AMD’s “PEPS” Research Pushes Neural Texture Compression Further, Cutting Model Parameters By 25% At Comparable Quality

“Applications of PEPS Beyond Neural Texture Compression​

Beyond neural texture compression, this new technique also has scope for use in SDFs (signed distance functions), which are used in 3D rendering. SDFs are notorious for requiring high-res grids that gobble up VRAM, so optimizing memory usage through lightweight neural compression models is important. When testing on the Pitted Stonefish SDF, Grid-PEPS was able to roughly match the IoU (Intersection Over Union, or how closely the reconstructed 3D shape overlaps the original) of non-PEPS methods with 8x more encoder parameters.

Implication For FSR Suite: Don't Expect To Run This On Your Radeon GPU Any Time Soon​

While this research is certainly interesting from a technical standpoint, it's hard to tell when it will become relevant for consumers. Currently, only NVIDIA has any kind of publicly available toolkits/demos for Neural Texture Compression, and there isn't a single game out there with a full NTC implementation. On the AMD side, support is even more sparse - in fact, AMD hasn't even officially come up with a brand name for it yet; in all their research, they use generic terms to refer to the technology. Regardless, it's good to see progress in this domain, especially as the RAMpocalypse ensures we'll be seeing 8GB GPUs well into the latter half of this decade.

News Source: AMD GPUOpen“

Source: https://wccftech.com/amd-peps-resea...rameters-by-25-percent-at-comparable-quality/

AMD is doing some good work here, Intel and Nvidia still appear to be ahead on the subject.

I really hope this doesn’t devolve into yet another competing standards situation.

Because if it affects compatibility then we don’t win.

Just means it'll be on PS6. Wouldn't shock me if it was exclusive to PS6.

DukenukemX said:

Just means it'll be on PS6. Wouldn't shock me if it was exclusive to PS6.

Click to expand...

This seem to be pure standard math optimisation that current gpu can run well enough (what added versus the current NTC model, if you can run current NTC adding this to it should not be an issue).

Anything major for the PS6 I doubt could be added now, high level ahrdware decision must be locked by now.

LukeTbk said:

This seem to be pure standard math optimisation that current gpu can run well enough (what added versus the current NTC model, if you can run current NTC adding this to it should not be an issue).

Anything major for the PS6 I doubt could be added now, high level ahrdware decision must be locked by now.

Click to expand...

Nah not gonna come to the PS5 or any of the other platforms any time soon.... maybe the Switch 2....

AMD claims that the algorithms aren't matrix multiplication-based, which would make you think they could maybe be backported to RDNA 2, but later on describes how the algorithm does make extensive use of AMD's WMMA functionality... WMMA, for those not familiar, stands for Wave Matrix Multiplication and is only available on RDNA 4 and up.

It could, in theory, with a lot of work, get backported to RDNA 2 like they are trying for FSR4, but considering they are another year or 2 out from getting this into the hands of the developers, which would then require additional years of working with it to get it into their titles, I suspect they have no intention of backporting it.

I mean, would I complain if they did? Certainly not! But I can't see them doing it, unless there is still a LOT of RDNA 2 GPUs kicking around in the wild some 4+ years from now, and if that is the case, then AMD has bigger problems.