Message boards :
Number crunching :
Question about dedicating CPU cores to GPU support w/Lunatics apps.
Message board moderation
Previous · 1 · 2 · 3
Author | Message |
---|---|
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
After creating the app_config.xml file however, the number/mix of tasks running hasn't changed; I'm still running 4xCPU "setiathome_v7 7.00" and 1xGPU "setiathome_v7 7.00 (Cuda32)". Against the single running GPU task it says "Running (0.04CPUs + 1NVIDIA GPU)". If it is finding it, you have the right name and location, but BOINC doesn't give any clues when there are errors. It is not picking up your v7 settings for cpu or gpu. Can only suggest you proofread carefully - post the file if you want another set of eyes. Did you use notepad to create it? XML editors cause problems like this. The cpu usage value of .1 will not reserve a core for S@Hv7, so you could use a value of .5 to do that if that is what you want. Another Fred Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop. |
Kieron Walsh Send message Joined: 2 Mar 00 Posts: 74 Credit: 43,502,325 RAC: 112 |
Hi Fred, My app_config.xml file content as placed in the C:\ProgramData\BOINC\projects\setiathome.berkeley.edu folder: <app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.1</cpu_usage> </gpu_versions> </app_config> A straight copy+paste from Claggy's advice. Grateful for any help. |
Fred E. Send message Joined: 22 Jul 99 Posts: 768 Credit: 24,140,697 RAC: 0 |
It is missing one line near the end. Should be: </gpu_versions> </app> </app_config> That may be enough to make the file invalid. Try it (with notepad). Another Fred Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
It's conventional, when preparing and comparing XML files, to indent the lines so the opening/closing tags can be matched. <app_config> <app> <name>astropulse_v6</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>setiathome_v7</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.1</cpu_usage> </gpu_versions> </app_config> That makes the missing </app> near the end stand out much more clearly |
Kieron Walsh Send message Joined: 2 Mar 00 Posts: 74 Credit: 43,502,325 RAC: 112 |
Thanks Fred, Working as planned now on the GPU side: 2xGPU. I also changed the CPU figure to 0.5 as you suggested and now have 3xCPU tasks running leaving one core spare for the GPU's to work with. All looking good. Thanks for your help guys. |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
A straight copy+paste from Claggy's advice. Sorry about that, I took it from a post on the forum, rather than from my setiathome directory. Claggy |
CElliott Send message Joined: 19 Jul 99 Posts: 178 Credit: 79,285,961 RAC: 0 |
I have two GTX 670s and two 760s on one mobo (Computer ID 7179664). All GPUs have 2048 MB of memory. As the load on the 4 GPUs was about 80% with 2 WUs per GPU, I changed <count>0.5</count> to <count>0.33333333</count> for 3 WUs per GPU. I am only seeing 0.6 WU per hour more with 3 WUs per GPU than my other computer, which has lower quality GPUs and only 2 WUs per GPU. Part of the problem is that with 3 WUs per GPU, the system has trouble starting a new WU on a GPU when a WU finishes. Watching Task Manager, what happens when a new WUs is started is that the RAM assigned to the process stays at 2447K for many seconds, then increases to 19,xxxK for many more seconds, then finally goes to 70-94,xxxK at which time the process is fully runnning. Watching Efmer's BoincTasks at the same time, the new WU task is taking 1.5 to 6 minutes to start. All during this time it is using no measureable CPU time, but is being assigned GPU time, and progress increases. When the WU finally "catches," i.e., memory use is at its final value (70-94,xxxK), progress goes back to zero, elapsed time is 1 - 6 minutes, and CPU time is a few seconds. With 3 WUs per GPU, GPU load is a solid 99%. Where do you think the problem lies in starting a new WU when running 3 WUs per GPU? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.