Hi Gang,

In my machine I have an nVidia GTX 285 and a GT 240.

I currently have BOINC set to use all GPUS and to report immediately.

Currently the 285 is running a "Long Run" and the 240 is running "ACEMD2:GPU Molecular dynamics..."

My agent is telling me that the 285 is only using 24% of its CPU and my 240 is only using 5%!

I have followed all of the instructions on this site for setting up my system to speed stuff up, but still my GPU utilization seems very low.

I have tested the clock speeds on the cards and they are both high/normal.

And I don't see this when they are crunching on Einstein@, Milkyway@, Docking@..etc.. Just GPUGrid..

Thoughts?

My agent is telling me that the 285 is only using 24% of its CPU and my 240 is only using 5%!

I have followed all of the instructions on this site for setting up my system to speed stuff up, but still my GPU utilization seems very low.

Thoughts?

Think you are mixing up your CPU and GPU.

Think you are mixing up your CPU and GPU.

Hrmm.. Well you might be right.. I know the difference between cpu and gpu, but I was assuming the percentage used as indicated by the agent I have was referring to the GPU usage for the WU's on the GPUs.. but it may actually be showing CPU usage, and not have any indication of how much GPU is being used..

Do GPUGrid WU's typically take a lot longer than other grid computer tasks? The log running one said it takes 8-12 hours on the fastest machines so that one I expected to be slow, but the other one seemed even slower and I typically can produce a Milkyway, Einstein or Docking WU in 2 hours or less.. Some as fast as a few minutes..

You can use GPUZ to see the % GPU utilization.

GPUGrid tasks are longer than some and shorter than others. Some Lattice and RNA tasks can run for several days and even weeks. Climate tasks typically take weeks to months to complete. At the other extreme some MW, VP and ibercivis tasks take just a few minutes. On high end GPU's tasks take between about 4h and 12h, so I would say they are slightly longer on average.

Thanks for your fast responses Skgiven, I appreciate it..

GPUZ seems to show my GPU's running at about 50% each.. So still not 100%.. Though this is on two Einstein@ tasks..

Also on a side note.. I was running two GPUGrid tasks and paused boinc to restart my machine.. On restarting boinc I had no GPUGrid tasks and my log reports this;

423 GPUGRID 3/20/2011 3:42:18 PM update requested by user
424 GPUGRID 3/20/2011 3:42:22 PM Sending scheduler request: Requested by user.
425 GPUGRID 3/20/2011 3:42:22 PM Not reporting or requesting tasks
426 GPUGRID 3/20/2011 3:42:24 PM Scheduler request completed

Why not requesting tasks?

At the other extreme some MW, VP and ibercivis tasks take just a few minutes.

You aint kidding.. I just had a MW take like 8 minutes..

Perhaps you did not get any GPUGrid tasks because you were running other GPU tasks (MW) or other project debt (due to project weightings) was in affect.

Perhaps you did not get any GPUGrid tasks because you were running other GPU tasks (MW) or other project debt (due to project weightings) was in affect.

Ah okay.. I'll have to look that up.. I dont know what project debt is yet..

Thanks bud!

I was running 50 to 60% on my 2 GTX 260's in SLI. I backed my CPU usage down to max of 70% and now run about 82% on each of my GPU's

You might want to check out gpu shark. its good at displaying relevant gpu information of all installed gpu's.

---