F@H GPU Bug?

ChrissyT88 · 22 Jun 2009 at 15:56

Hhmmm... Not too sure what to make of this. New computer has been running as per usual all day. Running on Server 2008 R2 RC1, two virtual machines are folding in hyper-V and a single GFX card client is running. Everything is normal....

I fired up handbrake and set about encoding an MKV file. Everything goes well, CPU usage jups to 100%, VMs slow down as expected, and all if fine. I opened FahMon to check the clients and noticed a 511 point WU was running a little faster than normal (~1000 ppd greater), but was at 97% completed. However, the next WU is something else entirely...

Im all for a single client giving 9000 ppd, but i know for a fact that these units typically give me closer to 6000 ppd. Interestingly, the ppd value keeps climbing if the client is left running, so something is obviously wrong. Is anyone else running Windows 7 with the latest drivers and experienced this? Im waiting for the encoding to finnish to see if i observe the same thing (which would suggest a dodgy WU), but that could be some time!

didgemaster · 23 Jun 2009 at 10:56

Not seen this myself, however I don't run my GPU clients on a 280 or in server 2008

I assume by now the work unit has finished what did the ppd turn out to be for that WU

*edit* just had another thought that may be relevant - how is your F@Hmon setup, does it average your points over the last frame, last three frames or the effective rate

ChrissyT88 · 23 Jun 2009 at 12:02

In the end i let the encoding finish, deleted the queue.dat file (forcing a restart of the work unit) and it continued to fold as normal, with FahMon indicating about 7000 ppd, which is sort of normal for this particular card.

FahMon is (and i think was) set up to estimate the ppd from only the last frame, although as i mentioned, it just got higher and higher as the WU progressed! Seems like a bug to me - somewhere something is getting confused! I haven't since tried to recreate this, but i may give it a go later on just to see. Its a bit weird!

Snapshot · 23 Jun 2009 at 12:10

I've found that the GPU console client does run faster if there are other applications running (apart from VMware player), typically going from 3700ppd on its own to 4200ppd with another FAH client running.

I can't get the GPU client and VMware player to play nicely together however. There are more than enough CPU cycles for the GPU client left by the VMware player but it runs really slowly. I've tried fiddling with the process priorities without success. I'm sure some-one else reported this here recently.