Ok, running both the above on one PC. CPU (i7 950) OCed @ 4.2 TFP 32 mins PPD 29k (6900) smp 7 GPU OCed @ 900/1800 TFP 2mins PPD 9k (6801) I'm happy GPU, but is the CPU a low???
My two 920 (@3.8 -smp 8) run about the same TPF with a 6900WU but HFM calculates that to 32000PPD. I don't run a GPU with them but my 460s @725 run between 9300-10300PPD with a 6801WU depending on the processor that is feeding them (and whether I am running a -smp on it)
my 920 (@4.0 -smp 7) gets 27-28k on 6900. my 460's (850/1700) produce just over 11k on p6801. what gpu's you using?
Like StreetSam says it seems that your TPF vs PPD estimation is quite a bit out for a P6900 especially considering your 950 clock speed. For reference you best check the SMP2 calculator at linuxforge.net. It says you should expect 34k PPD with a TPF of 32 minutes. Obviously with -smp 7 TPF increases so for 29k PPD you are looking at a TPF of closer to 35 mins. Personally I wouldn't bother with -SMP 7, increasing the GPU core priority to idle will mean maximum cycles for the bigadv without decreasing the GPU's output . However, I don't have a Fermi card which I hear is CPU hungry. HFM.net seems to calculate PPD factoring the time/date the work unit was downloaded so if the core has been halted at any point the PPD calculation will alter. Also check what calculation mode is selected in HFM.net i.e last frame, last 3 frames, effective rate or total frames. I can't vouch for the 'impact' of GPU3 units on Fermi cards but I'm running my i7 950 @ 4.2GHz/1600MHz for a TPF of 30:51, PPD is 35691 (P6900 unit uninterrupted @ 90%) according to HFM & that is with a GTX260 running a non-Fermi GPU3 WU. The GPU core only adds about 2 mins to the TPF. Running solo my i7 on a P6900 reaches 37k-ish!
The concensus on the Stanford forums is that GPU3 does take out some CPU time, but usually SMP(slightly lower) + GPU gets more PPD than SMP alone. I've seen this happen when I stopped an SMP job for a day. It still completed but with a lower credit. I believe (though I haven't been and checked) that the bonus calculation involves real time since download, and not just active running time. This is because the bonus scheme was originally set up to encourage people to return SMP projects as fast as possible. Just found the formula, it is explained here http://folding.stanford.edu/English/FAQ-PointsNew under the subheading of 'Bonus Point Formula' and below
Tesla, reset both to what you have said, will see where this goes. Ifs its shoddy, I will remove the GPU and place in another machine, just the heat from 3 GPUs folding is a concern.
With an i7-980X running bigadv WUs and two GTX460s both running GPU3 clients, there was a 1.6% 'processor grab' by each GPU, so CPU ppd dropped by about 3% (in the 40k pts range) BUT gained over 20k ppd from the GPU clients - an overall net gain. I guess the slower the CPU and the fewer cores it has, then this 'grab' becomes more significant. For some reason on my 'old' Q6600 with one GTX460, the 'grab' is the proverbial "square root of sod all"! So my scientific explanation is goosed
I'm still shooting for stability at 4.2 myself so good luck. Dude regarding the core priority btw, it's idle for the CPU & low for the GPU. Easy to mix up without checking.
All good and work AOK: i7 950 @ 4.2 (200*21) TPF 30 mins PPD 37K (SMP) GTS 450 @900/1800 TFP 1min 50 PPD 10K (GPU3) 47K PPD Running the GPU3 is having little impact on CPU Temps all good. Cheers guys
