You can't get full GPU utilization on Windows, at least not on GPUs as large as the 1070. 99 % must have been an extraordinary WU, or only a short-term reading. For average utilization to go above 90 % on large GPUs, Linux is needed. (That's due to some differences in the driver stacks, and it applies to Folding@Home and several but AFAIK not all other GPU-based distributed computing applications.)suddenly, the gpu utilization i read in msi afterburner is around 80-85% instead of 99% like it was on the test run.
That said, I don't know what PPD and what utilization-% to expect for 1070 and 970.
I can only speak theoretically, as I don't have comparable hardware: Zen's SMT generally scales pretty well with independent computing threads/ processes. But I wonder about how good Windows' kernel is in keeping unwanted SMT load off of the core which runs the GPU feeder process. Only way to find out is to watch PPD and GPU utilization over multiple WUs.Ok, got F@H installed and running on my Ryzen R5 1600 rig (stock CPU clocks) with GTX1070ti Gaming (X?) (stock clocks).
It defaulted to 10 CPU threads, should I reduce that to 5? How well does Ryzen's SMT fare with F@H?
But frankly, CPU slots smaller than 24 threads are hardly worthwhile (due to low quick-return bonus and due to receiving too many low-scoring WUs of the 0xa4 type, compared with the better scoring 0xa7 type), unless it's about an entire rack full of such hosts, and electricity is cheap at the place in question.