps on the 13-year-old game.

There's not much more to add. Outside of the two DX9 games, most of the changes aren't something you'd really notice as a gamer. It's worth mentioning that improvements in minimum FPS on the DX9 games are often even larger than the average FPS gains — CSGO had a lot of stuttering on the first couple of test runs with the 3490 drivers, and that's mostly gone now, but our use of the median result of five runs partially obscures this fact.

Last we have the 4K testing results, though we didn't do any DXR tests so the overall chart is down to 10 rasterization games. The three ACM-G10 based graphics cards show 3–7 percent aggregate gains in performance using the full test suite, but only a 1–5 percent increase if we omit the two DX9 games.

Mass Effect 2 now shows improvements of 10–35 percent, with the 16GB card benefiting the least, and the factory overclocked ASRock card benefiting the most. CSGO does show modest gains as well this time on the two Intel Limited Edition cards, while the ASRock card only shows a 3% improvement. Again, there was a lot of variability between runs in CSGO, and the minimum fps does show much better results with 35–50 percent higher 99th percentile framerates.

Playing Intel Arc, the Iteration Game

It's been an interesting six or so months since the Arc A380 first appeared. Given all the claimed performance gains, we were hoping to see more… consistency across our test suite. The latest drivers generally do outperform the launch drivers, but some issues remain, and that's on a still relatively limited test suite of 15 games.

None of our findings are particularly surprising, including the continued oddities. AMD and Nvidia have been playing the drivers game for decades, and even they have occasional problems. Intel has had graphics drivers for integrated graphics solutions for decades as well, but the performance on tap was so low that there often wasn’t any pressure to even try to optimize for newly released games. Creating dedicated graphics solutions changes the user expectations, and now Intel is playing catch up.

The most important things to note with the driver updates coming out of Intel are that there’s a regular cadence, and that older DirectX 9 games got some much-needed TLC to get them running more smoothly. It’s rare that more than two weeks passes without some new driver being announced, and Intel has also been doing better on getting Game Ready drivers out for bigger launches, including "Game On" drivers for the Hogwarts Legacy and Atomic Heart launches over the past couple of weeks.

If Intel can keep that up, plus add the occasional larger overhaul that provides more universal improvements, in a couple of years we hopefully won't even need to have a serious discussion about Intel’s GPU drivers. And of course, we're only scratching the surface with our game performance testing.

The games, settings, resolutions, and even test sequences can absolutely affect performance and potential gains from newer drivers. Maybe a different CSGO map would have yielded bigger performance improvements from the newer drivers. Or maybe the 12900K CPU was holding us back a bit more than the 13900K Intel used in its testing. We're looking into our CSGO results and may do some further testing and update this article as appropriate.

Regardless, let’s hope that Intel’s GPU and driver teams get the needed time to continue working on future drivers and architectures. Right now, it doesn’t sound like many people are biting on Arc graphics cards. Intel recently dropped the official price of its A750 Limited Edition card to $249, down $40 or 14 percent from the launch price. Big companies don’t do that if parts are flying off the shelves. How long can Intel continue to bleed money on consumer GPUs? Perhaps more importantly, can Intel afford to not invest more money into consumer GPUs?

There’s a lot riding on Arc Battlemage, which is currently slated to arrive in 2024. Intel's Raja Koduri has also said that the company is more interested in competing with mainstream GPUs than worrying about halo cards, with $200–$300 being the sweet spot. We’ll have to wait and see if Intel can narrow the gap between Arc and its AMD and Nvidia competition come next year — which, incidentally, is when we expect we'll see the next generation Nvidia Blackwell and AMD RDNA 4 GPUs launch.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Minecraft RTX officially released to the public in April 2020, after first being teased in August 2019 at Gamescom 2019. Since then, a lot of things have changed, with regular updates to the game and a whole slew of ray-tracing capable graphics cards that didn't exist in early 2020. The best graphics cards now all have DXR (DirectX Raytracing) and VulkanRT, and you can see the full view of how the various GPUs stack up in other games by checking out our GPU benchmarks hierarchy. But how does Minecraft RTX run these days?

To be clear, we're not talking about regular Minecraft here. That can run on a potato, even at high resolutions. For example, the Intel Arc A380 — which as you'll see below falls well below a playable 30 fps with ray tracing on — plugs along happily at 93 fps at 4K without DXR, using 4xMSAA and a 24 block rendering distance. Or put another way, performance is about ten times higher than what you'll get with ray tracing enabled.

So why would anyone want to turn on ray tracing in the first place, if the performance hit is so severe? First: Look at all the shiny surfaces! Minecraft RTX is a completely different looking game compared to vanilla Minecraft. Second, with cards like the Nvidia RTX 4090 now available, you can still get very good performance even at maxed out settings.

We've now tested Minecraft RTX on every DXR-capable graphics card, ranging from the lowly AMD RX 6400 and Nvidia RTX 3050 up to the chart topping 4090. We're using the latest drivers as well: AMD 23.2.2, Intel 31.0.101.4369, and Nvidia 528.49. (The Intel drivers are the main change with this update, as the latest release has improved Minecraft RTX performance quite a bit.) We'll have more details on the ray tracing implementation of Minecraft below, but let's get to the benchmarks first, since that's likely why you're here.

Minecraft RTX Testing Details

We've been using Minecraft RTX — or just Minecraft, as it's normally referred to these days — for a while now in our graphics card reviews. If you want to see just how punishing DXR calculations can be, it's great for stressing lesser GPUs. We're now running on a state of the art test system, with a Core i9-13900K with 32GB of DDR5-6600 memory, connected to a 4K 240Hz Samsung Neo G8 32 monitor. Full details on our test PC are in the boxout.

We previously tested Minecraft RTX using other CPUs and different memory conditions. While those do matter to some extent, especially if you're using something like an RTX 4090, we feel that anyone trying to run Minecraft RTX will likely have at least 16GB of memory and a reasonably capable CPU. For these tests, we're only looking at the impact of the GPU, and we've more or less maxed out all of the other hardware to eliminate bottlenecks as much as possible.

We're testing at four settings: 1920x1080 with 8 RT Chunks rendering distance, and 1920x1080, 2560x1440, and 3840x2160 with a 24 RT Chunks rendering distance — the maximum available. We're focusing on native rendering for the most part, but we do have a couple of DLSS results in the charts to show how that can affect things.

Note that in vanilla Minecraft, the rendering distance can be set as high as 96 chunks (it used to be 160, or perhaps that varies by map). Higher rendering distances will put more of a strain on the CPU, but certainly running with 64 chunks isn't uncommon.

Our test map is Portal Pioneers RTX, which includes a helpful benchmarking setup. You can find various generic RTX texture packs to use for your own maps, though you can't simply enable ray tracing effects without such a pack. That's because the ray traced rendering requires additional information on the various blocks to tell the engine how they should be rendered — e.g. are they reflective, do they glow, etc.

One other important note is that DLSS upscaling can't be tuned, it's either on or off with an Nvidia GeForce RTX card. As far as we can tell, for 1080p and lower resolutions, DLSS uses 2x upscaling (Quality mode), 1440p uses 3x upscaling (Balanced mode), and 4K and above use 4x upscaling (Performance mode). The blocky nature of Minecraft does lend itself rather well to DLSS upscaling, however, so even 4x upscaling at 4K still looks very good.

Minecraft RTX Graphics Card Performance 

We've tested virtually every card at every setting that could possibly make sense, including plenty of settings that don't make sense at all! Basically, we stopped testing higher resolutions and settings once a card dropped below 20 fps, though we did test all of the cards at both 1080p settings just for the sake of completeness. Let's start with the "easiest" setting, 1920x1080 and an 8 RT chunk render distance.

If you've seen results like this before, you won't be surprised to see that Minecraft RTX runs far better on Nvidia RTX hardware than on anything else — and that's not even factoring in DLSS upscaling. AMD does offer several GPUs that can now break 60 fps, but even the RX 7900 XTX falls behind the RTX 3080, and the fastest previous gen RX 6950 XT trails the RTX 3060 Ti.

To get performance above a steady 60 fps (meaning, the 1% minimum fps is also above 60), you'll need at least an RTX 2080 Ti, or alternatively an RTX 2060 basically gets there if  you turn on DLSS. If you're only looking to clear 60 fps average performance, the RTX 2080 Super and above will suffice. For a bare minimum playable experience of over 30 fps (average), the RX 6650 XT and above will suffice.

Intel's Arc Alchemist GPUs have had problems with Minecraft RTX since they first gained support. The latest 3469 drivers help quite a bit, though some oddities remain and we suspect there's still room for improvement. At least 1080p with an 8 chunk RT render distance is now playable on the A750 and above. The Arc A750 usually lands pretty close to the RTX 3060, where here even the A770 16GB ends up being quite a bit slower.

Finally, let's talk about DLSS upscaling, with the three sample cards of RTX 2060, RTX 3080, and RTX 4090. The RTX 2060 performance gets a massive boost of 72%, taking it from a borderline 41 fps to a very playable 70 fps. The RTX 3080 sees a decent 52% jump, going from 107 fps to 162 fps. There are limits to what DLSS can do, however, and while the RTX 4090 does improve its average fps by 24%, the 1% lows actually drop a few percent due to the DLSS overhead.

Maxing out the RT Chunks render distance at 24 increases the demands on the GPU a decent amount, at least in areas where you have a relatively unobstructed view. Because the BVH tree — Bounding Volume Hierarchy, used to help optimize the ray tracing calculations — gets constructed on the CPU and passed over to the GPU, some of the hardest hit cards with the increased view distance are the fastest cards. The RTX 4090 performance drops by 31% for example, while the RTX 3070 only loses 15% compared to 8 RT Chunks.

If you're looking for at least 60 fps, the RTX 2080 Ti now marks the bare minimum GPU you'll need, while 60 fps minimum basically requires an RTX 3080 or faster. Only AMD's RX 7900 XTX/XT cards can break 60 fps, and in both cases minimums fall quite a bit short of that mark. Dropping your fps target to 30 fps opens the door to more cards, but you'll still need at least an RTX 2060 to get minimums above 30, or an RTX 3050 to get the average fps to 30.

This is another good example of how much faster Nvidia's GPUs are at ray tracing calculations compared to AMD. In traditional games, the RTX 3050 delivers about 23% lower performance than an RX 6600, but in Minecraft RTX it basically ties the RX 6700 XT — a card that's 94% faster in our rasterization GPU benchmarks!

Intel's ray tracing hardware normally looks much better than this as well, for example the Arc A750 lands between the RX 6700 XT and RX 6750 XT in Cyberpunk 2077, one of the most demanding DXR games around (outside of Minecraft). Right now, Intel's Arc A700 cards come in just behind the RX 6600. That's with the updated drivers, but when you look at the 1440p results below, you'll see that something still seems off.

Last, let's talk DLSS again, which is still in Quality mode here. The gains on the RTX 2060 are 65% now, slightly lower than above so perhaps other bottlenecks are becoming more of a factor. The RTX 4090 at the other extreme sees a 22% increase again, again a touch less than before. Finally, our "middle of the road" RTX 3080 gets a 65% improvement. In other words, the DLSS performance increase appears to be mostly resolution dependent in Minecraft RTX, even though the ray tracing workload increased.

Stepping up to 2560x1440 with maxed out settings at 24 RT Chunks, we've dropped the three slowest GPUs from our benchmarks (A380, RX 6400, and RX 6500 XT), but we certainly could have skipped testing a bunch of the other cards. Roughly half of the tested cards can't keep the 1% lows above 30 fps, and only a handful of cards can even average 60 fps — six to be precise, at least when we're not using DLSS. And we're not even looking at a worst-case scenario, as there are certainly more demanding maps available in Minecraft RTX.

If you want to fully clear 60 fps, the only card that can do that at native 1440p is the RTX 4090. To average 60 fps or more, you'll want at least an RTX 3080 Ti. Also note how close the RTX 3080 10GB and RTX 3080 12GB are in performance, which indicates Minecraft RTX hits the RT Cores and GPU shaders far more than it depends on memory bandwidth. We'll see that pattern continue even at 4K below.

At least there are still 22 cards that can still break 30 fps average, which is much better than when Minecraft RTX first became available. Back then, only the RTX 2070 Super, RTX 2080, RTX 2080 Super, and RTX 2080 Ti were able to reach "playable" performance — and the Titan RTX as well, if you want to include that status symbol of GPU-dom. But in our updated tests, the RTX 2070 Super now falls short and you'll need an RX 6800 XT or faster.

Intel's Arc A700 cards are barely slower at 1440p than at 1080p as well. Now they land between the RX 6650 XT and the RX 6700 10GB, which seems pretty reasonable overall. The fact that they're only 1–2 fps slower than at 1080p means there's room for improvement in the 1080p results, though we wouldn't expect too much change at 1440p.

DLSS now switches to the "Balanced" algorithm with 3X upscaling, which means it can provide and even more significant boost to performance. The RTX 2060 is basically unplayable at just 22 fps native, but with DLSS it more than doubles performance to hit 46 fps. Even the behemoth RTX 4090 gets a 52% increase in performance thanks to DLSS, going from 102 fps to 156 fps. The RTX 3080 plays middle fiddle again with a 75% increase, jumping from 57 fps to 100 fps and delivering a far better gaming experience.

Wrapping things up with our 4K testing, we've naturally dropped most of the GPUs from testing. Where we had 38 GPUs in total at 1080p, we're down to 19 now, and half of those are again struggling — unless you want to turn on DLSS, which you should if you have an Nvidia card. We just wish you could decide whether to use Performance, Balanced, or Quality mode manually rather than the game devs deciding what's appropriate.

