|
Title: 'nice'/prioritizing multiple opencl processes in same GPU? Post by: HanSolo on June 09, 2011, 04:23:04 AM I've noticed running two poclbm's against the same GPU improves rate about 2%. So far so good.
Now especially the last few days solo-mining into bitcoind 0.3.21, the bitcoind seems slow to respond, causing short stalls ("Problems communicating with bitcoin RPC") on both processes. These are most common when bitcoind accepting inbound connections.. I fear it could be bad-faith peers swamping others with bad traffic. Another thread mentioned pointing the second process at a different pool to fill gaps. Tried that again so far so good. But of the two processes the one that's most busy lately seems to get the most GPU.. for example if one process has even a few seconds head start, it keeps completing most of the hashes. So every RPC stall means almost all hashing power goes to the pool. Is there anything that works like Linux 'nice' for prioritizing processes using the GPU? I really want one process to dominate except when its stalled. I tried regular 'nice' but it had no effect. Title: Re: 'nice'/prioritizing multiple opencl processes in same GPU? Post by: CentroniX on June 09, 2011, 04:44:37 AM If I understood another post correctly set the dominant one with -f1, and the second with -f50. The f1 will always take priority over the f50.... Again, if I understood it right.
Title: Re: 'nice'/prioritizing multiple opencl processes in same GPU? Post by: fnord123 on June 09, 2011, 05:22:27 AM Phoenix doesn't seem to like -f, anybody know how to do this with Phoenix?
Title: Re: 'nice'/prioritizing multiple opencl processes in same GPU? Post by: PcChip on June 09, 2011, 06:04:22 AM Set one at High, one at Above-Normal. It doesn't do 99%/1% like you're wanting, but it seems to do about 60/40 for me.
However even though the miner clients displayed a small increase in total hashing power, the combined displayed hashrate from both pools showed a lower overall rate so I stopped. Anyone know why ? |