Running CPU and GPU Clients Question

[Smartazz]

Hey all, I'm currently crunching WCG on my 4 CPU cores and SETI@home and Collatz Conjecture on the two video cards. Should I leave a core free for the video cards or would it not make much of a difference to them? Thanks in advance.

 

[biodoc]

It might be worth testing since you have 2 cards on the same machine. If the WUs for the GPU projects show similar run times for completion you could suspend WCG and then check if the WU run times for the GPU projects are reduced. You could also open up the task manager and monitor cpu usage & total cpu run time for the WCG & GPU projects before and after suspending WCG.

 

[Smartazz]

Is watching the gpu usage percentages a reliable method of testing this?

 

[salvorhardin]

The best way is to check if an extra core decreases your gpu runtimes. In my experience watching gpu usage isn't reliable. I recently had to stop crunching moowrapper on my 4830 because it would interfere with poem which was running on my 5850. When running both projects both gpus were pegged at 98% usage, but when I stopped moowrapper usage on my 5850 dropped to 55% and my runtimes decreased by almost half.

 

[zzuupp]

The best way is to check if an extra core decreases your gpu runtimes. In my experience watching gpu usage isn't reliable. I recently had to stop crunching moowrapper on my 4830 because it would interfere with poem which was running on my 5850. When running both projects both gpus were pegged at 98% usage, but when I stopped moowrapper usage on my 5850 dropped to 55% and my runtimes decreased by almost half.

IIRC, doesn't moowrapper try to spread itself across all GPUs.

 

[salvorhardin]

I have 2 boinc directories setup so I can crunch 2 different gpu projects at once. Each one is setup to ignore a gpu so they don't interfere with each other.

 

[Sunny129]

Hey all, I'm currently crunching WCG on my 4 CPU cores and SETI@home and Collatz Conjecture on the two video cards. Should I leave a core free for the video cards or would it not make much of a difference to them? Thanks in advance.

IIRC, SETI@Home Multibeam ATI GPU tasks require ~4% of a CPU core per task. S@H Astropulse i'm a little less sure about, even though that's primarily what i crunch when i'm running S@H (its been a while since i've done SETI)...but IIRC, Astropulse ATI GPU tasks consume ~5% of a CPU core each. Collatz ATI GPU tasks consume only ~1% of a CPU core each. i understand that when a task's status in BOINC reads something like "running (0.05 CPUs + 0.5 ATI GPUs)," the "0.05 CPUs" is just an estimate...but if you monitor those tasks in the task manager (as biodoc mentioned) while they run, you'll see that although CPU resource consumption oscillates above and below the estimate quite a bit for each task, the estimates are pretty accurate. again, these figures are based on ATI GPUs, so if you're crunching any of the projects you mentioned with nVidia GPUs, i have no idea if they consume any more or any less of the CPU while crunching the same tasks.

the point of all that is that IF you're crunching on ATI GPUs, the CPU component of GPU tasks for either the SETI@Home or Collatz projects may or may not be more or less negligible, depending on your hardware setup. to give you an example, i crunch Einstein@Home, LHC@Home, and Test4Theory@Home on all 6 cores of my 1090T CPU while running 2 Milkyway@Home tasks simultaneously on my HD 5870 GPU. like the projects you mentioned above, MW@H ATI GPU tasks only consume 0.05 of a CPU core. so does running 6.1 CPU cores worth of work on only 6.0 CPU cores slow things down? yes - both the CPU and GPU tasks suffer...but again, its negligible...at least on my ATI GPU crunching machine it is.

now consider my other box, which again crunches Einstein@Home, LHC@Home, and Test4Theory@Home on only 5 of of the 6 CPU cores, and 3 Einstein@Home BRP4 tasks simultaneously on my GTX 560Ti. these GPU tasks consume 0.20 of a CPU core each, for a total of 0.60 of a CPU core. the estimate of 0.20 CPU cores per GPU task is again pretty accurate as i've confirmed by monitoring these tasks in the task manager. so unlike my ATI box, where it isn't necessary to leave a CPU core free for the 0.1 CPUs worth of GPU tasks, it IS worth it to leave a CPU core free for the 0.60 CPUs worth of GPU tasks running at any given time on my nVidia box. obviously everything runs fast and smooth on the nVidia box b/c i don't fully utilize the CPU (i only utilize on average ~5.6 of 6 cores). if i force a 6th CPU task, both CPU and GPU tasks suffer noticeably. running 6.6 CPU cores worth of work on only 6.0 CPU cores takes a much bigger toll than trying to run 6.1 CPU cores worth of work on only 6.0 CPU cores.

sorry for the long-winded response, but hopefully it gives you some insight. basically, its all going to come down to experimentation b/c everyone's mixed and matched hardware setup is different.

 

[Smartazz]

Thanks for the advice. The completion times barely change so I figure I'll leave all four cores dedicated to WCG.