As expected, the RTX 4090 mostly breezes along with just over 60 fps, and it's the only card to break that barrier. Even 30 fps is a difficult target without upscaling, with the RTX 3080 Ti and above proving sufficient. An interesting aside is that the two RTX 3080 cards, one with 10GB on a 320-bit interface and the other with 12GB on a 384-bit interface, end up with very similar performance. Minecraft RTX at least doesn't seem to strongly depend on memory bandwidth based on those results. Meanwhile, none of AMD's GPUs can break 30 fps, and the fastest RX 6000-series cards fall in the high teens.

With DLSS now using Performance mode and 4X upscaling, you get a huge boost to framerates by enabling upscaling. The RTX 2060 nearly triples its performance and is mostly playable at just a hair under 30 fps. RTX 3080 improves by 160%, with its upscaled performance passing the native RTX 4090. And the RTX 4090 effectively doubles its performance.

While you might think 4X upscaling will leave noticeable artifacts, at 4K and with a game like Minecraft — meaning it's very blocky by default, and even the text is intentionally pixelated looking — the resulting output still looks almost as good as native rendering. Considering how demanding 4K would otherwise be, DLSS is basically a prerequisite to get good performance out of that resolution in Minecraft. And at least the frames rendered in Minecraft are true frames and not the DLSS 3 Frame Generation frames.

Ray Tracing vs. Path Tracing 

That wraps up our performance test, but to understand why Minecraft RTX is so demanding, we need to briefly describe how it differs from other RTX enabled games. Nvidia says that Minecraft RTX uses 'path tracing,' similar to what it did with Quake II RTX, where most other games like Control and Cyberpunk 2077 only use 'ray tracing.' For anyone who fully understands the difference between ray tracing and path tracing, you probably just rolled your eyes hard. That's because Nvidia has co-opted the terms to mean something new and different.

In short, Nvidia's 'path tracing' in Minecraft RTX just means doing more ray tracing calculations—bouncing more rays—to provide more effects and a higher quality result. Actually, Nvidia also now lists Portal RTX as the only game with "full RT" if you check out Nvidia's list of RTX enabled games (which can mean games with DXR, DLSS, DLAA, or some combination of those).

Path tracing as used in Hollywood films typically means casting a number of rays per pixel into a scene, finding an intersection with an object and determining the base color, then randomly bouncing more rays from that point in new directions. Repeat the process for the new rays until each ray reaches a maximum depth (number of bounces) or fails to intersect with anything, then accumulate the resulting colors and average them out to get a final result.

That's the simplified version, but the important bit is that it can take thousands of rays per pixel to get an accurate result. A random sampling of rays at each bounce quickly scales the total number of rays needed in an exponential fashion. You do however get a 'fast' result with only a few hundred rays per pixel—this early result is usually grainy and gets refined as additional rays are calculated.

Ray tracing is similar except it doesn't have the random sampling and integral sums of a bunch of rays. Where you might need tens of thousands of samples per pixel to get a good 'final' rendering result with path tracing, ray tracing focuses on calculating rays at each bounce toward other objects and light sources. It's still complex, and often ray and path tracing are used as interchangeable terms for 3D rendering, but there are some technical differences and advantages to each approach.

Doing full real-time path tracing or ray tracing in a game isn't practical yet, especially not with more complex game environments. Instead, games with ray tracing currently use a hybrid rendering approach. Most of the rendering work is still done via traditional rasterization, which our modern GPUs are very good at, and only certain additional effects get ray traced.

Typical ray traced effects include reflections, shadows, ambient occlusion, global illumination, diffuse lighting, and caustics, and each effect requires additional rays. Most games with ray tracing support only use one or two effects, though a few might do three or four. Control uses ray tracing for reflections, contact shadows and diffused lighting, while Hogwarts Legacy can use ray tracing for reflections, shadows, and ambient occlusion.

Compare that with Minecraft RTX, where you get ray tracing for reflections, shadows, global illumination, refraction, ambient occlusion, emissive lighting, atmospheric effects, and more. That's a lot more rays, though it's still not path tracing in the traditional sense. Unless you subscribe to the 'more rays' being synonymous with 'path tracing' mindset.

(Side note: the SEUS PTGI tool is a different take on path tracing. It depends on Minecraft's use of voxels to help speed up what would otherwise be complex calculations, and it can run on GTX and AMD hardware. There's also the RTGI ReShade tool that uses screen space calculations and other clever tricks to approximate path tracing, but it lacks access to much of the data that would be required to do 'proper' path tracing.) 

Closing Thoughts 

There's a lot to digest with Minecraft and ray tracing. The game in its 'normal' mode can run on everything from potatoes—along with smartphones and tablets—up through the beefiest of PCs. Cranking up the render distance in the past could cause a few oddities and put more of a load on the CPU, but extreme framerates aren't really needed. Minecraft RTX significantly ups the ante in terms of the GPU hardware requirements.

The lighting, reflections, and other graphical enhancements definitely make a big difference, both in performance as well as visuals. Minecraft has never looked so pretty! Nor has it looked so dark when you're mining underground — bring lots of torches. The core survival and exploration gameplay hasn't changed, of course, but makers who spend their time building intricate Minecraft worlds now have a host of new tools available.

It's now almost three years since Minecraft RTX first became available to the public, and a lot has changed in that time. Where formerly only RTX 20-series GPUs were available, we now have the RTX 30- and 40-series cards, AMD's RX 6000- and 7000-series GPUs, and even the handful of Intel Arc offerings. As you can see from our testing, Nvidia's GPUs still generally provide the best experience, and DLSS delivers another level of performance that currently can't be matched. It's too bad the game hasn't been updated with FSR 2 or XeSS support.

When Apple moved to its own silicon, its Mac computers lost a feature called Boot Camp, which let people dual boot between macOS and other operating systems, like Windows. Now, Microsoft is officially sanctioning running Windows 11 on Macs with M1 and M2 processors. Today, it published a new support page entitled "Options for using Windows 11 with Mac computers with Apple M1 and M2 chips."

The options to do so aren't necessarily new — Microsoft is pushing Windows 365 Cloud PCs or by using Parallels Desktop 18 — but it's the first time we've seen Microsoft provide Windows guidance to Mac users since Boot Camp went away. In general, it seems to be pointing it only toward business users.

Windows 365 Cloud PCs are currently available in business and enterprise plans, starting at $31 per user per month. In this case, you can stream Windows 11 to a Mac (or any device) through the software-as-a-service. The benefit of this is that Windows 365 offers nested virtualization, allowing you to run emulators and get full compatibility with features that require virtualization, like running Android apps from the Amazon app store.

Using Parallels Desktop 18 may be more accessible for most people, but the standard edition for home and student use starts at $99.99 a year, with professional and business edition subscription services at $119.99 and $149.99 per year. Microsoft says it's authorized for running Arm versions of Windows 11 Pro and Windows 11 Enterprise.

Windows 11 on Arm has many of the same restrictions on a Mac as it does on PCs designed for Windows. "The Arm version of Windows 11 has limitations that can impact your ability to use various types of hardware, games, and apps, including those that rely on DirectX 12 or OpenGL3.3 or greater," the support page reads. In other words, don't expect to use your M2 Max's GPU cores for gaming with Windows 11 on Arm.

But going the Parallels route also doesn't allow for nested virtualization, so you won't be able to run the Windows Subsystem for Android or Linux (to run Android apps from the Amazon app store or run a GNU environment in Windows), or the Windows Sandbox or Virtualization-based security (VBS) to separate some memory from the OS.

Additionally, 32-bit Arm apps from the Microsoft Store in Windows don't work on Macs with M1 and M2 processors (Apple stopped supporting 32-bit apps with macOS Catalina). But Microsoft says that 32-bit Arm apps "are in the process of being deprecated for all Arm versions of Windows." Ideally, you would run 64-bit Arm apps on Mac, but you can also emulate x64 or x86 apps.

We don't expect that this news means that Microsoft will also make the Arm versions of Windows 11 more available to the public. Microsoft explicitly told me that "Parallels Desktop is currently the only authorized option that supports Windows 11 in VMs on Macs." Parallels has been used to run Windows 11 for quite some time on M-series Macs, however, and installing Windows 11 is a simple part of the app's setup.

It isn't clear why Microsoft is now publishing this page, other than to make it explicit which methods of getting Windows 11 on a Mac are officially sanctioned. But hey, if it helps someone find what they're looking for, we're all for it. It's not as exciting as the prospect of dual-booting macOS and Windows 11 on Arm, though. We can only dream.

Last week, a collector put up a rare 3dfx Voodoo 5 6000 graphics card for auction on eBay. This was a fully functional prototype, with engineering work from former 3dfx employee Jindrich Semenec to get the card working with 8xFSAA. Bidding quickly eclipsed $12,000, and the final sale price ended up at $15,000. Here's what we know for now.

The seller purchased the card originally from a friend, as it was something of a holy grail for graphics card collectors. Over time, however, they felt there wasn't quite as much emotional attachment to the Voodoo 5 6000, and so they decided to put the card up for auction so that someone else could enjoy it.

Getting the card running in a modern PC isn't possible, as it uses the long since discontinued AGP 4X interface. AGP rapidly dwindled in use with the advent of PCI Express in 2003, though the Voodoo 5 series of cards first appeared in 2000, and 3dfx filed for bankruptcy in 2002. But if you have an old Pentium III or similar rig still kicking around, you could potentially still run the Voodoo 5 6000 in it.

As for specs, the card uses four of the Napalm 30 architecture VSA-100 chips clocked at 166 MHz. Each chip features two pixel shaders and no vertex shaders (as those weren't around in 2000). There are also two texturing units and two render outputs (ROPs) per chip, and the card used SLI — Scan Line Interleaving, not the Nvidia Scalable Link Interface — to put all four chips to use. Each GPU was also paired with 32MB of SDR memory running at the same 166 MHz as the GPU core, for a total of 128MB.

The VSA-100 chips were fabricated on a 250nm process, with a size of around 112mm^2 and just 14 million transistors per chip. Even the lowest of integrated graphics solutions these days will use far more transistors, though a lot of those go toward things like AVC, HEVC, and/or AV1 encoding/decoding support, in addition to the various graphics duties. With four chips total, the Voodoo 5 6000 was rated at up to 60W of power use — try not to let it go to your head!

While the card could pull most of the needed power over the AGP connection, it does feature a barrel-style external power adapter. Note also that the only output available is a single VGA connector, though that's no surprise considering the age of the hardware.

Just how fast would the Voodoo 5 6000 be compared to modern graphics cards? Even if we ignore the lack of support for newer APIs (DirectX 6 was the maximum at the time), it's not going to set any records. Four pixel shaders at 166 MHz yields a theoretical performance of 2.66 Giga-ops per second. That would put it roughly on par with the Radeon 9700 that came out in 2002, according to our GPU benchmarks hierarchy.

Or maybe we need to add in some additional "ops" for the ROPS and TMUs — honestly, it's been so long since we moved to unified shaders that I hardly remember how they might compare with newer GPUs. Suffice it to say you won't be running Windows 11 or any remotely recent games on the Voodoo 5 6000. But that's assuming you even want to, as more likely than not this will end up enshrined on a shelf rather than sitting in a PC that can actually be used.

Whatever the case, this is certainly a rare piece of gaming history. 3dfx Interactive was one of the driving forces behind early 3D games on PC, and it also powered some arcade machines in the late 90s. RIP, 3dfx and Voodoo.

We're trying to get additional details on who purchased the card, so if you're one of the two involved parties and would like to talk, drop me a note.

MTT S80 benchmarks have revealed that China's homemade graphics card cannot compete with today's best graphics cards. The performance is lacking, but the PCIe 5.0 gaming graphics card doesn't play nice with many games, some going as far back as 2013.

The MTT S80 is the first Chinese homegrown graphics card to support the DirectX API, representing a giant leap in domestic gaming. When Moore Threads announced the MTT S80, the chipmaker worked with up to 60 games. We would later find out that the only catch is that only 11 out of the 60 titles are on the official support list, whereas MTT S80 could run the remaining titles, but performance would be an issue. The MTT S80 supports both DirectX 9 and DirectX 11; however, there's still a long road ahead, longer for the latter than the former.

PC enthusiast Löschzwerg, who kindly shared the MTT S80 benchmarks, has recounted the struggles that the graphics card had to go through with some of the titles. According to the reviewer, old titles, such as Dota 2 and Tomb Raider, were non-functional. The former crashed instantly when a game was started, and the latter refused to start.

Counter-Strike: Global Offensive (CS: GO) was playable on the MTT S80. Löschzwerg noted the graphics driver hampered the MTT S80's performance because sometimes GPU utilization didn't even hit 50%. The China-made graphics card's performance on Elder Scrolls V: Skyrim was mediocre, comparing the low to very high preset. The former was a bit better, but it wasn't anything to brag about.

Löschzwerg highlighted that the geometry performance of the MTT S80 is abysmal. According to his observations, frame rates dipped substantially on CS: GO and DUST II in scenes with abundantly visible geometry. Tessellation was another one of the MTT S80's flaws. Due to the early maturity of the driver (200.2), tessellation does more harm than good. Benchmarks and games crashed when the tessellation was active, so the only stopgap solution was to disable it altogether.

Driver performance and compatibility get the best of even the biggest chipmakers, such as Intel. Unfortunately, the first Arc driver made Alchemist graphics cards underperform on older APIs, such as DirectX 9. Fast-forward a few months, and DirectX 9 performance has improved as much as 43% since the initial launch. It just serves to demonstrate that it's enough to have a good product without an equally good driver.

In comparison, Moore Threads is a rookie and doesn't have the same resources as Nvidia, AMD, or Intel. As a result, it'll probably take the Chinese chipmaker a long time to iron out the bugs and have a fully functional driver that doesn't hold the hardware back.

Microsoft released some Patch Tuesday fixes last November but has only today admitted they have been causing errors with Intel graphics hardware running DirectX or Direct3D applications. The issues were introduced with KB5019980 / KB5019961 for Windows 11 users and KB5019959 for Windows 10 users, notes Neowin. Microsoft’s KB releases target known Windows quirks and wrinkles, yet this isn’t the first time that they have come with their own crop of issues.

The Microsoft-induced DirectX or Direct3D application bugs are described as intermittent, which might be one of the reasons why they persisted for so long without even an official workaround.  Microsoft said, "After installing KB5019980, you might receive an error with apphelp.dll on Windows devices using Intel graphics drivers with versions 26.20.100.7463 up to 30.0.101.1190.” Additionally, it explained, “This issue might happen intermittently and affects apps which use DirectX or Direct3D to render part or all of their content.” It appears to be the case that apphelp is being an app hindrance.

Looking on the bright side, Microsoft now admits and acknowledges this apphelp.dll bug, and while it says that a resolution will be in an upcoming update, there is a workaround that affected users can try.

Microsoft says the apphelp.dll bug affecting DirectX and Direct3D titles can usually be sidestepped by applying the latest Intel drivers. It recommends users “install an Intel graphics driver with a later version than 30.0.101.1190.” That driver was released back in December 2021, so we are pretty sure Intel Arc discrete graphics users aren’t clinging to this outdated software. However, depending on processor generation, some PCs might not offer newer graphics drivers for iGPUs.

On the topic of Intel graphics drivers, we have been pleased to report the sterling progress the new entrant into the PC dGPU market has made in recent weeks/months. Only last week, Intel released a new graphics driver providing a fantastic DX9 performance uplift – targeting an area where the launch driver was very weak.

China doesn't have many homebrew graphics cards, so expectations were high when graphics card manufacturer Moore Threads revealed the MTT S80. It's hard to tell whether the MTT S80 has what it takes to compete with the best graphics cards, but hopefully, graphics card collector Löschzwerg's latest benchmarks can provide some insight.

While Moore Threads is green behind the ears, the company has strong leadership. Former Nvidia global VP and China GM Zhang Jianzhong founded Moore Threads in 2020, so the Chinese fabless semiconductor company is a newcomer to the graphics card game. Besides being China's domestic graphics card, the MTT S80 has garnered a fair amount of hype outside the country since it's the first PCIe 5.0 gaming graphics card to hit the market. Nvidia's GeForce RTX 40-series (Ada Lovelace) and AMD's Radeon RX 7000-series (RDNA 3) products are still on PCIe 4.0.

The MTT S80 is the successor to the MTT S60 and still leverages the same MT Unified System Architecture (MUSA) architecture. It supports modern APIs, including CUDA, DirectX, OpenCL, OpenGL, and Vulkan. The Chunxiao GPU, manufactured under the 12nm process node, powers the MTT S80. Like Nvidia and AMD, the MTT S80 embraces AV1 encoding support in addition to other popular formats, such as H.264, H.265, and VP9.

The MTT S80, which supports PCIe 5.0 x16, wields 4,096 MUSA cores operating at 1.8 GHz to offer up to 14.4 TFLOPs of FP32 performance. This places the Chinese graphics card between the GeForce RTX 3060 (12.7 TFLOPs) and GeForce RTX 3060 Ti (16.2 TFLOPs) or, alternatively, the Radeon RX 6750 XT (13.3 TFLOPs) and Radeon RX 6800 (16.2 TFLOPs). Moore Threads outfits the MTT S80 with 16GB of 14 Gbps GDDR6 memory across a 256-bit memory interface. This arrangement is suitable for a maximum theoretical memory bandwidth of up to 448 GBps, on equal footing with the GeForce RTX 3060 Ti.

The MTT S80 features your typical dual-slot design with a triple-fan cooling solution. Of course, it's a gaming graphics card; some RGB eye candy is inevitable. The graphics card utilizes an 8-pin EPS connector, yes, the kind that you find on motherboards. It has a 255W TBP (total board power), and one EPS connector supplies up to 300W. The MTT S80 has the same outputs as Nvidia's flagship GeForce RTX 4090. In addition, you receive three DisplayPort1.4a outputs and one HDMI 2.1 port to support up to four 8K displays.

To say that benchmarking the MTT S80 was difficult is an understatement. Löschzwerg emphasized that the graphics card's performance and driver were wonky. The GPU utilization wasn't optimal, and the graphics card rarely showed its full potential, suggesting a lack of driver optimization. Tessellation doesn't work with the current driver and causes crashes on Unigine Heaven, 3DMark 11, and Final Fantasy XIV: Endwalker. Surprisingly, Resizable BAR works fine.

Moore Threads MTT S80 Benchmarks

The MTT S80 delivered between 16 FPS and 31 FPS (actual gameplay) in Crysis at 1080p on the high preset with DirectX 9. In F.E.A.R., with 1080p and maximum preset, the graphics card ran the game between 54 FPS and 186 FPS. Meanwhile, the average framerate for Half-Life 2: Lost Coast at 1080p and maximum settings was 132.2 FPS.

The MTT S80 finished the Final Fantasy XIV: Endwalker benchmark with a score of 4,312 points. Unfortunately, the reviewer used custom settings so we couldn't compare it to our results. However, the score slides between the 4,000 to 5,999 range, equivalent to standard performance, meaning the graphics card has what it takes to run the game on default settings. The Arc A770 scored 18,674 points in the same benchmark, outpacing the MTT S80 by 100%.

The results revealed that the Arc A770 outperformed the MTT S80 by over 170% in 3DMark06 on both presets. Intel's graphics card also delivered 155% higher performance in 3DMark03. With the more demanding preset, the Arc A770 still beat the MTT S80 by a 54% margin.

The MTT S80 scored 551 points in Unigine Heaven at 1080p 8xAA on the ultra preset. Another Twitter user provided the scores for the GeForce GTX 680 and the Radeon HD 7950 Boost. The former had a 114% higher score, while the latter showed a 101% better score than the MTT S80.

The power metrics are a real shocker, though. The test system, which comprises the Core i5-10400, ASRock B560M-HDV, and 32GB (2x16GB) of DDR4-3200 memory, idles around 22W. With the MTT S80, the idle power was about 131W, conveying that the graphics card consumes 109W while idling. On the other hand, the peak system power consumption was 315W, so the MTT S80 consumed 293W. Again, the values are ridiculously high. For example, even the enthusiast-grade GeForce RTX 4070 Ti is a 285W graphics card.

With the driver's current state, the MTT S80 is potentially leaving some performance on the table. Officially, the graphics card supports around 20 DirectX games, but performance is a hit or miss. The graphics card performs better on DirectX 9 titles than on DirectX 11. However, there's still a long way to go regarding game compatibility.

The MTT S80 retails for 2,999 yuan or $442.65 on JD.com, a popular Chinese online retailer. Unfortunately, it won't be a suitable option since gamers can still purchase Intel, AMD, or Nvidia graphics cards in China. However, if restrictions on exports to China get harsh, the MTT S80 could gain relevance.

The Intel Arc Alchemist architecture brought a third competitor to the best graphics cards, and while it can't top the GPU benchmarks in terms of performance, there's certainly a strong value proposition. At the same time, there are areas where the drivers still need tuning — Minecraft with ray tracing being one example that comes immediately to mind since it's one of the games in our standard test suite. But Intel hopes to encourage further adoption with the latest announcement of a $40 price cut to the Intel Arc A750, bringing it down to just $250.

It's difficult to say exactly how many people have purchased Intel Arc graphics cards — mobile or desktop — since they first became available in the spring of 2023. I've seen claims that Intel dedicated GPUs accounted for up to 4% of total sales in Q4 2022, but so far Arc GPUs don't show up as individual entries on the Steam Hardware Survey (which will be updated with January data shortly, so maybe that will change). Certainly, dropping the price of the A750 by 14% can't hurt.

It's not just about lowering the price, either. Since launching with the 3490 drivers back in October 2022, Intel has delivered three WHQL drivers and at least four beta drivers. The latest beta, version 4090, became available last week, and improving performance and compatibility has been a key target for all of the driver updates.

DirectX 9 performance, an area that Intel hadn't really focused on prior to the Arc launch, has been one of the biggest benefactors of the newer drivers. Intel claims that, across a test suite of thirteen games, average framerates at 1080p have improved by 43%, and 99th percentile fps has improved by 60%. At 1440p, the average fps increased by 35% while 99th percentile fps improved by 52%.

Granted, the test suite for DX9 games isn't so much about making games that ran poorly suddenly run well. The worst performing of the suite, Stellaris, looks to have performed at about 75 fps with the launch drivers, whereas it's now getting more like 130 fps. And Half-Life 2 went from just under 400 fps to about 600 fps. Even so, the overall experience has improved, and framerate consistency and frame times are much more stable.

It's also interesting that Intel continues to show the Arc A750 as an RTX 3060 competitor, mostly ignoring (in charts) AMD's own RX 6600. That's probably because AMD has a much stronger value proposition, with the RX 6600 regularly selling for $225, give or take. Our testing puts it slightly below the A750 (and RTX 3060), but the price cut does make up for the higher power use on Intel's GPU.

Intel is also continuing to push its XeSS (Xe Super Sampling) AI upscaling algorithm as an alternative to Nvidia's DLSS and AMD's FSR technologies. The adoption rate isn't nearly as high, but considering how new Intel is to the dedicated GPU arena, getting 35 games to support XeSS in the first six months or so is pretty decent.

Another feature of the Arc GPUs that's more than just "pretty decent" is the video encoding and decoding support. Arc was the first modern GPU to offer full AV1 support, and the quality of the Quick Sync Video encoding goes head to head with Nvidia's best (with AMD trailing on previous generation GPUs, though we still need to look at the latest RDNA 3 chips).

But it's not all sunshine and flowers. Our own testing of Bright Memory Infinite (using the Bright Memory Infinite benchmark on Steam) and Minecraft shows there's still room for improvement. Another interesting aspect of the Arc GPUs that we've discovered is that, using a Samsung Odyssey Neo G8 32-inch monitor, the DisplayPort connection can only run at up to 4K and 120 Hz, while Nvidia's RTX 20-series and later (using DP1.4a) all support 4K and 240 Hz via Display Stream Compression.

Ultimately, lowering the price of the A750 by $40 probably won't change the minds of millions of gamers, but it does make the overall package more attractive. Intel has also added Nightingale and The Settlers: New Allies to the software bundle for anyone who purchases a new Arc graphics card or system equipped with an Arc GPU. As we've noted before, Intel may not have the fastest cards on the planet, but the value proposition is certainly worth considering.

Which does bring up an interesting question: What's happened with the Arc A580? That's supposed to have the same 8GB of GDDR6 as the A750, but with 24 Xe-cores instead of 28 Xe-cores (3,072 shaders vs. 3,584 shaders on the A750). It also has a lower TBP of 175W compared to 225W and a Game clock of 1700 MHz, or at least that's the theory. With the new price on the A750, the space for an A580 continues to shrink, but maybe Intel could still release something in the $199–$219 range. We're still waiting...

The full slide deck from Intel is included below, for reference.

AMD’s venerable Polaris architecture graphics cards are showing their age, with one of the hottest new titles simply refusing to run. The PC game on everyone’s lips in January 2023, Forspoken, has DirectX 12 feature-level 12_1 API requirements that the stoic Radeon RX 400 and RX 500 series graphics cards can’t satisfy.

That the aging AMD Polaris graphics cards have finally started to be too old to party like they used to was highlighted by Redditor xCurio. They contrasted the inability of Polaris to run the PC game du jour, Forspoken, when the even older GeForce GTX 960 or GTX 970 cards (Maxwell architecture) can be used.

If we check out the published minimum system requirements for Forspoken, it can be seen that gamers are expected to use a system with at least an AMD Radeon RX 5500 XT 8GB in the graphics card section or Nvidia GeForce GTX 1060 6GB. These are RDNA 1 and Pascal architecture graphics cards, respectively. Despite the clear minimum recommendation for an RDNA 1 or better GPU, it is still a surprise that the older Polaris isn't even able to run Forspoken.

AMD released a new Adrenalin graphics driver with optimizations ready to support Forspoken just a couple of days ago, but this driver is for use with its Radeon RX 7000 (RDNA 3) graphics cards only. Curiously, the red team hasn't provided any driver updates for earlier Radeon RX series GPUs since  December 8, while the RX 7000 series have received four driver updates since that time. So if you have a Vega, RDNA 1 or RDNA 2 architecture graphics card, we hope Forspoken won't be an unpleasant experience due to technical issues. We're also hopeful that AMD has a new Adrenalin graphics driver complete or almost ready in the pipeline for pre-RDNA 3 architecture graphics card owners.

In the meantime, Forspoken isn’t receiving the warmest reception from gamers. For example, if you check the Steam user reviews section, many recommend holding off for a few weeks or months until patches have ironed out the launch wrinkles. Another benefit of this strategy is that it could also receive a price cut. Pleasingly, there is a free demo to try if you are curious about this epic action RPG adventure game.

