Folding Forum

Hello

I've noticed that Radeons have a fairly low PPD number compared to GeForces in the same range.

The Radeon RX 7900 XTX performs much worse than an RTX 4070 in most cases. Considering its specs, I'm surprised.

Is there an optimization issue with AMD GPUs? Or is the RDNA architecture unsuitable for this type of computing?

Thanks

Optimisation issue

Optimisation issue from AMD or FAH/OpenMMM?

TRINITAS wrote: ↑Thu Oct 02, 2025 6:57 pm Optimisation issue from AMD or FAH/OpenMMM?

FAH

FAH is using opencl for AMD, and CUDA for nVidia.
OpenMM now supports HIP, but FAH would need to build HIP Fahcore.
HIP FAHcore has been in development for years now, still don't know when it will come out

Okay. I understand better.

I'm surprised that OpenCL isn't suitable. Some people tell me it's an API that doesn't evolve.

TRINITAS wrote: ↑Thu Oct 02, 2025 7:16 pm Okay. I understand better.

I'm surprised that OpenCL isn't suitable. Some people tell me it's an API that doesn't evolve.

That API is long dead

I imagine it's the same phenomenon for Intel ARC B, that FAH uses Intel's API (OneAPI if I remember correctly)

TRINITAS wrote: ↑Thu Oct 02, 2025 7:59 pm I imagine it's the same phenomenon for Intel ARC B, that FAH uses Intel's API (OneAPI if I remember correctly)

No, FAH uses opencl for Intel

I meant, does Intel have its API? (Like HIP for AMD, or CUDA for NVIDIA)

TRINITAS wrote: ↑Fri Oct 03, 2025 8:08 am I meant, does Intel have its API? (Like HIP for AMD, or CUDA for NVIDIA)

Yes, they do. OneAPI, as you mentioned