gpuowl tuning

[M344587487]

Right now mem clock ceiling is around 900MHz. I know the card can go further because I've done it for mining but that was on windows and the tools just worked. The best so far at 1401MHz core is 0.945V core, 900MHz mem, 2.34ms/it drawing 142W. Compared to 2.47ms/it at 100W it's way worse. There's more to do between 1269 and 1401MHz but I think I will have to flash with Vega 64 bios to progress further and probably will. It's a shame the bios are signed though.

[xx005fs]

Originally Posted by M344587487

Right now mem clock ceiling is around 900MHz. I know the card can go further because I've done it for mining but that was on windows and the tools just worked. The best so far at 1401MHz core is 0.945V core, 900MHz mem, 2.34ms/it drawing 142W. Compared to 2.47ms/it at 100W it's way worse. There's more to do between 1269 and 1401MHz but I think I will have to flash with Vega 64 bios to progress further and probably will. It's a shame the bios are signed though.

That to me sounds like Hynix HBM as my card with Samsung could overclock way further than that even with Vega 56 stock BIOS. I reckon that you can change the bios switch to power saving bios and try to lower power even further.

Originally Posted by preda

I use ROCm 1.9.1, Ubuntu 18.04 with Linux kernel 4.18.8.
With dual Vega64 (air with the standard "blower" cooler).
Here are my observations:

1. ROCm is in general faster then amdgpu-pro (better compiler, producing better ISA code).
2. My sweet-spot is p-state 5 (rocm-smi --setsclk 5), which results in 1401MHz, GPU fan at 2300 RPM (automatic), 150W power, 75degC temperature.

If I set the frequency higher (p-state 6, or 7, or automatic (default)), the GPU quickly reaches 82-84 decC and there does thermal throttling. This thermal throttling results in worse performance then p-state 5, so it's a lose-lose: higher temperature, higher power use, lower performance.

I do not set the fan speed manually, I leave it on automatic, which is enough cooling for 150W with 75C.

Impressions of ROCm + RX580 on Debian:
1. faster than amdgpu-pro for large exponents
2. automatic and default settings, core clock 1319MHz, memory clock 2000MHz, fan 2130 rpm, power 144W, 77 deg C

Originally Posted by preda

I use ROCm 1.9.1, Ubuntu 18.04 with Linux kernel 4.18.8.
With dual Vega64 (air with the standard "blower" cooler).
Here are my observations:

1. ROCm is in general faster then amdgpu-pro (better compiler, producing better ISA code).
2. My sweet-spot is p-state 5 (rocm-smi --setsclk 5), which results in 1401MHz, GPU fan at 2300 RPM (automatic), 150W power, 75degC temperature.

If I set the frequency higher (p-state 6, or 7, or automatic (default)), the GPU quickly reaches 82-84 decC and there does thermal throttling. This thermal throttling results in worse performance then p-state 5, so it's a lose-lose: higher temperature, higher power use, lower performance.

I do not set the fan speed manually, I leave it on automatic, which is enough cooling for 150W with 75C.

Originally Posted by SELROC

Impressions of ROCm + RX580 on Debian:
1. faster than amdgpu-pro for large exponents
2. automatic and default settings, core clock 1319MHz, memory clock 2000MHz, fan 2130 rpm, power 144W, 77 deg C

At 1319MHz core clock it is in ROCm profile 6.
GpuOwl Version 5.0 https://github.com/preda/gpuowl/comm...a518a6913e9cff


Timing 4.48 ms/sq with 88M exponent.