We now have three major players vying to be the best graphics cards: AMD, Nvidia, and Intel. The two incumbents recently released their next-generation GPU architectures, with Nvidia Ada Lovelace taking on AMD RDNA 3. The Intel Arc Alchemist architecture also came out last year, but you can certainly make the argument that it was really designed to compete with Nvidia Ampere and AMD RDNA 2. Beyond the hardware, though, software and drivers play a critical role in extracting maximum performance from the GPU, and drivers are becoming increasingly complex.

Consider all the APIs that need to be supported: DirectX 9/10/11/12, OpenGL, Vulkan, OpenCL, and others besides. Then you have potentially multiple generations of hardware — Nvidia's latest drivers support five different architectures while AMD's drivers support ... well, AMD's a bit of a special case right now, as one set of drivers only supports the latest RDNA 3 GPUs while the other set supports RDNA 2, RDNA, Vega, and Polaris, but we're told the two will get a unified driver in the future.

Intel in contrast is at an advantage, since it doesn't have a huge back catalog of GPUs to support... unless you count integrated graphics. And Intel does support 11th Gen Tiger Lake, 12th Gen Alder Lake, and 13th Gen Raptor Lake integrated graphics with its drivers. That's a bit interesting, because there's a huge gulf in performance between many of those integrated solutions and the dedicated Arc GPUs.

One thing we've noticed over the years has been the steadily bloating size of GPU driver downloads. Which got us thinking about the current status quo. You might think Nvidia's drivers would be larger, since it also has some extra stuff like CUDA and DLSS that it needs to support. You'd be partially correct. At present, here's how the download sizes stack up:

We've become accustomed to seeing larger download sizes on Nvidia's drivers, though they did shrink quite a bit back when Nvidia killed off 3D Vision support (RIP). Besides the drivers, there's also the included GeForce Experience app — a 124.8 MiB download if you grab it separately, but it's normally packaged with every Nvidia driver.

The newer AMD drivers strike us as somewhat odd, however. Why are the drivers that only have to support two GPUs 8% larger than the drivers that support probably over 100 different GPUs? Maybe the unified AMD drivers won't be much larger and it's just a matter of tying a few remaining pieces together.

But AMD and Nvidia have nothing on Intel when it comes to driver bloat. 1.21 GB? Great Scott! How could they have been so careless? But seriously, you have to wonder what exactly is taking up so much space. (Update, 3/24/2023: Intel managed to slash the driver download in size in half within two months of this article going up. Interesting timing!)

Granted, Intel is newer to this than AMD and Nvidia. It also has tried to stuff in a bunch of extra features, including XeSS support and things like Smooth Sync. But AMD and Nvidia have that sort of stuff as well. More to the point, we can't help but wonder if the larger, presumably unoptimized file sizes might also correlate with unoptimized code that's full of extra cruft. Also, will the drivers get even larger when Battlemage arrives?

Thank goodness the best SSDs have become pretty affordable, as these days you can grab a fast 1TB NVMe drive for around $85, or alternatively a 2TB SATA drive. Unpacking and installing the drivers will of course consume even more space, but that's nothing compared to 100GB and larger game install sizes — unless you have a data cap and like to download drivers on a regular basis. Bon appétit!

Innosilicon showcased its Fantasy 2 graphics card at the 17th China Chip Excellence Awards Ceremony, where the consumer-targeted GPU scooped up the China Chip Excellent Technology Innovation Product Award. 

Thanks to this industry event we now have some great new images of the graphics card (H/T to Twitter user @Loschzwerg_3DC). There's still no word on consumer availability of this GPU — even though it was supposed to become available last August.

If you haven't been following the Innosilicon Fantasy 2 graphics card news closely, it's a PowerVR architecture consumer product, claimed to be capable of "4K-level high- performance rendering," while using very little power.

Don't expect too much in terms of performance from this product when it gets into the hands of reviewers, especially from the fanless model in the new images. Its 1.5 TFLOPs FP32 performance and 10 TOPS (Int8) computing power aren't exactly the stuff of gamers' dreams. That said, it's good to see PowerVR back in desktop PC products: its ultra-low power consumption (5-15W), as well as its broad graphics API (OpenGL, Vulkan, and DirectX) and OS support (Windows, Linux, Android), are particularly welcome.

Innosilicon claims that the Fantasy 2 graphics card is capable enough for "daily needs" such as office use, entertainment, CAD, and other applications. But when it says the GPU "performs well in games," we suspect it's talking about casual 2D gaming titles rather than the latest AAA 3D PC releases.

Those with the thirst for PowerVR power might be better served by the Fantasy 1, which is supposed to target GeForce RTX 3060 grade performance (take that with a big pinch of salt). This card has been confirmed by ImgTec to feature the BXT-32-1024 GPU, delivering 5 TFLOPs FP32 performance, and coming kitted out with 16GB of GDDR6X on a 128-bit interface. 

Muddying the already opaque waters regarding these elusive GPUs, there are also reported to be two variants of the Fantasy 1, with the Type A (outlined above) using a single BXT-32-1024 GPU, and the Type B version featuring a dual-GPU-equipped PCB.

Chinese tech news sites, forums, and social media have almost no hands-on information regarding any of these Innosilicon Fantasy graphics cards, even though we're now firmly into 2023. As it stands with the phantom launches and no-shows, it doesn't look likely that the Fantasy 2 will make it onto our highly competitive best graphics cards list.

Just yesterday, Intel released a massive driver update to its Arc Alchemist GPUs, resulting in a near 2x FPS (frames per second) increase in DX9-based titles. Now, according to a report by GamingOnLinux, we know how Intel was able to accomplish this task in some games. Intel has added another translation layer to its latest graphics driver, featuring Steam’s DXVK translation layer. This is the same translation layer Steam uses in its Proton API to translate Windows games to Linux (like we saw on the Steam Deck), allowing Intel to translate DX9 code to the open-source Vulkan API.

According to Intel’s blog post from yesterday, Intel is apparently only using the DXVK translation layer in some cases; its driver will take a hybrid approach of opportunistically utilizing a combination of API techniques that take advantage of translation layers using one or more modern API implementations. As a result, Intel won’t be using DXVK for the entirety of its DX9 processing, and will only use DXVK when it can provide better performance over Intel’s DX12 emulation technique.

Intel never actually stated it was using DXVK in its official blog post, but thanks to Gaming On Linux, we now know DXVK translation is what Intel is referring to, based on a readme document Intel published on the same topic. Unfortunately, we do not know how effective DXVK actually is with Intel’s implementation, but it must be a very effective solution for Intel to implement the translation layer into its latest driver at all.

Intel’s Arc GPUs - and its associated Xe integrated graphics, no longer feature native DX9 support. Instead Intel has opted to rely solely on translation layers to replicate native rendering. One such example is Intel’s utilization of the Microsoft D3D9On12 mapping layer, which translates DX9 commands to DX12.

Intel’s reliance on translation layers may not be ideal, but it actually gives the company a shortcut in garnering good DX9 GPU performance quickly. Translation layers allow Intel to use DX12 optimizations for both modern DX12 titles and for older DX9 games, which severely cuts down the development time needed to optimize both new and old APIs. This was a necessary move for Intel, because it needed a way to compete with Nvidia and AMD, which have nearly twenty years of experience developing DX9 drivers for discrete GPU hardware.

The awesome part about DXVK is that Intel no longer needs to rely on Microsoft’s DX12 emulation layer to play DX9 games on Arc. With DXVK, Intel can now run DX9 games on other operating systems such as Linux, and gain additional performance benefits that might not be available in Microsoft’s emulator. 

Intel has made significant progress with DirectX 9 performance on its Arc Alchemist graphics cards. The latest 31.0.101.3959 driver brings up to 1.8X higher average FPS and up to 2.3X improved 99th percentile FPS on the Arc A770, one of the best graphics cards on the market.

When Arc first came out, it was already public knowledge that the Intel graphics cards perform poorly on games with legacy APIs. Nonetheless, Intel had pledged to improve performance over time, and the chipmaker has delivered on its promise. As a result, the new Arc display driver offers a substantial performance increase in DirectX 9 titles, including CS:GO, League of Legends, Starcraft 2, Guild Wars 2, Payday 2, and Stellaris.

Compared to the previous driver (31.0.101.3490), the new v3953 driver delivered 1.79X higher average FPS in CS:GO at 1080p with high settings. It's a tremendous jump in framerates when comparing 177 FPS on the old driver versus 318 FPS on the new driver. At 1440p with ultra settings, we're looking at 1.77X better performance. The v3953 driver doesn't just make the framerates jump through the roof. It also ensures smoother gameplay. On the v3953 driver, the Arc A770 exhibited 2.26X higher 99th percentile FPS at 1080p and 2.15X at 1440p.

Intel's test system was pretty standard for a high-end gaming PC. It consisted of a Core i9-12900K (Alder Lake) processor, 32GB (2x16GB) of Dominator Platinum DDR5-5600 C36 memory, and an Asus ROG Maximus Z690 Hero motherboard. Intel used the Arc A770 for testing; however, the Arc A750, Arc A380, or their mobile counterparts should enjoy similar performance benefits from the v3959 driver.

Intel explained that modern APIs, such as DirectX 12 and Vulkan, are still the main focus; however, the company's opting to use a hybrid approach. As a result, there will be occasions when a title runs on the native DirectX implementation and other circumstances where it'll leverage translation layers to go from DirectX 9 to a modern API.

It's incredible what the Intel engineers could achieve with the Arc driver in less than two months since the Arc graphics cards debuted. Nonetheless, Intel has promised further improvements for titles on legacy APIs, meaning it could just be a matter of time before we see enhanced performance for games based on DirectX 10 or DirectX 11.

If you're looking for a decent budget to midrange graphics card, this Asus RTX 2060 Cyber Monday deal on Amazon is about as good as we're likely to see. Priced at $179, down from a previous $279 and beating the former best price on an RTX 2060 of $219 by $40, this one will likely go fast. Note that you may have to click "other sellers" to get this price — it's not always the top option.

If you check our GPU benchmarks hierarchy, you'll see that the RTX 2060 performs a few percent faster than the RTX 3050 — and that applies to ray tracing games as well as rasterized games. That's largely thanks to its having a 192-bit memory interface and 336 GB/s of bandwidth, compared to a 128-bit interface and just 224 GB/s on the 3050.

It's not as fast as AMD's RX 6600 in rasterization games, though it does rank ahead of it for DXR (DirectX Raytracing) performance. Now that the RX 6600 $189 deal is gone, this is about as good as it gets for a "budget" GPU, and it will easily beat cards like the RX 6500 XT, GTX 1650, or Arc A380.

The RTX 2060 has had an interesting run of late. After EVGA announced it was exiting the GPU market, we saw the RTX 2060 drop down to $219 for a short time. Since then, prices have climbed back into the upper $200s, but for a time the RTX 2060 was the best value on an Nvidia GPU. It can now stake that claim again, as it's about $75 less than the RTX 3050 it still outperforms, and matches the lowest current price on an GTX 1660 — the GDDR5 non-Super model.

Turing has been out since 2018, and the RTX 2060 originally launched in early 2019. Four years later, we'd normally expect supply to be long gone, but the GPU shortages of the past two years have kept it in production. With cryptocurrency mining (on GPUs) now effectively dead for the time being, this should finally represent the end of the line for the venerable TU106 GPUs and the Turing Architecture.

Nvidia has released a new Game Ready Driver — version 526.98, which is jam-packed with feature updates, including support for the RTX 4080, as well as G-Sync and V-Sync support for DLSS 3 frame generation, HDR10+ Gaming, and RTX I/O DirectStorage.

Arguably the most anticipated feature — aside from RTX 4080 support — is official support for V-Sync and Nvidia G-Sync with DLSS 3 frame generation. Support for these two frame-syncing technologies was noticeably absent during the initial release of DLSS 3, but it's finally here with the new 526.98 driver — giving gamers a tear-free experience with frame generation enabled. 

That said, support isn't as liberal as we hoped. For now, it's limited to the Nvidia driver version of V-Sync, rather than in-game implementation. This means you have to go to the Nvidia Control Panel and manually enable V-Sync within the 3D settings profile the game you're playing (or enable it globally). 

G-Sync support is more straightforward, and only requires the tech to be enabled in the control panel once. DLSS 3 support of G-Sync works for both G-Sync Compatible and G-Sync Ultimate displays. If you still have tearing problems, you can enable V-Sync (manually) in combination with G-Sync.

For those worried about increased input lag, Nvidia says DLSS 3's combination with Nvidia Reflex technology will help decrease input lag even with V-Sync enabled. However we suspect you'll still see slightly higher input lag due to the way V-Sync works.

RTX I/O GPU Decompression Update

The next major feature is support for Microsoft's DirectStorage 1.1 API, in the form of RTX I/O. This feature allows RTX GPUs to comply with the DirectStorage standard, which lets them decompress and load game assets directly from high speed storage solutions with aid from the CPU. This will significantly decrease game load times in titles that support the technology.

RTX I/O's feature set will also extend past Microsoft's DirectStorage API in the form of Vulkan extensions. These extensions will give Vulkan-based titles access to the same performance enhancements as DirectStorage while bypassing Microsoft requirements such as Windows 10, 11, and DirectX 12.

