I'm not sure I could use Xfire with such disparate cards but No these are running independently.



buffers are set at 0.5 days and 0.7 days ( a rough setting that just works so I did not try anything else)

My only interest in having a large cache is in keeping these running during "server maintenance" so if any of those folks that decide these limits are reading.... Could you please take a look at the downtime statistics and adjust the Maximum cache accordingly? .... Pretty please

EDIT:
My assumption is that the fastest of cards runs at no more than 120 wu's an hour. Does anyone get more? (info needed for setting this figure)

I have only a buffer of 90 wus or so on a 7950 so be happy with your 300 per card ^^ i dont get more units..

On the 2P octo core with dual 7950's, I seem to be doing one wu in about 9-13 minutes. So with 24 simultaneous GPU threads, I think that comes out to a bit more than 120/hour since 12min/wu on 24 threads would be 30 seconds per wu.

My current 'ready to start' back log is only 17 tasks even though I have it set to a cache of 10 days for all of my GPU only machines. So yeah, it would be very nice to see a change to how the cache is handed for high HCC-GPU throughput rigs.

I know that I said my buffer went up to around 2*600 some hours ago but now is back to just above 600 again...what to do? Well I better live with or go back to folding if the problem persists so my workshop stays warm and some science gains.

Well, things of fringe relevance will be overlooked (ignored). Even when your host "projects" them HCC-GPU jobs at 3.4 hours (riddled since my measly Q6600 does them CPU version in under 2 hours, and projects them at 2), guessing you run a CPU/GPU mix, you'd still be able to cache allot more if only you did what I proposed... set that cache/buffer to a saner level. A dual Octo with one day should have at 3.4 hours about 112 tasks queued. Others are getting their max, so what else is different, what science mix are you running on that device?

edit: Are these tasks running in High Priority? (It would if either Connect about every / Minimum Work Buffer is set to 10 days). And for client 7.0.xx, the Additional Work buffer only works for 1 work fetch... As soon as the buffer is above the Minimum setting [if you have that at zero or very low], the clients stops fetching till it drops again below minimum.

Saturday night live, tell us tomorrow what happened when you've decided to reduce the Minimum to the suggested 1.0 days and the additional buffer to e.g. 0.50 days value.