Intel XeSS Explained: Everything you need to know about Intel’s Upscaler

MicroCenterOfficial

Intel might be a third player in the graphics card industry, but it’s anything but a third wheel. With a capable range of mid-range graphics cards, some impressive advances in driver performance, and exciting technologies that make Intel’s GPUs that much better, it’s got a lot going for it. The keystone of Intel’s GPU technologies is Xe Super Sampling, or XeSS, Intel’s dynamic upscaling algorithm. Its own version of NVIDIA’s Deep Learning Super Sampling (DLSS) and AMD’s FidelityFX Super Resolution (FSR).

XeSS is the new kid on the block with a lot to prove, but it’s already making waves for how effective it is as an upscaling solution, and for its wide-ranging support. With Intel’s current focus on more affordable graphics cards and its enormous install base of onboard GPUs, this kind of technology is keenly important. Fortunately, it’s already doing well, and is getting better all the time.

Here’s everything you need to know about Intel XeSS.

What is Intel XeSS?

XeSS is a technology that is able to raise the perceived resolution of a game to make it look more detailed, without the usually associated cost of doing so. Where 4K resolution is a lot more demanding on a graphics card than 1440p, XeSS can make it possible to run the game at something approximating native 4K without the added overhead.

XeSS achieves this by actually rendering the game at a lower resolution and using its artificial intelligence-assisted algorithm to upscale each frame to make up the shortfall in detail. This can be used to improve the FPS of a game or play it at a resolution your graphics card otherwise wouldn’t be capable of. You can also use it to give you more FPS headroom for enabling other graphical settings, like ray tracing, which would otherwise tank the frame rate.

There are actually two versions of XeSS. There’s the one that runs on Intel’s own graphics cards, with added acceleration from the dedicated XMX (Xe Matrix Extension) cores found only on Intel’s GPUs, and a version that runs on graphics cards from other manufacturers and on older Intel GPUs without XMX cores. The Intel version is more efficient thanks to hardware acceleration, but also leverages a more advanced upscaling algorithm, delivering better visuals.

Both versions of XeSS offer the same range of quality modes to pick from, though. They range from close-to-native resolution with mild upscaling, to heavier upscaling on much lower resolution source frames. The former has a much lighter effect on performance, but doesn’t exhibit upscaling artifacts in the same way, while the latter provides a much greater uplift in performance, at a greater effect on visual acuity.

Like NVIDIA and DLSS, Intel has also performed some algorithmic training of its XeSS technology, letting it take rendered frames, upscale them, and then compare the output against a higher-resolution rendered image to see how effective it was. It could then try again, by smoothing out the errors, artifacts, and problems, and then again, and again, until it learns how to better upscale the various frame and data inputs it receives.

This is what makes XeSS capable of working across a wide range of games and not just on ones where the algorithm was specifically trained on that game – as was the case with the earliest versions of competing upscalers.

How does XeSS work?

On the surface, XeSS works in a similar manner to AMD’s FSR and NVIDIA’s DLSS. It takes a range of inputs from the game, including a frame rendered at a lower resolution – say 1080p – and uses a combination of machine learning and those various inputs to output a frame that looks like it was rendered at a higher resolution – like 1440p or 4K.

Under the hood, XeSS is a little more complicated. It draws on several key pieces of information to generate the higher-fidelity frames. It starts with a lower-rendered-resolution frame from the game, combines it with motion vectors – effectively, the player’s different perspectives between frames – and frame history elements, like color grading, lighting, and depth data, to inform what the eventual frame should look like.

All of that data is maintained as an ongoing “history” of the game, and they better inform the next frame that XeSS upscales, and the next one, and the next one. That lets the algorithm perform better after successive frames have been upscaled.

As one final pass before pushing the frame to your screen, XeSS performs temporal anti-aliasing (TAA) on the frame. That smooths out any jagged edges that the algorithm spat out and replaces the in-game anti-aliasing technology, letting it handle all of the rendering in one pipeline.

All XeSS modes, whether they’re running on Intel graphics cards, or somebody else’s, and whether on the highest performance mode, or the most ultra of quality settings, work like this. However, where non-Intel GPUs generate those higher-resolution frames using generic DP4a instructions (which are useful for AI calculations and can run on any hardware) Intel’s XeSS completes it using a more advanced algorithm that leverages its XMX cores.

Intel hasn’t revealed the precise mechanism for how this alternative algorithm works, but it leverages machine learning and the AI acceleration of its XMX cores to deliver what it claims is a higher quality image, whilst retaining the same performance advantages of the standard XeSS algorithm.

Intel updated XeSS to version 1.1 in March 2023, introducing a number of important changes to the way it works. It accelerated Intel’s XMX kernels, allowing for greater performance from XeSS on Intel GPUs – some showed a 25% performance improvement using XeSS 1.1 over XeSS 1.0. It also improved the algorithm speed on non-Intel GPUs by developing faster DP4a kernels. That’s particularly important, since it not only means NVIDIA and AMD GPUs get greater benefit from XeSS, but that gamers using Intel’s integrated GPUs on its processors can see improved performance, too.

XeSS 1.1 also introduced a new upscaling model that was better equipped to weed out errors in temporal stability, such as flickering and ghosting.

XeSS modes: From Ultra Quality to Performance

Like other upscaling algorithms, XeSS has several performance modes you can pick from that range from close to native quality with a comparatively minimal improvement to performance, to a much lower rendered resolution with a big improvement in performance, but with more obvious visual artifacts.