HDR10+ Gaming Standard

Another gaming focused update in driver 526.98 is support for the HDR10+ Gaming standard. This new standard is an enhanced version of HDR10+ designed specifically for video games that adds Source Side Tone Mapping, auto calibration, and an automatic low-latency mode for TVs. Nvidia says this new feature is especially beneficial for games, since games can query HDR Luminance and color information directly from the display. This ensures the color reproduction of HDR10+ is perfectly produced no matter how strong or weak the display's HDR properties.

For HDR10+ gaming support to work, you'll need a GTX 16 series GPU or a GeForce RTX GPU, as well as a certified HDR gaming display.

Game Ready, Monitor Support, and CUDA 12

Finally, driver 526.98 features Game Ready support for several new titles, support for G-Sync Compatible displays, and support for CUDA 12.

Game Ready support includes the upcoming Spider-Man: Miles Morales, WarHammer 40,000: Darktide, WRC Generations, and World of Warcraft: Dragonflight. There's alao a DLAA update for Monster Hunter Rise. 

Monitor support includes 12 new displays from Philips, Eve, Corsair, Asus, and AOC. Ten of these monitors are 27- and 32-inch IPS and OLED panels with 144 Hz to 280 Hz refresh rates, while the other two are OLED TVs (ranging from 48 inches to 77 inches) with up to 120 Hz refresh rates at 4K resolution..

For more details, check out Nvidia's full patch notes here.

A new mobile Ryzen 9 CPU might just have made its first appearance in a benchmark, bringing 12 Zen 4 cores to laptops for the first time. If it exists, that is. This is definitely a rumor, for now, coming as it does from hardware leaker Benchleaks’ Twitter account.

The purported AMD Ryzen 9 7845HX APU appears in a benchmark listing for Ashes of the Singularity: Escalation, a strategy videogame with a reputation for being hard on the processing cores. The rig appears to use the integrated Radeon GPU, has 8GB of RAM, and ran the game at 1080p under DirectX 11 with its graphics settings on ‘crazy,’ which is the level above ‘extreme.’ It averaged just over 50fps.

With 12 cores and 24 threads - appearing in the benchmark as 24 cores, as was the case with the 12-core 7900X - this is the first Ryzen HX processor to be equipped with more than eight cores. HX processors were added to AMD's range in the Ryzen 5000 generation and signify even greater performance than the ‘normal’ H series, which are high-performance mobile APUs. The mainstream mobile APUs get a U on the end of their name, while an M series chip is for low-power mobile.

Videocardz speculates that this could mean the Ryzen 9 7845HX is a repurposed desktop model, and it certainly could be a member of the Dragon Range family announced back in June. This makes it one of the first mobile chips to be identified using AMD's new naming convention: the 7 at the beginning signifies 2022, the 8 is its market segment, and can confusingly mean it’s either a Ryzen 7 or 9 (though the CPU name is pretty clear). The 4 is for the Zen 4 architecture, and the 5 means it’s the upper model in its segment - a lower-class chip would have a 0 here.

Direct comparisons are hard to find, as most other benchmarks on the site show AMD HX chips from earlier generations running discrete GPUs, using DirectX 12 or Vulkan, or having their graphics options set completely differently. A 12-core Ryzen 9 5900X using the integrated GPU under Vulkan and with 32GB of RAM managed an average score of 112FPS, for example, suggesting DirectX 11 isn’t the best API to really show off the new chip's capabilities.

China-based Moore Treads on Friday began sales of its latest MTT S80 graphics card based on the Chunxiao architecture for client PCs. The add-in-board promises compute performance comparable to Nvidia's GeForce RTX 3060 Ti, but its performance in real games remains unknown. Meanwhile, it is available at JD.com, a store that delivers worldwide. However, there is a catch.

A GeForce RTX 3060 Ti-Like GPU from China

Moore Threads' MTT S80 graphics board is based on the company's Chunxiao graphics processor featuring 4,096 stream processors, 128 tensor cores, 256 texture units and 256 render outputs clocked at 1.80 GHz. In addition, the card carries 16GB of GDDR6 memory with a 14 GT/s data transfer rate connected to the GPU using a 256-bit interface (up to 448 GB/s peak bandwidth). 

From a compute performance point of view, the MTT S80 delivers up to 14.4 FP32 TFLOPS, slightly below the peak compute performance of Nvidia's GeForce RTX 3060 Ti (16.2 FP32 TFLOPS), which is one of the best graphics cards. Meanwhile, the actual performance of the MTT S80 AIB in real games needs to be tested. 

The GPU comes with drivers that support Microsoft's DirectX application programming interface, Khronos Group's OpenGL/OpenGL ES, proprietary MUSA, and multiple specialized APIs. However, Moore Threads openly says that its DirectX support is in its infancy, and only 20 popular titles are initially supported (e.g., Call of Duty, Crossfire, Counter-Strike, Diablo 3, League of Legends, etc.), so do not expect much from the card for now. Meanwhile, the company says that it is working closely with developers of Unity and Unreal Engine as well as designers of actual games to ensure compatibility and performance with the two popular engines. 

The Chunxiao graphics chip also has a capable video engine that can handle AV1, H.264, and H.265 codecs for up to 8K videos. Unfortunately, Moore Threads does not disclose which players support its hardware-assisted video decoding. Meanwhile, the MTT S80 AIB has three DisplayPort 1.4 and one HDMI 2.1 interface. 

Interestingly, the MTT S80 is the world's first client graphics card with a PCIe 5.0 x16 interface, as Nvidia's Ada Lovelace GPUs support a PCIe Gen4 bus. AMD has yet to disclose the host interface of its latest RDNA 3 GPUs in general and Radeon RX 7900 XT and Radeon 7900 XTX in particular.

Only Available in Bundles

The Moore Threads MTT S80 graphics card is now available at JD.com (via VideoCardz) for ￥2999 ($373 without VAT). However, the AIB does not come alone, as it is bundled with the Asus TUF Gaming B660M-PLUS D4 motherboard priced at ￥1029. If you subtract the value of the motherboard, then the graphics board will cost $245 without VAT, which is relatively cheap given its compute capabilities. Still, since the performance of the product is unclear, we can only wonder whether it is actually worth $245. 

Another thing to point out is that according to VideoCardz, the initial release of the MTT S80 will be very limited, and it is unclear how many units were actually made. It is also unclear when the second batch is set to come and how big it is.

A New Kid on the Block?

There are about a dozen known developers of graphics processors in China. Some target only datacenters (including AI and HPC), while others design client GPUs yet only aim local market. Founded in October 2020, Moore Threads designs GPUs for many applications (except HPC) and seems intent on selling its products outside of China. This is why the company makes its MTT S80 available at JD.com, which ships worldwide and works with developers of popular engines and games to ensure compatibility with western titles. 

Will it succeed? That depends on multiple factors, including engineering talent availability, willingness to invest, and ability to execute. Moore Threads is a very young company, and while so far it has released four AIBs (MTT S10, MTT S50, MTT S60, and MTT S2000) with two incoming (MTT S80 and MTT S3000), we can only wonder whether it has enough people to address all of its declared market opportunities. After all, AMD, Intel, and Nvidia employ thousands of people to work on gaming and datacenter GPUs. 

Furthermore, it remains to be seen whether western companies will work closely with a Chinese entity even on such things as video games. Given the strict curbs the U.S. government imposes against Chinese semiconductor and supercomputing sectors, some companies may consider it risky to work with China-based high-tech companies. In fact, there are things that they might deem potentially toxic.

The Chunxiao GPU in its MTT S80 form is aimed at consumer applications and therefore is not designed for any artificial intelligence or high-performance computing applications. Also, Moore Threads has the MTT S3000 32GB server grade board that offers a 15.2 FP32 TFLOPS throughput and supports datacenter-oriented features, such as up to 32-way GPU partitioning as well as SR-IOV PCIe virtualization. Meanwhile, the Chunxiao graphics processor supports FP32, FP16, and INT8 precision. Therefore, assuming that it also supports an appropriate instruction set, it can theoretically be used for AI or even some light HPC workloads that do not need FP64.  

Based on what we know about the Chunxiao graphics processor, we believe it is aimed primarily at gaming and entertainment applications. The datacenter MTT S3000 will likely mainly be used for remote Android games rendering, virtual desktop infrastructure (VDI), and video streaming services.  

Whether the MTT S3000 can actually be used for non-consumer AI and technical computing applications in datacenters efficiently is something we do not know. We have reasonable doubts given the form factor, power consumption, and performance. Yet, if the U.S. further tightens its sanctions against the Chinese high-tech industry, there might be ramifications even for a consumer-oriented company like Moore Threads. 

Microsoft teased the arrival of its DirectStorage 1.1 API almost a month ago, and it is now available for developers to get started with its newly implemented GPU decompression functionality. Remember, Direct Storage 1.1's headlining ability is accelerating game load times by as much as 40%, depending upon the game and PC hardware configuration.

In a blog post heralding the arrival of DirectStorage 1.1, Microsoft wrote mostly to the audience of developers, pointing to guides, resources, and describing what’s new from a technical viewpoint. The key confirmation is that “GPU decompression and GDeflate [are] now available.” But there are a number of performance increases and bug fixes to Direct Storage in general, too.

The utility value of DirectStorage going forward will depend on a variety of things. On the consumer side, PC gamers will need modern hardware to feel any benefits, with at least a DirectX 12 + Shader Model 6.0 GPU, plus a fast, modern storage device like one of the latest NVMe SSDs. 

In the demos showcased last month, Microsoft shared a screenshot (see how to take screenshots in Windows) where DirectStorage 1.1 was used in a pretty ideal setup, and it loaded a ‘game’ scene up to 3X faster than an earlier API. Moreover, the GPU-powered decompression reduced CPU loads significantly. But, of course, whether your CPU or GPU has more ‘spare time’ will depend a lot on the game engine and your hardware.

Importantly for the adoption of this data loading acceleration API, some optimized drivers are already ready. Microsoft linked to support pages from key PC hardware partners like AMD, Intel, and Nvidia in its blog. If you follow the respective links, you can learn more about each vendor’s work to support DirectStorage 1.1.

AMD highlights the important fact that “DirectStorage is a feature that must be enabled by (game) application developers to realize the benefits.” Thus, end users can’t just wait for a new GPU driver. Sadly, AMD hasn’t confirmed whether it has already implemented DirenctStorage 1.1 support in a shipping driver or if it is something we have to look forward to.

Nvidia says that the current Game Ready Driver (version 526.47) includes optimizations for GDeflate. It provided some performance comparison charts showing the clear advantages offered by GPU decompression in game / level loading.

Intel says its Arc graphics driver 101.3793 includes DirectStorage 1.1 optimizations for Intel Xe architecture on systems with NVMe SSDs. It also kindly shared some performance examples.  In its example featuring a system using an Intel Core i9-12900K CPU and Arc A770 16GB GPU, the decompression bandwidth was increased from 7.88 GB/s to 21.67 GB/s.

So, game developers can, at last, begin in earnest implementing DirectStorage 1.1, with the carrot at the end of the stick being significantly shorter loading times and faster game data streaming. We look forward to seeing the improvements to available games coming as patches, as well as new games coming with DirectStorage 1.1 enabled from launch.

The Nvidia GeForce RTX 4090 delivers new levels of performance, lands at the top of our GPU benchmarks hierarchy, and is the fastest of the best graphics cards. For all that, it's not the most efficient of GPUs, but that's mostly due to design decisions. We can improve efficiency — and potentially reduce the risk of melting 4090 power adapters — through power limiting. This is basically a quick and easy alternative to underclocking and undervolting, and the opposite of overclocking.

For our testing, we've taken Nvidia's RTX 4090 Founders Edition and run it through eight of our most demanding tests to see what happens to performance, clock speeds, temperatures, and power requirements at the various settings. We tested in 10% increments, starting at 120% and dropping down to just 50% to provide the full range of options.

Before we get to the results of our testing, note that without increasing the GPU clock speed, the higher power limits don't really do much for performance. We've tested RTX 4090 overclocking elsewhere and found that in our standard gaming test suite, even at 4K ultra settings, overclocking only increased performance on the RTX 4090 by about 4%. In our more demanding DXR (DirectX Raytracing) test suite, however, overclocking proved a bit more helpful and increased performance by about 9% overall.

In short, the more demanding a game happens to be at whatever settings we test with, the more impact we'll see from changing the power limit. Conversely, there are a lot of games that don't tax a GPU like the RTX 4090 much at all, even at 4K and maxed-out settings — Microsoft Flight Simulator is a perfect example of this, given its CPU-limited nature. Such games may run just as fast at lower power limits, but that's because the GPU wasn't hitting its 450W TBP (Total Board Power) limit to begin with.

RTX 4090 Power Limiting Test Setup

We're using our existing Core i9-12900K test PC. We would likely increase power consumption from the graphics card if we were to upgrade to the new Core i9-13900K, but probably not by too much. We're using our DXR test suite for the power-limited testing, with the addition of A Plague Tale: Requiem, a recent release that also happens to support DLSS3. We wanted to include at least one set of tests using DLSS2 and DLSS3 (we tested both) to see what that did for performance and efficiency.

The other games we use for testing are Bright Memory Infinite Benchmark (far more impressive looking and demanding than the actual game), Control Ultimate Edition, Cyberpunk 2077, Fortnite, Metro Exodus Enhanced Edition, and Minecraft. We mostly use the maximum quality setting with all ray tracing options enabled, without DLSS of any form, though we didn't enable "Psycho" lighting in Cyberpunk. Note also that we enabled HairWorks and Advanced PhysX in Metro Exodus Enhanced, which is slightly more demanding than the previous testing we've conducted using that game. In other words, these results should only be compared within this article, as other reviews may have used slightly different settings.

Again, we've run all eight of the above gaming tests at eight different power limits on the RTX 4090 Founders Edition. Besides the default 100% limit, we tested at 110% and 120% to see if simply increasing the power limit would help performance. Then we also tested at 90%, 80%, 70%, 60%, and 50% to try and improve efficiency.

We capture all performance data using Nvidia's FrameView utility, which also collects GPU clock speeds, temperatures, and power consumption data. While the power consumption reported by the software isn't exactly the same as real-world power use measured using external tools like Powenetics, looking at the results indicates that Nvidia's software reports figures that are within about 10W of what we measured using Powenetics, with the added benefit being that we're able to collect a bunch of data much more quickly.

With that preamble out of the way, let's hit the test results. 

RTX 4090 Overall Power Efficiency

We have two separate charts, one showing our usual performance data (average and 99th percentile FPS) and the second showing power, temperature, and GPU clocks. Note that we didn't attempt to adjust the GDDR6X clocks, so potentially, as we drop the power limit, the memory will consume a slightly larger portion of the overall power budget. We'll start with the overall results, showing the geometric mean (equal weighting to all scores) of our eight tests.

In terms of performance, it's immediately clear that there's almost no change either by raising the power limit by up to 20% or by dropping it by 10%. There's also only a very slight 2.3% drop in frame rates with an 80% power limit — measurable but not particularly meaningful. Even with a 70% power limit, performance is still just 5% slower than at stock, which drops to 10% slower at 60% and 23% slower at 50%.

In terms of pure efficiency, every power-limited configuration ends up being better than the stock settings. Performance per watt, at 4K and in the most demanding games, measured 0.175 FPS/W at 100%. Efficiency drops at 110% and 120%, to 0.169 FPS/W and 0.166 FPS/W, respectively. Going in the other direction, it improves to 0.179 at 90%, 0.194 at 80%, 0.214 at 70%, 0.234 at 60%, and finally 0.237 FPS/W at 50%.

Again, that's "better" across the suite of power limits. However, gains are clearly tapering off at 50%, and pure performance still matters — most people don't buy a $1,600 graphics card just to limit performance while improving efficiency! But if you're worried about the adapter melting and are waiting for a replacement to arrive, cutting power use should certainly reduce the risk.

The average power use across our eight game tests was 402.3W at stock, which increased to 428.3W with the 120% power limit — so a theoretical 20% increase in power limit only caused a real-world increase of 6.5% due to other system bottlenecks. More interesting is the drop in power use, clocks, and temperature as we limit the power. At 80%, which seems to be the sweet spot on the efficiency curve, the average power use is 353.8W, about the same as what you'd see from an RTX 3090 or 3080 Ti. GPU temperature also dropped a couple of degrees, and clock speed still averaged 2,611 MHz.

With the 50% power limit, the average power draw was only 224W, GPU temperature was just 53C, but the core clocks are now down to 1,644 MHz. So performance would probably still be competitive with an RTX 3090 Ti, but we'd still be more inclined to increase the power limit to 60% or 70%. 

Bright Memory Infinite Benchmark Power Efficiency

The Bright Memory Infinite testing results are very similar to the overall chart. Performance barely changes at all, with up to 20% more power available due to clock speeds hitting their limit. Dropping the power limit to 80% decreases the power use by 14% (355W instead of 415W) while only reducing performance by 3%.

The power draw at stock on the reference RTX 4090 Founders Edition is still well over 400W, though it's nowhere near the 450W TBP maximum. There have been reports that the RTX 4090 "only uses about 350W while gaming," but those are clearly not universal truths. In CPU-limited scenarios, GPU power use will drop significantly, but in the most demanding games — particularly ray tracing games — you'll regularly see more than 400W at 4K ultra settings.

Control Ultimate Edition Power Efficiency

Control Ultimate Edition is one of the more demanding games in our test suite, at least as far as power consumption is concerned. At stock, the 4090 FE used 424W of power, so only 26W below the TBP. That should mean it will be affected more by power limit changes than some of the other games.

In practice, things aren't all that different from Bright Memory Infinite. Power use at 80% drops to 357W, a 17% reduction, while performance only drops by 2.1%. It's possible lengthier gaming sessions might show more of a difference, as time constraints limited our testing to a few minutes per game, but clearly the 4090 at stock settings isn't targeting maximum efficiency. 

Cyberpunk 2077 Power Efficiency

Cyberpunk 2077 has a reputation as being one of the most demanding games around. Interestingly, it used less power at stock in our tests than several of the other games. We measured 413W of power use at 100%, which dropped to 351W at 80%.

The ideal balance between power and performance once again seems to be 70% or 80%. At 80%, you get 96% of the performance with 85% of the power use, while the 70% limit gives you 94% of the performance at 75% of the power use. (You don't use 70% or 80% of the base power because the power limit stems from the 450W TBP.) 

Fortnite (DXR) Power Efficiency

Fortnite, even with maxed-out ray tracing settings and without DLSS, ends up being one of the least demanding games in our test suite — and if you're running without ray tracing effects, it would have even lower power requirements. At stock, the RTX 4090 only consumes 375W of power, and even at a 120% power limit, it only hits 407W. Performance losses from reducing the power limit should be even lower, though actual power savings may not be as significant either.

The 90% and 80% power limit both deliver similar performance, at least within the margin of error — 3% slower than stock, with power use of 368W and 346W, respectively. The 70% limit drops power use to just 309W, an 18% reduction versus stock, while performance is down 9%. The best setting for Fortnite, based on our testing, seems to be 60%: you get 91% of the base performance while using 29% less power.

Metro Exodus Enhanced Edition Power Efficiency

Somewhat surprisingly, Metro Exodus Enhanced Edition ended up as the most demanding game from our test suite. We're using the Extreme preset, but we also enable HairWorks and Advanced PhysX. Even at stock settings, the RTX 4090 uses 432W, which jumps up to 465W with a 110% power limit and 483W with a 120% limit. That's interesting, considering none of the other games even managed to break 450W at the 120% setting. Unfortunately, the performance gains from the higher power limit are still marginal, at 1.2% and 1.8%.

Given the baseline measurements, reducing the power limit as expected causes a slightly larger drop in performance. At an 80% power limit, actual power use drops by 18% while performance drops by 5%. Each additional 10% drop in power reduces performance, though 70% still yields 92% of the base FPS. The final two power limits create a larger dip in performance, and efficiency at the 50% setting is actually worse than at 60%. 

Minecraft Power Efficiency

Minecraft without ray tracing can run on a potato, but with DXR enabled, it's often quite demanding. It uses "full path tracing" to provide extensive lighting, reflection, and shadow effects, but it seems we've hit the limits of what it needs with the RTX 4090, as the base power use is only 380W. Like Fortnite, that means we see less of a performance impact by limiting power than in some of the other games.

The optimal setting on the RTX 4090 Founders Edition looks to be 60%, which reduces power use by 30% but only drops performance by 8%. Technically the 50% setting still improves performance per watt, but the additional 10% reduction in power use also causes an 11% decrease in performance. 

A Plague Tale: Requiem Power Efficiency

A Plague Tale: Requiem is the only game where we tested with DLSS 2 and DLSS 3 (Frame Generation) just to see how much that affects power use. In theory, Frame Generation should reduce the GPU load while increasing performance — though it also increases latency. And that's precisely what we see in our testing.

With DLSS 2 Super Scaling enabled, in Quality mode, power consumption at stock settings is 398W. That's slightly lower than several other games but not the lowest we tested. DLSS 3 drops that to 387W, so there is not a huge difference overall, and it also breaks the 144 fps mark.

With DLSS 3 enabled, framerates stay above 144 fps at the 80% power limit and fall just below that mark at 70%, so you only lose 1% and 4% of the base performance. However, power use doesn't drop as much because of the lower starting point: it's down just 8% on the 80% setting and 19% on the 70% setting. With DLSS 2, the performance and power impact was slightly larger but hardly worthy of discussion.

Nvidia RTX 4090 Power Efficiency: Closing Thoughts

There's an ideal spot on the voltage and frequency curve for maximum efficiency, which is what Nvidia likes to target with its laptop GPUs. For desktops, however, chasing maximum performance at the cost of efficiency is often the name of the game. For example, with the RTX 4090, the final ~5% in performance basically requires 15–20% more power.

Given the issues users have encountered with the apparently faulty 16-pin power adapter, anyone who already purchased an RTX 4090 and who still uses the adapter might want to look into reducing power consumption — along with removing the side panel on the case and doing their best to avoid putting any stress on the cable connector. While lower power use doesn't guarantee the connector won't fail and melt, it should at least make it less likely to happen while we wait for what will almost certainly be a manufacturer recall of the adapter.

On the bright side (for Nvidia), the actual cost of the adapter cable should be relatively trivial compared to the cost of an RTX 4090 graphics card. Another bright side, this time for users, is that you can improve overall efficiency by 20–30% if you run a 4090 with a 70% or 60% power limit… but you'll also reduce performance on your shiny new and not yet melted extreme GPU.

Results with other RTX 4090 cards will also likely differ from what we've shown here. Factory overclocked models often run at higher voltages and don't necessarily work as well at reduced power limits, at least in our experience. Still, what we've done here was simple enough. All you need is a utility like MSI Afterburner and then set an appropriate power limit.

We're certainly curious to see how AMD responds to the 4090 melting adapter debacle. Of course, we know AMD has no intention of using the 16-pin 12VHPWR connector on its upcoming RDNA 3 GPUs, but AMD has been beating the efficiency drum for the past few generations. So RDNA 3 might end up being a lot more efficient than Ada Lovelace, at least if you stick with Nvidia's default power settings. Will it be faster, though? That remains to be seen.

• MORE: Best Graphics Cards
• MORE: GPU Benchmarks and Hierarchy
• MORE: All Graphics Content

Microsoft says the DirectStorage 1.1 API is coming soon to Windows PCs. With this release, Microsoft will enable GPU Decompression on PCs per its previously shared roadmap. The primary benefit of GPU Decompression is that PC gamers could see load times reduced by as much as 40% depending on the game and the PC hardware configuration.

DirectStorage 1.0 was introduced for Windows (10 and 11) in March 2022. It was designed to bring speed, bandwidth and latency optimizations to the Windows storage subsystem, which would scale with storage performance. Thus faster storage, like the latest NVMe SSDs, deliver greater overall benefits. Version 1.1 of the API builds on this promise with GPU Decompression, reducing the processing of graphical assets using the CPU and moving this workload to the GPU, which is adept at massively parallel processing and decompression tasks.

Microsoft explains in its DirectX Developer Blog that modern games typically require massive amounts of data to build and render the immersive worlds we navigate. This could mean hundreds of gigabytes of compressed data loaded to run the latest AAA titles. “When a game is run, the assets are transferred to system memory, where the CPU decompresses the data before it is finally copied into GPU memory to be used as needed,” says Senior Microsoft Program Manager, Cassie Hoef. “The transfer and decompression of these assets on gaming devices contributes heavily to load times and limits how much detail can be included in open world scenes.”

DirectStorage 1.1 has an optimized GPU-powered decompression algorithm to deliver loading performance benefits of up to 40% to the PC gaming experience. To back up its claims, Microsoft built an optimized demo where scene loading was up to 3x faster, and CPU loads were drastically reduced. Above, you can see a GPU with GDeflate (left) loading a scene in 0.8 seconds versus the CPU with Zlib (right) taking 2.36 seconds.

DirectStorage was initially described as a Windows 11-exclusive technology before Microsoft caved and facilitated its use in Windows 10 gaming experiences. Like the previous 1.0 version of the API, there are still said to be “additional optimizations in the IO stack available to Windows 11 users.” A DirectX 12-capable GPU supporting Shader Model 6.0 is necessary, and users should have game data on a high-speed NVMe SSD for optimal gains.

It will be interesting to see this technology applied in modern games, as in many titles on my mainstream 2022-built PC, it is the GPU that is usually maxed out, not the CPU.

Developers will be able to play with DirectStorage 1.1 later this year, and when it goes mainstream, we expect GPU driver updates from all major players. Please don’t hold your breath, though, as we are yet to see even one AAA game with DirectStorage 1.0 support released on Windows. Forspoken, by Square Enix, was going to spearhead DirectStorage on Windows this month but has been rescheduled for late January 2023.

YouTube channel Benchmark Lab has decided to test Nvidia's new GeForce RTX 4090 in a unique fashion — by pitting it against two RTX 3090s in SLI. 

That's right, SLI — Nvidia's obsolete multi-GPU tech that Benchmark Labs managed to run using some unknown SLI-enablement magic. The channel was able to run SLI decently well in several DX11, DX12, and Vulkan titles, but ultimately the 3090s couldn't beat the RTX 4090 (though they got close). 

Before we jump to the results, it's worth noting that all of this should be taken with a grain (or several) of salt. SLI has been unsupported for years at this point, and getting it to work with most modern titles — especially with DX12 and Vulkan — would require...unorthodox methods. We're not sure how Benchmark Lab managed to enable SLI in most of the games tested — again, specifically the DX12 versions — or if the results were doctored or extrapolated. That said, the results are pretty interesting (if they are, indeed, accurate). 

Benchmark Lab tested 9 games: Spider-Man Remastered, Cyberpunk 2077, Watch Dogs Legion, Microsoft Flight Simulator 2020, Minecraft RTX, God of War, Witcher 3, Horizon Zero Dawn, and Read Dead Redemption 2. Testing was performed at 4k resolution using a variety of different quality settings and DLSS modes. 

