Thursday, August 21st 2025

AMD FSR Leak Accidentally Tips Potential Support for Older GPUs

AMD recently released its FidelityFX Super Resolution SDK 2.0 to accompany the shift to the machine learning-based FSR 4, but in a bizarre twist, the company also seemingly accidentally open-sourced the libraries to FSR 4 via its AMD GPUOpen project, giving tech-savvy browsers a sneak peek at the inner workings of FSR 4. The files have since been removed from AMD's GPUOpen library, but quick fingers managed to snag screenshots of the file directory, which reveals enough. One of the biggest takeaways was the inclusion of 8-bit integer (INT8) libraries, which use less resources but are also less precise than those used in RDNA 4, indicate that there is or may have been some intention to release FSR 4 for AMD's RX 7000 GPUs. FSR 4 was initially announced to be exclusive to AMD's Radeon RX 9000 series GPUs, but that may not be the case for long.

There have previously been hints that AMD FSR 4 could be coming to older RDNA hardware, and there have even been mod projects that have successfully forced it to work on older hardware, but this is a substantial official signal that AMD may be working on porting it to older hardware. There is also speculation that the INT8 libraries may be destined for the PlayStation 5 Pro, but the file type seems to indicate otherwise—PS5 uses .PSSL, while these shaders were packaged as .HLSL, suggesting they were for PC use.
Sources: @uzzi38 on X, VideoCardz
Add your own comment

28 Comments on AMD FSR Leak Accidentally Tips Potential Support for Older GPUs

#26
shuffy_losys
Hecate91The issue isn't subjective at all and IMO it's valid criticism when there are numerous titles that have come out being terribly unoptimized for the level of hardware it takes to run a game well.
TomorrowThere's certainly some of that, but if a person has a 5090 and 9800X3D and they cant get a stable 60fps with a smooth frametime graph, then it's justified to call that game unoptimized. Not just by that user, but by others as well. Unless it's some super-duper path traced next Crysis or something that looks stunning.
The problem is still that even if we have these numbers, it's still subjective. Maybe someone worth a gajillion dollars doesn't care about how much they have to spend on hardware and they don't care as long as they get 60FPS with it.

Anyway, I'm going to stop here because we can argue until the heat death of the universe about what "optimized" means.
Posted on Reply
#27
tpa-pr
If AMD add official support for FSR4 on my RDNA3 card and/or Intel/Nvidia cards it's another tick in the "why I buy AMD" column. I like their open technology philosophy.

The ironic part is that FSR4 support might come to RDNA3 by the time I've upgraded to UDNA, but we'll see!
Posted on Reply
#28
ValenOne
mrnagantPS5 Pro is 300 TOPS INT8. Xbox Series X is 97 TOPS INT4. X supports INT8 as well, so maybe in the 50 TOPS range.

Radeon 9060 is (default/boost) 97/171 TOPS INT8.

PS5 Pro has the same INT8 throughput as the 9070 XT!

DLSS/XeSS use INT8. It seems like with Blackwell, DLSS started supporting FP8. XeSS also supports FP8. FSR2/3 doesn't look to support FP/INT8, but rather uses FP16. (Might be why quality is meh, to keep it performant, and supports basically all GPUs). FSR4 supports FP8 with some work on INT8. So really, FSR4 could technically run on anything that supports INT8 IF they supported INT8. I imagine the PSSR will likely be the INT8 implementation of FSR4.
Heck, RDNA2 even supports INT8.

This would be why RDNA 3.5 doesn't support FSR4. No native FP8 path.

Wonder why AMD hasn't ported to INT8 when XeSS and DLSS use it. Is FP8 a better path forward than using INT8, so wanted to slingshot themselves forward rather than play catchup? Then backport to INT8?
So, seems like FSR is going with a walled garden now, even though Intel and Nvidia can both run FP8. FSR3 is the fallback. :P
Read www.amd.com/en/products/graphics/desktops/radeon/9000-series/amd-radeon-rx-9070xt.html

For RX 9070 XT
Peak 8-bit Precision (INT8 Matrix) Performance = 389 TOPs
Peak 8-bit Precision (INT8 Matrix) Performance with Structured Sparsity = 779 TOPs

Try again.
Posted on Reply
Add your own comment
Nov 19th, 2026 06:44 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts