One of my clients is behaving in an unusual way; I have 3x cards in this machine. Two are GTX 970s and one is a GTX 980. For starters, the 980 seems to be misread by the client, because when running properly, the card which produces the numbers that I would expect from a 980 (400K+ PPD) is labeled as a 970, and the client which is detected as being a 980 is only producing the numbers you would expect from a 970. Now on to the second part... This machine should be capable of 1M PPD, and to get it there all three cards were running with a slight overclock. They ran well like that for several days at the end of last week and through the weekend. Late on Sunday night I noticed that one of the clients had been crashing and returning bad work units, and not knowing which card was which client, I immediately dropped the clocks on all three back to stock and forced the slot to request new work. It was assigned work, downloaded it and began folding again, and has been stable ever since. But, for the past day and a half it's been stuck at no more than 70-80K PPD because the GPU core is not being fully utilized. Can anyone explain why the client is behaving this way or suggest how I can get it back to normal?
I'm afraid I've not tried 3 GPU's in the same box ... I think some of the other folks have though, so maybe they can help out ...
No worries Doc; I'm now running three machines with three cards in them and it's the first time I've come across this. Thankfully I have got it behaving properly again as far as output goes, but have no idea what actually causes this. I reinstalled the client, drivers and rebooted the system but the F@H control window still identifies the slot which is clearly running on the GTX 980 as a GTX 970. Anyway, I'm seeing 1.1M PPD on that box again. so I'm happy enough to leave it like that.