In Spider-Man Remastered, with maxed out settings (Ray Tracing and DLSS Balanced mode enabled), the RTX 3090s in SLI achieved 80 - 85 frames per second (fps) on average, while the RTX 4090 managed a noticeably higher 95 fps average. 

In Cyberpunk 2077, with maxed out settings (Ray Tracing and DLSS Balanced mode enabled), the RTX 3090s in SLI managed 50 fps while the RTX 4090 averaged 70 fps.

In Watch Dogs: Legion, with maxed out settings (Ray Tracing and DLSS Quality mode enabled) the gap closed a little: the RTX 3090s in SLI averaged 73 fps, while the RTX 4090 averaged 80 fps.

The results were even closer in Microsoft Flight Simulator with maxed out settings (Ray Tracing and DLSS Quality mode enabled). The RTX 3090s in SLI averaged 80 fps, while the RTX 4090 averaged 83 - 85 fps. (However, note that Microsoft Flight Simulator is known to be CPU-bound, especially at higher frame rates.)

In Minecraft RTX, with max settings (Ray Tracing enabled), the RTX 3090s in SLI averaged 70 fps while the RTX 4090 hit a slightly higher average of 75 fps. And in God of War at max settings (DLSS Quality mode enabled), the RTX 3090s in SLI hit an average of 103 fps while the RTX 4090 did quite a bit better with an average of 120 fps.

For the rest of the titles, check out the video below: 

RTX 3090 SLI Is The Only Setup That Can Get Close To The RTX 4090

With the exception of the Cyberpunk 2077 results, the RTX 3090s in SLI were around 8 - 15% slower than the RTX 4090. For an SLI implementation, this isn't bad at all — and it's the only GPU configuration we've seen so far that can get close to the RTX 4090's performance bracket. For some perspective, the next-closest GPU is the RTX 3090 Ti — which trails the 4090 by over 50% based on our tests. 

Sadly, the 3090s in SLI setup wasn't being used to its full potential, as the secondary 3090 consistently maxed out at 45% utilization. This is one of the pitfalls of SLI setups, in which bad optimization leads to less-than-superb GPU utilization on the secondary card. In theory, if we could get anywhere close to 100% utilization on both GPUs, we would probably see the 3090s in SLI outperform the 4090 by a good margin. 

But the chances of that happening are extremely unlikely, since Nvidia officially killed off SLI in the form of physical bridges with the RTX 40 series. SLI support has dwindled so much over the past several years that it's basically only useful in synthetic benchmarks at this point. 

There are technically ways of enabling SLI (unofficially) in games that don't support it — like in the case of the benchmarks seen in this video. But these results are unpredictable at best, and usually result in system instability or severe micro-stuttering problems while gaming. 

There is some hope for SLI, however. Multi-GPU workloads are very common in the enterprise space, and Nvidia even has multi-GPU technologies that don't require an NVLink or SLI bridge. 

On the gaming side, modern APIs such as DX12 and Vulkan do have the capability to render frames to two completely different GPUs in tandem. So there's a chance multi-GPU tech might make its way back to the gaming space, eventually (but whether developers want to support multi-GPU technology for gaming is a different story). 

Today is the release date for Nvidia's GeForce RTX 4090, and with it comes a brand new driver — version 522.25, which is jam-packed with features. Aside from RTX 4090 and game ready driver support, Nvidia has added a slew of DX12 optimization updates, which apply to 14 titles and give all RTX GPUs a performance boost (depending on title and resolution).

This is one of the most feature-rich updates we've seen from Nvidia in a while — it might be worth installing if you don't update your graphics drivers often. Nvidia's DX12 optimizations steal the show, offering performance gains of up to 25% in titles such as Assassin's Creed Valhalla. 

The optimizations include the following 14 titles:

Assassin’s Creed Valhalla: up to 25% (1080p)
Battlefield 2042: up to 7% (1080p)
Borderlands 3: Up to 8% (1080p)
Call of Duty: Vanguard: up to 12% (4K)
Control: up to 6% (4K)
Cyberpunk 2077: up to 20% (1080p)
F1(R) 22: up to 17% (4K)
Far Cry 6: up to 5% (1440p)
Forza Horizon 5: up to 8% (1080P)
Horizon Zero Dawn: Complete Edition: up to 8% (4k)
Red Dead Redemption 2: up to 7% (1080p)
Shadow of the Tomb Raider: up to 5% (1080p)
Tom Clancy’s The Division 2: up to 5% (1080p)
Watch Dogs: Legion: up to 9% (1440p)

Most of Nvidia's DX12 optimizations don't significantly improve performance, but they're still better than nothing. On the low end, Nvidia's optimizations provide a 5% boost in Shadow of the Tomb Raider and Tom Clancy's The Division 2. The best performance results come from F1 22 with a 17% performance boost at 4k resolution, and Assassin's Creed Valhalla, which saw a 25% bump at 1080p resolution.

According to Nvidia, these DX12 optimizations include shader compilation optimizations, reduction of CPU overhead, and resizable bar profiles for Forza Horizon 5 and F1 22 (the last of which only applies to resizable bar supported GPUs — RTX 30 series and newer).

The rest of these optimizations will work with any RTX graphics card, so it's a free upgrade for anyone packing an RTX GPU — but your mileage will vary based on your system configuration. Nvidia's results are coming from a test bed consisting of an RTX 3090 Ti, RTX 3080 Ti, RTX 3070 Ti, and RTX 3060 Ti paired to a Core i9-12900K processor with 32GB of RAM. 

RTX 4090 Support

This driver update is also the first update supporting Nvidia's all-new GeForce RTX 4090 GPU, and is required to run the new flagship card (if you're planning on buying one today). 

The RTX 4090 is Nvidia's first RTX 40-Series GPU operating on the Ada Lovelace GPU architecture. In our review of the Founders Edition variant, we saw performance improvements of 50 - 70% over the RTX 3090 Ti.

DLSS 3, DLSS 2, and Game Ready Support

Nvidia also announced several new DLSS 3- and DLSS 2-supported titles as well as game ready support for A Plague Tale: Requiem, Gotham Knights, Scorn, and Uncharted: Legacy of Thieves Collection.

Newly-supported DLSS 3 titles include:

A Plague Tale: Requiem (available Oct. 18)
Bright Memory: Infinite (available Oct. 12)
F1 22 (updates coming soon)
Justice ‘Fuyun Court’ New Graphics Showcase (available Oct 12)
Loopmancer (available Oct. 12)
Marvel's Spider-Man Remastered (available Oct. 12)
r (available in beta for Xbox Insider Program members Oct. 17)
SUPER PEOPLE (available now in early access)

Newly-supported DLSS 2-only titles include:

Blind Fate: Edo no Yami
Dakar Desert Rally
Destroy All Humans! 2 - Reprobed
Gotham Knights
Uncharted: Legacy of Thieves Collection

It's worth noting that all DLSS 3-supported games will also support DLSS 2 at the same time — going by the name "DLSS Super Resolution" in game menu options. DLSS 3 is the frame amplification/generation version of DLSS 2, and can be turned on and off in game settings.

Intel Arc A750 and A770 are about to reset pricing expectations on October 12. Intel Arc Alchemist has been in development for several years, and after delays and misstarts, we're finally in the home stretch. The Arc A770 and A750 will launch on October 12, and they're setting their sights firmly on the midrange market for the best graphics cards.

Intel provided additional details on what we'll see when the cards arrive in just a couple of weeks. Earlier this week, during the Intel Innovation keynote, Intel CEO Pat Gelsinger revealed a starting price of $329 for the A770. But which A770 was Pat referring to?

It's the A770 8GB model, to the surprise of no one, but the other Arc A700-series cards are priced equally aggressively. Intel's own Arc A770 16GB Limited Edition will cost $349, just $20 more than the baseline A770 8GB. $20 for double the memory? Yes, we'll take that in a heartbeat, even if it probably won't matter that much in most games, at least today. But the bigger deal is arguably the Arc A750, which will have 8GB and a starting price of just $289.

Intel's not being shy about performance comparisons, going straight after Nvidia's nominally $329 RTX 3060 — except that actual GPU street pricing for that card still starts at $369. Amazon has the PNY RTX 3060 XLR8 for $377, while Newegg has the MSI RTX 3060 Ventus for $369 (plus a 15% off code right now if you hurry). That might be the first time we've seen an RTX 3060 go for less than its MSRP since it first launched in January 2021.

Even at price parity, though, Intel still thinks it may have the upper hand. To that end, it provided a full suite of 67 gaming benchmarks with the A770, A750, and RTX 3060. The games were tested at both 1440p high and 1080p ultra, which seems a reasonable target for midrange hardware.

Intel also provided several "bang for the buck" slides showing the A770 and A750 going up against the RTX 3060, but the catch is that it used a "current" price of $418 for the RTX 3060. That might have been the going rate for the card last month, but GPU prices are quite volatile right now, and we're able to find much better prices just one week after Intel collected its data.

As with all manufacturer-provided benchmarks, we can't vouch for the performance data shown in this slide. Intel tested both cards with a Core i9-12900K, 32GB of memory, and an Asus Z690 DDR5 motherboard. While it didn't show the raw framerates, even the lower-priced Arc A750 showed generally better performance across the test suite.

Intel made a point of testing games that use all three major graphics APIs as well: DirectX 12, Vulkan, and DirectX 11. Games that support DX12 or Vulkan were tested using that API, as for Intel that generally means better performance. That might not be true of the RTX 3060, however, where DX11 at times still runs better than DX12 or Vulkan — again, a dose of skepticism is certainly warranted.

DX11-only games not surprisingly remain the lowest performing class overall, with about ten of the 17 games in that group showing better performance on the 3060. The DX12 titles meanwhile showed Intel Arc A750 leading in 32 of the 43 tested games, and in Vulkan the A750 was ahead in six of the seven games.

Unlike AMD, Intel isn't obviously avoiding comparisons in ray tracing games. It also showed off performance in 17 different games that use ray tracing. Even there, an area where Nvidia has traditionally reigned supreme for the past four years, the Arc A770 was faster in 13 of the tests.

Intel also discussed how building the Arc drivers using the integrated drivers as the starting point proved problematic. It's worked hard over the past several months to improve the situation, and that's beginning to pay off. The DX12 Optimized titles are games where Intel has specifically delivered targeted performance improvements, while the other DX12 games are just using standard code. As time passes, a lot more games should start to fall into the "optimized" category, with commensurate performance.

The full details of the Arc desktop lineup remain unchanged, but it's nice to see the A580 still waiting in the wings. That won't be launching on October 12 alongside the A700 cards, but given the $289 price point on the A750, we can't help but wonder how low Intel will go. $249 would put it right against the RTX 3050 and RX 6600, though AMD's card easily outclasses the 3050 at that price point.

That brings up another interesting topic: How do the Arc A750 and A770 fare against AMD's RX 6650 XT starting at $300? That's typically 10–20% faster than the RTX 3060, according to our GPU benchmarks hierarchy, at least in our standard non-ray tracing test suite. (In the DXR suite, the 3060 ends up being 25–30% faster.) Things could be close, which is why we still need to run the benchmarks ourselves. That will be happening over the next two weeks as we prepare for the retail launch.

There are other things to consider as well. Standard apples-to-apples gaming performance is a good starting point, but Nvidia has a lot of games with DLSS support now available. At present, there are only a few XeSS-enabled games. Intel showed performance improvements ranging from 37% to as much as double the fps with XeSS Performance mode at 1440p.

We tested Shadow of the Tomb Raider with XeSS earlier this week, and a few other games are now publicly available with XeSS support (like Death Stranding). On the A380, XeSS was a bit problematic, but Intel says a lot of that might be due to trying to scale down the algorithm to a significantly slower GPU. We'll see how it does on the faster Arc GPUs soon enough.

But now Nvidia has DLSS 3 waiting in the wings, clearly trying to stay one step ahead of the competition. That's not coming to RTX 3060, however, and who knows when we'll see an RTX 40-series competitor to Arc A770 (our guess: next spring).

Intel Arc GPUs are set to launch in just a couple of weeks, and we're excited to test them out and see how they stack up to the AMD and Nvidia competition. We're also glad to see a potential true mid-range GPU launch for the first time in far too long. Intel almost certainly wouldn't have priced Arc this low had it been ready six months ago, but fears that it might try to stick to earlier rumored pricing are at least safely put to rest now.

As Intel shifts from the Alchemist launch over to Battlemage, which it once again reiterated is where most of the graphics team is now working, we can only hope it will execute better on Arc. The A750 and A770 both look promising right now, and we'll see them in action soon enough. But at this point, we're now looking on to AMD RDNA 3 and Nvidia Ada Lovelace. The sooner Battlemage can join the fray, the better.

The full presentation deck can be viewed below. Again, performance per dollar data needs to be taken with a healthy helping of salt, as the 3060 prices have dropped potentially $40 or more in the interim. And who knows where they'll be in two more weeks when Intel's competing cards officially arrive?

The first benchmark results of the Intel Arc A580 RI (we don't know what "RI" means) have been added to the Ashes of the Singularity database. This occurred just hours after Intel published specifications of its upcoming Arc Alchemist graphics cards, including the midrange Arc A580 — a card that should go up against midrange GPUs in our GPU benchmarks hierarchy, and it might even have a shot at displacing one of the best graphics cards if it's priced right.