There are five XeSS quality preset modes you can select from. They are as follows:

• Ultra quality: This has a render scale of 77% and a scale factor of 1.3x.
• Quality: This has a render scale of 66% and a scale factor of 1.5x.
• Balanced: This has a render scale of 58.8% and a scale factor of 1.7x.
• Performance: This has a render scale of 50% and a scale factor of 2.0x.
• Ultra performance: This has a render scale of 33.3% and a scale factor of 3.0x.

Note: While these are the standard presets for the various XeSS quality modes, custom parameters can be set by a game developer. That means that in some games, the precise input resolution for the different quality modes may vary from these figures, and the modes available to you may be fewer.

As an example of how you might use one of these quality modes, if you were targeting 4K resolution in a game and utilizing ultra-quality mode, the input resolution from the game would be 2,957 x 1,663 (77% of the output resolution). It would then be upsampled by XeSS at 1.3x its input resolution, to deliver a full 4K resolution of 3,840 x 2.160.

Extrapolate that out to the other quality modes and you’ll see that the difference between input and output resolutions increases dramatically as you stray into the performance modes. Ultra Performance is beyond anything offered by NVIDIA or AMD’s upscaling algorithms and is unlikely to see much use since it will have the greatest impact on the game’s visuals, but it does offer those running onboard graphics the option of playing more demanding games by significantly cutting down on the input resolution.

In reality, most gamers utilize the Ultra Quality and Quality modes, since you can’t tell much of a difference between them and native resolution, and yet they still deliver an impressive uplift in performance.

What graphics cards do you need for XeSS?

One of the greatest strengths of XeSS is that it works on just about any GPU, whether that’s a high-end graphics card from one of the latest generations, an entry-level card from a few years ago, or even onboard graphics on your CPU. If you have a GPU, XeSS should be able to give it at least a minor uplift in performance. That’s particularly useful for anyone trying to game on integrated graphics.

XeSS works on AMD graphics cards, NVIDIA graphics cards, and Intel graphics cards, as well as onboard GPUs from both Intel and AMD. However, that’s only with the standard XeSS algorithm using DP4a instructions. To make use of the more advanced, accelerated XeSS algorithm, you need to run it on an Intel Arc graphics card with XMX AI accelerators.

At the time of writing, the only desktop graphics cards that offer that support are:

• Intel Arc A770
• Intel Arc A750
• Intel Arc A380

Intel XeSS is also available on certain mobile graphics chips, including:

• Intel Arc A770M
• Intel Arc A730M
• Intel Arc A550M
• Intel Arc A370M
• Intel Arc A350M

How do you enable XeSS?

Make sure you have a compatible graphics card or chip as listed above, and the latest Intel GPU drivers – download them here – and are playing a game with XeSS support.

Open the game’s Settings menu and navigate to the graphics settings. You may have to look through the menu to find it and it may be hidden in Advanced settings or something similar. Find the XeSS toggle and switch it to “On”. You may also be given a choice of quality modes. Most gamers prefer to use Ultra Quality or Quality modes but try playing around with it to see which you like the most. You’ll get greater performance from those with greater render scales but be aware that that does introduce greater chance of visual artifacts.

Some games will let you adjust the sharpening XeSS implements, too. This can be useful if the game implements a heavy-handed anti-aliasing solution that loses some detail around the edge of objects. Drag the slider around until you find a point you’re happy with.

The future of XeSS

Intel hasn’t provided much in the way of hard detail about its future plans for XeSS, other than they exist. Considering how much emphasis both AMD and NVIDIA are putting on the development of their own upscaling algorithms – and the positive effect it has had on in-game performance – it seems almost certain that Intel will continue to invest heavily in XeSS. It will go hand in hand with the development and launch of its future generations of graphics cards.

Intel is also in a unique position with its dominance of the mobile market meaning there are many millions of Intel GPU-equipped CPUs out there. XeSS has the potential to help many more of those laptops play games than ever before if it continues to improve onboard GPU performance.

One area where Intel may continue to develop XeSS is with frame generation. This is a technology that has been introduced by NVIDIA in the latest version of DLSS 3, and AMD is planning to introduce with FSR 3 later in 2023. At the time of writing, XeSS doesn’t offer any kind of frame generation, but considering it already leverages AI hardware for the existing XeSS, it’s no stretch to imagine Intel developing its own frame generation technique in the future.

Which games support XeSS?

At the time of writing XeSS is supported in over 50 current-generation games. They include standout titles like:

• Dying Light 2: Stay Human
• Hogwarts: Legacy
• Grid Legends
• Gotham Knights
• The Rift Breaker
• Returnal
• Starship Troopers Extermination
• Forza Horizon 5
• Ghostbusters: Spirits Unleashed
• Farming Simulator 2022
• Enlisted
• Death Stranding
• Hitman III
• Shadow of the Tomb Raider

Intel has promised many more supporting games in the future, too.

More from the Micro Center Community:

Looking for more information about Intel GPUs? We’ve revisited the Arc series recently, as well as benchmarks for the Arc A770 and launch breakdowns! And if you can’t find what you’re looking for, don’t hesitate to post a new discussion and the Community will be happy to help!

McMuffin719

This is an awesome review of Intel's XeSS upscaling features, as well as its promising competition against NVIDIAs and AMDs standings for the mid-range market. Good job!

ChaosDTV

the arc cards are becoming a new fav of mine

Vladdy

Super useful information! Go Intel. It's nice to have more competitors.

-V

Shoan041

That awesome! I can't wait to see where intel goes with GPU technology in the future!

ricardo_ortega

I didn’t know that XeSS is compatible with not just intel cards but with AMD and Nvidia as well but only with limited features since it’s more for the arc cards. Good to know.