Friday, August 1st 2025
Intel Opens Up XeSS 2 to NVIDIA and AMD GPUs with SDK 2.1.0
Intel on Thursday released the XeSS 2 SDK version 2.1.0, with which it announced support for non-Intel GPUs. This would be any GPU that supports Shader Model 6.4, which are GeForce GTX 10-series "Pascal" or newer from NVIDIA, and Radeon RX 5000 RDNA from AMD, or newer. The update opens up not just super resolution (XeSS SR), but also frame generation (XeSS FG) and low latency (XeLL), however, using XeLL requires XeSS FG to be enabled, it cannot be exclusively used.
The release of XeSS 2 SDK would make it possible for game developers to update their binaries to support XeSS 2, and expose the feature-set to even non-Intel GPUs. At a hardware-level, XeSS 2 would rely on DP4a compute to run its super resolution and frame generation models. Only on Intel Arc "Alchemist" or "Battlemage" GPUs will the technology leverage XMX accelerators.
Source:
Intel Graphics
The release of XeSS 2 SDK would make it possible for game developers to update their binaries to support XeSS 2, and expose the feature-set to even non-Intel GPUs. At a hardware-level, XeSS 2 would rely on DP4a compute to run its super resolution and frame generation models. Only on Intel Arc "Alchemist" or "Battlemage" GPUs will the technology leverage XMX accelerators.

14 Comments on Intel Opens Up XeSS 2 to NVIDIA and AMD GPUs with SDK 2.1.0
Currently on a rx7800xt so no need for an upgrade but when that time comes, I hope Intel has something interesting to go for.
Also while this is excellent news I just keep saying this, this is SUCH a boring time in gaming, pls lets just have 1 upscaler that fits all, one that works on older hardware but detects support on newer cards for more advanced features (shadermodel 6.4+ / AI cores) and then run the better version on those regardless of brand......
I just hate this mismash of support and then, if its supported, what version, its just so silly.
Like I played KF3's server test and FSR looked like crap, I have no clue what version it used but it just sucked, XeSS was much better (kinda blurry in motion but good enough)......and yet FSR (which we know is atleast V3 here) looks prefectly fine for 95% in GTA Enhanced.
Just these different versions and then different versions within the same ffing upscaler....its a mess, I want it to be over already!
Seems the entire ARC divison is staying in front of the great scythe. The new boos can axe the entire GPU team in any second.
Also, worth noting while Intel supports every GPU with Shader model 6.4, they did explicitly recommend RTX 30 or RX 6000 series or above.
Is this still true for XeSS 2 upscaling?
Regardless, for AMD handhelds with RDNA3+ forcing XeSS through optiscaler might be an excellent choice.
Someone actually did "hack" FSR 4 to work on RX 7900 XTX on Linux and it was barely faster than native at 4K Quality, at lower output resolutions it didn't offer performance benefits over native.
Or maybe FSR 3.1 is not compatible with Nvidia's libraries or Nvidia's financial support.
I was reading this yesterday
RDNA 4's Unreal Engine 4 ray-tracing stutters may not be AMD-specific | Tom's Hardware
with the subtitle
"Even Intel's B580 suffers from the same stuttering issue."
Then in the article with big letters we can read thisand I have to wonder what kind of incompatibilities do developers find with newer AMD(and Intel) software and hardware when using proprietary libraries from Nvidia. Maybe those libraries are not just unoptimized for use with AMD or Intel hardware, they are also incompatible with anything new or somewhat new from the competition. Maybe those libraries aren't updated often enough to offer better compatibility with latest versions of FSR or XeSS, or be 100% compatible with RDNA4 or Battlemage, so developers don't integrate AMD or Intel features fast enough to avoid ending up with a problematic game.P.S. The comments in the article are sad and funny at the same time. People just realizing what is happening the last 15 years.
AMD is also partly responsible, they could always offer to help with the implementation. It is possible that they did, but we would never know.
Actually, with AMD having declared themselves a software company they should have done it already.