Testing unreleased hardware is always a challenge as drivers may not be optimized and software might need tweaking. But these Arc A580 performance numbers raise more questions than answers. The main one is whether the published results were indeed obtained on an Arc A580, and whether it was running final clocks and decent drivers.

@Benchleaks discovered four "Intel Arc A580 Graphics RI" benchmark results in the AOTS database (1, 2, 3, 4) obtained on a system based on an unreleased Intel 16-thread processor with a 2.50GHz default clock rate (and an unknown turbo frequency) and equipped with 16GB of memory. The results were obtained with the Min_1080p preset. We decided to compare leaked results of Intel's upcoming midrange Arc A580 graphics board to our own test results of Intel's Arc A380 as well as results obtained on Nvidia's entry-level graphics cards from the AOTS database.

It should be noted that some of Arc A580 results were obtained using a DirectX 11 renderer and some using a DirectX 12 renderer. Considering that Intel openly admitted that it did not optimize drivers of its discrete GPUs for the DirectX 11 application programming interface, AOTS DX11 results are barely worth consideration. Meanwhile, we added the best Arc A580 results with DX11 and DX12 renderers into the table just to show the difference.

Note: we did not include performance numbers for Nvidia graphics cards obtained on systems based on high-performance CPUs.

Now, when compared to Nvidia's GeForce GTX 1650, our own test results for the Arc A380 look pretty good, which is logical as this is an entry-level gaming GPU based on a several years old architecture. The RTX 3050 results meanwhile look bad because they're using DX11 — we couldn't find any RTX 3050 DX12 numbers.

So Jarred pulled out a few other GPUs and tested the same settings to put things into proper perspective, all tested on the same i9-12900K platform. You can see how things stack up, and it doesn't lend much credence to the supposed A580 results. The Arc A580 appears to be slower than the Arc A380 in these AOTS benchmarks. There can be several reasons for these odd results.

• The CPU in the system used to benchmark the Arc A580 is limiting performance so badly that a GPU which is supposed to be three times faster than the Arc A380 turns out to be slower.
• Drivers used for the Arc A580 were not optimized for the ACM-G10 GPU.
• The Arc A580 RI is a mobile part that has a power limitation.
• This is not an Arc A580 and someone cheated the benchmark into thinking that it's dealing with this card when it's actually something else.

Truth to be told, Ashes of the Singularity is not a good GPU benchmark since the game was launched in 2016 and should be a fairly easy nut to crack for modern high-end graphics processors, especially with the Min_1080p preset. Meanwhile, the benchmark favors AMD GPUs and takes huge advantage of fast CPUs — it's really more of a CPU benchmark than a GPU benchmark, particularly with the Min_1080p setting. Also, AOTS tends to show better performance on its second run (though not always!), which suggests that there is more variance between runs if you're not careful.

In other words, all AOTS benchmark results need to be viewed with a helping of salt, and perhaps a slice of lemon and a shot of tequila at the viewer's discretion. Keeping in mind that we are dealing with a pre-release piece of hardware with unknown drivers, it's really hard to draw any conclusions about the performance of Intel's Arc A580 in AOTS, and we're not even sure this is a real Arc A580. We should hopefully have actual hardware in hand where we can run our own tests in the not-too-distant future, based on what we're seeing on the Intel Graphics channel.

Jarred Walton contributed to the story.

Regarding the best graphics cards, compatibility is just as important as performance. While AMD and Nvidia don't have many issues in that department, Intel's Arc Alchemist still has some kinks to work out since the company's graphics cards struggle with legacy APIs.

Unlike other publications focusing on the Arc A380's performance, German news outlet PC Games Hardware took the graphics card for a spin through 50 games that launched between 1999 and 2022. The site used Gunnir's Arc A380 Photon 6GB OC for the tests, one of three custom Arc A380 models available on the market.

Overall, the Arc A380's compatibility across the 50 titles was pretty decent, considering the state of Intel's Arc drivers. The graphics card only had problems with nine games out of the 50 that PC Games Hardware tested. If we want to simplify the results, Arc A380 would have issues with every 1 out of 5 games. It's not an awful ratio, but Intel's drivers ultimately have a lot of optimization headroom.

It did surprise us that the Arc A380 didn't play nice with Doom Eternal, Quake 2 RTX, Halo Infinite, or Spider-Man Remastered, considering that these titles leverage the Vulkan or DirectX 12 APIs. Of course, Arc should work remarkably with games on modern APIs. But then again, let's not confuse performance with compatibility.

PC Games Hardware's results also revealed that Arc runs flawlessly on APIs that precede DirectX 12. Popular games, such as Counter-Strike: Global Offensive (DirectX 9) or The Witcher 3 (DirectX 11), worked without hiccups. However, it's surprising that the German publication left out other most-played Steam, including Dota 2 (DirectX 11), Apex Legends (DirectX 11), and Lost Ark (DirectX 11), out of its testings.

Do remember that PC Games Hardware tested for compatibility. Performance, on the other hand, is another story. Intel has admitted that Arc performs poorly on DirectX 11 and older APIs. The chipmaker aims to close the performance gap between legacy APIs and DirectX 12, but it's a long work in progress or a "labor of love forever," as Intel Fellow Tom Petersen calls it.

The Arc A380 has been available in China for a couple of months now. The ASRock Challenger Arc A380 model recently arrived on U.S. soil via Newegg for $139.99, and it's already sold out. We're still waiting for Intel's Arc A5 and A7 series, which will hopefully land before the end of the year.

The Arc A380 may not be one of the best graphics cards around. However, Intel's entry-level Arc Alchemist graphics card has what it takes to run Crysis, which was the golden standard for gaming.

YouTuber PCGH Benchmarks & Analysis recently took the Gunnir Arc A380 for a spin in Crysis, which didn't faze the budget Arc graphics card. Do note that this is the original Crysis, which came out 15 years ago, and not Crysis Remastered, which debuted two years ago. Notwithstanding, the decade-old title didn't faze the Arc A380 at all.

The Arc A380 performed admirably in Crysis, delivering over 60 FPS at 1080p (1920x1080) on very high settings and anti-aliasing set to 4x MSAA. We also saw glimpses of the Arc A380 peaking over 80 FPS. Unfortunately, it's public knowledge that Arc doesn't offer outstanding performance outside DirectX12 or Vulkan titles. Nonetheless, the Arc A380 still offered great performance numbers in Crysis, a title that leverages DirectX 10. The YouTuber reported that the game ran beautifully, and we didn't see any frame stuttering.

Intel's current Arc drivers are holding the Arc A380 back, so the graphics card could still have some performance in the tank. Currently, the drivers lack optimization for legacy APIs and require more development to tackle APIs before DirectX 12. However, Intel recently admitted to reusing its integrated graphics software for Arc, which explains why the drivers gimp the performance on Arc Alchemist's discrete graphics card.

At the beginning of the year, Intel had promised to ship river four million discrete graphics cards in 2022. So far, the chipmaker has only released the Arc A380, which is only available in China. The company is still optimistic that it'll be able to get desktop Arc into gamers' hands later this year.

Native DX9 hardware support is officially gone from Intel's Xe integrated graphics solutions on 12th Gen CPUs and A-Series Arc Alchemist discrete GPUs. To replace it, all DirectX 9 support will be transferred to DirectX 12 in the form of emulation.

Emulation will run on an open-source conversion layer known as "D3D9On12" from Microsoft. Conversion works by sending 3D DirectX 9 graphics commands to the D3D9On12 layer instead of the D3D9 graphics driver directly. Once the D3D9On12 layer receives commands from the D3D9 API, it will convert all commands into D3D12 API calls. So basically, D3D9On12 will act as a GPU driver all on its own instead of the actual GPU driver from Intel.

Microsoft says this emulation process has become a relatively performant implementation of DirectX 9. As a result, performance should be nearly as good, if not just as good, as native DirectX 9 hardware support.

This DX9 change from Intel appears to be a very good move as a result. Intel can now divert driver development resources towards DirectX 11 optimizations -- which we know is very bad at this time, and Intel won't suffer performance consequences as a result, with DX9 optimizations "outsourced" to Microsoft entirely.

According to Microsoft, with how performant D3D9On12 is, it will be interesting to see if Nvidia and AMD follow the same path as Intel. But, there could be consequences to the API translation, including higher CPU usage (since the translation is software accelerated) and potential side-effects with older games. Nvidia and AMD also have almost 20 years of driver experience with DirectX 9, which might result in performance losses with the DX12 emulation layer.

Intel, on the contrary, only has experience with DirectX 9 on its integrated graphics, which does not translate into the experience with its much higher-performing discrete graphics. So it makes a lot of sense that Intel is immediately transitioning to emulation as it gets closer to launching Arc worldwide.

In a new video on its YouTube channel, Intel has shared fresh benchmarks of the chipmaker's upcoming Arc A750 Limited Edition graphics card. According to Intel's 1080p and 1440p results, the Arc A750's performance is on the same level as Nvidia's GeForce RTX 3060, one of the best graphics cards.

The cumulative results claim that the Arc A750 delivered up to 3% higher performance than the GeForce RTX 3060 at 1080p and up to 5% in 1440p across 43 DirectX 12 titles. The Arc A750 performed slightly faster than the GeForce RTX 3060 in games that use the Vulkan API as well. Intel recorded up to a 4% performance delta at 1080p and 5% in 1440p in favor of the Arc A750.

Intel hasn't officially revealed the specifications for the Arc A750. However, the graphics card will likely come with 24 Xe cores, 3,072 shaders, and 12GB of GDDR6 memory across a 192-bit interface. In addition, the boost clock speed probably hovers around the 2,300 MHz mark. Meanwhile, Intel used EVGA's GeForce RTX 3060 XC Gaming for comparison, one of the faster custom GeForce RTX 3060 on the market, flaunting a 1,882 MHz boost clock.

Intel benchmarked the Arc A750 and GeForce RTX 3060 identical systems powered by the Core i9-12900K, the current Alder Lake flagship. The testbeds also had 32GB of (2x16GB) Corsair Dominator Platinum RGB DDR5-5200 C38 downlocked to 4,800 MHz memory, MP600 Pro XT 4TB SSD. Intel used Windows 11 and the balance power plan for the tests.

One thing to note is that the Arc A750 was on Intel's engineering driver, whereas the GeForce RTX 3060 used the GeForce 516.59 WHQL driver. Arc Alchemist's drivers are still a work in process, so that could be holding the Arc A750 back in the gaming benchmarks. Intel has admitted that Arc underperforms in older APIs, so the company only used DirectX 12 and Vulkan titles.

Intel still has a lot of time to get its driver act together. The company hasn't committed to a specific timeframe for desktop Arc Alchemist's launch. However, the chipmaker stated in a recent blog post, "Intel Arc GPUs are scheduled for release later this year."

According to a recent Intel Q&A, the company confirmed that driver optimizations for Arc GPUs -- relating to poor performance in DirectX 11 and 9 games -- is going to be a constant work in progress with no end goal in mind. Basically, Intel's lack of experience in the discrete GPU driver space will prevent their GPUs from being competitive with older APIs for quite some time.

This was made very apparent by a review from LinusTechTips, where he saw a 50% performance delta between the DX11 and DX12 versions of Shadow of the Tomb Raider running on an Arc A770. In DirectX 11, the A770 only saw around 38FPS, while in DirectX 12 mode, that frame rate bumps up to a whopping 80 FPS.

For the uninitiated, DirectX 11, Direct X 9, and other older APIs behave very differently from the modern ones like DirectX 12 and Vulkan. These older APIs rely heavily on the GPU driver itself to do a lot of the heavy lifting when it comes to tweaking and configuring lower-level GPU settings unseen by the user. 

This behavior was intentional in an effort to reduce some additional heavy lifting for game developers. As a result, driver optimizations play a massive role in dictating the gaming performance of a GPU with these older APIs. 

This is a night and day difference compared to DirectX 12 and Vulkan, where a lot of this driver baggage has been transferred to the game engine itself, with game developers being responsible for handing lower-level optimizations such as video memory allocation (this is why DirectX 12 and Vulkan are referred to as "low level" APIs).

The bad news for Intel is they have very little experience with these APIs surrounding discrete graphics (in comparison to iGPs). Nvidia and AMD, on the other hand, have more than a decade of experience in the field and know all the little details and odd behaviors DX 11 and DX 9 might have.

As a result, Tom Petersen from Intel says the road towards better performance in APIs like DirectX 11 will be a "labor of love forever." It is a sad truth, but a truth nonetheless. These optimizations don't happen overnight, and there are infinite ways to optimize GPUs for DirectX 11 and its predecessors. This fact holds true even for experienced companies like AMD, which has seen big DirectX 11 driver gains in recent years.

Integrated Graphics Experience Has Made Things Worse for Intel

At first glance, it's easy to assume Intel's experience with integrated graphics would be beneficial. But unfortunately, it has not helped matters and has even made things worse for the company.

In a report we covered a week ago, CEO Pat Gelsinger noted that it made a fatal error on the driver side of development and falsely assumed that it could take its integrated graphics driver stack and apply it to its discrete Arc GPUs.

This strategy showed Intel that its integrated graphics driver stack was utterly inadequate to run Intel's much more powerful Arc GPUs since the architectural differences between its iGPs and dGPUs are massive.

We suspect this could be a big reason Intel's Arc GPUs suffer exceptionally in Direct X 11. If Intel had started from scratch with a dedicated GPU driver stack, the developers would have had more time to optimize for older APIs.