I'm at ~1.9 M PPD currently, from 3 Pascal GPUs.
2 more GPUs to come as I find time to put it all together.
With all stops removed, the 3-GPU host may be good for ~2.3 M PPD at a bit less power efficiency, but dramatically increased coil whine. The latter is a problem. I am thinking about rearranging which hosts are in which rooms to maybe run the GPU host in the noisier mode for extended time.
I'll put CPUs towards F@H only temporarily as a test. I think last year they reachedcirca almost Maxwell efficiency (PPD/W) with the right WU types; I want to find out whether it's worse or better this year. (Edit: In F@H, near GPU-like efficiency on CPUs is only possible with high core count servers and requires Linux, such that a single folding slot occupies all hardware threads.)
The new 2-GPU host is meant to be running Linux; if this turns out workable I may set up a Linux partition on the 3-GPU host too. Also I am thinking about pulling the Xeon from it and putting an i7 in there, like it had originally before a sad accident, but am still undecided.
2 more GPUs to come as I find time to put it all together.
With all stops removed, the 3-GPU host may be good for ~2.3 M PPD at a bit less power efficiency, but dramatically increased coil whine. The latter is a problem. I am thinking about rearranging which hosts are in which rooms to maybe run the GPU host in the noisier mode for extended time.
I'll put CPUs towards F@H only temporarily as a test. I think last year they reached
The new 2-GPU host is meant to be running Linux; if this turns out workable I may set up a Linux partition on the 3-GPU host too. Also I am thinking about pulling the Xeon from it and putting an i7 in there, like it had originally before a sad accident, but am still undecided.
Last edited: