Folding Forum

If I'm right, E-cores mostly just lack some power-hungry SIMD features. All things that CPU folding needs, but what about GPU folding? All the core needs to do is poll the GPU in a busy loop and do occasional sanity checks. Would an E-core be able to keep up without starving the GPU?

I am thinking about having a few RTX 5080 GPUs and pinning the threads to the E-cores, and then reserving the P-cores for folding. Is this feasible or would an E-core not be able to keep up for some reason?

You can use any CPU for GPU folding.
However Nvidia GPUs still get upset if any of the CPU cores are being used for CPU folding.
By upset, I mean lower performance.

It's probably not too much work to get the Nvidia GPUs happy, at the expense of isolating a few cores from the scheduler entirely.

At least that works for a mobile Nvidia GPU I had access to and, unless my understanding of the kernel scheduler is totally wrong, should work for anything else, as long as it's not overheating and throttling the clock.

I've tested it a bit more and find that P-cores are actually a bit more efficient for feeding the GPU. I've also managed to prevent CPU folding from interfering with GPU folding. As long as the threads are pinned and the GPU core is not losing boost because of heat from the other cores, there's no PPD impact.