Quote from: cryptoyes on August 09, 2019, 11:04:48 AM

Your miner gets gradually slow on systems with many GPUs. It starts off with higher hashrate for the first few minutes and then slowly drops by up to 30% in speed. This is a sign of too much driver comms, keeping the CPU swamped with irq requests (Gminer had the same problem a few months ago but they fixed it).

You should find a way to drop CPU utilization and driver usage. You should also add an option to disable nvml polling of temperature, fans, power etc.

13x gtx1070, i3-7100, 8 GB ram, Ubuntu x64, nvidia 415.27, cuda 10.0

Gminer 150,5: 280-290 H/s, steady. CPU utilization is 40-50%. I can communicate with the GPUs while it's mining as usual.

miniZ 150,5: starts at ~250 H/s for a about 1 minute, then gradually drops to ~180 H/s within ~2 minutes. CPU is kept at 100% at all times and any GPU command issued hangs (e.g. nvidia-smi hangs entirely). The nvidia driver is totally hogged.

I tried lowering --intensity and the CPU utilization drops, but the hashrate drops significantly too, even if I use --intensity 99. Max I can get is about ~195 H/s.

I also tried --oc1 and --oc2. Also tried --telemetry 0 (doesn't disable nvml polling). Nothing made any difference (even though I have them overclocked and at lower power limits).

I'm willing to test a beta if you're willing to work on this. PM me. I manage a small farm.

EDIT: Retested Gminer on 150,5. Gets 280-290 H/s steady, with 40-50% CPU utilization.

Hi cryptoyes,
thank you so much for your availability.
We have been unable to reproduce this problem. However, we made a few changes when introducing the new Beam algo, for the next release.
Maybe you can try it and let us know if you see a difference in cpu usage. It should be out within ~24h.
We also introduce an option to disable nvml.
Cheers