CUDA_Opencl_100

Message boards : Number crunching : CUDA_Opencl_100
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Cliff Harding
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1432
Credit: 110,967,840
RAC: 67
United States
Message 1329478 - Posted: 20 Jan 2013, 17:29:17 UTC

I started successfully running the nVidia version of cuda_opencl_100 the other day today I noticed something that I don't like and need help understanding what is happening to decide if there is a bug or if it is just me.

I am running BOINC 7.0.44, _x41zc cuda 5.0, under Win7(64-bit) with 1x EVGA GTX660SC/2Gb & 1x EVGA GTX460SE v1/1Gb. Normally I run at 80% CPU (6 cores/6 CPU tasks) and 3 tasks each on the graphic devices. When running Einstein or Milkyway opencl the number of CPU tasks running do not decrease, running on 6 cores. However when running SETI opencl, the number of tasks reduced by 1 core/1 task. At first I thought that it was because of
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>


I removed the suspected sub-parms and after restarting BOINC and it is still running on 5 cores/5 tasks with 2 opencl tasks. How do I increase the number of CPU tasks back to 6 and still have 2 opencl tasks running without increasing the percentage of processors, or is this a bug that needs to be reported?


I don't buy computers, I build them!!
ID: 1329478 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1329482 - Posted: 20 Jan 2013, 17:41:06 UTC - in response to Message 1329478.  
Last modified: 20 Jan 2013, 17:45:40 UTC

I started successfully running the nVidia version of cuda_opencl_100 the other day today I noticed something that I don't like and need help understanding what is happening to decide if there is a bug or if it is just me.

I am running BOINC 7.0.44, _x41zc cuda 5.0, under Win7(64-bit) with 1x EVGA GTX660SC/2Gb & 1x EVGA GTX460SE v1/1Gb. Normally I run at 80% CPU (6 cores/6 CPU tasks) and 3 tasks each on the graphic devices. When running Einstein or Milkyway opencl the number of CPU tasks running do not decrease, running on 6 cores. However when running SETI opencl, the number of tasks reduced by 1 core/1 task. At first I thought that it was because of
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>


I removed the suspected sub-parms and after restarting BOINC and it is still running on 5 cores/5 tasks with 2 opencl tasks. How do I increase the number of CPU tasks back to 6 and still have 2 opencl tasks running without increasing the percentage of processors, or is this a bug that needs to be reported?

I set it like that in the app_info so the OpenCL AP app isn't slowed because of lack of CPU time, both the Nvidia and ATI OpenCL AP apps perform better with a free core,
change it to the following if you want:

<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>

Claggy
ID: 1329482 · Report as offensive
Profile Cliff Harding
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1432
Credit: 110,967,840
RAC: 67
United States
Message 1329495 - Posted: 20 Jan 2013, 18:02:24 UTC - in response to Message 1329482.  

I started successfully running the nVidia version of cuda_opencl_100 the other day today I noticed something that I don't like and need help understanding what is happening to decide if there is a bug or if it is just me.

I am running BOINC 7.0.44, _x41zc cuda 5.0, under Win7(64-bit) with 1x EVGA GTX660SC/2Gb & 1x EVGA GTX460SE v1/1Gb. Normally I run at 80% CPU (6 cores/6 CPU tasks) and 3 tasks each on the graphic devices. When running Einstein or Milkyway opencl the number of CPU tasks running do not decrease, running on 6 cores. However when running SETI opencl, the number of tasks reduced by 1 core/1 task. At first I thought that it was because of
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>


I removed the suspected sub-parms and after restarting BOINC and it is still running on 5 cores/5 tasks with 2 opencl tasks. How do I increase the number of CPU tasks back to 6 and still have 2 opencl tasks running without increasing the percentage of processors, or is this a bug that needs to be reported?

I set it like that in the app_info so the OpenCL AP app isn't slowed because of lack of CPU time, both the Nvidia and ATI OpenCL AP apps perform better with a free core,
change it to the following if you want:

<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>

Claggy


I will try your suggestion. But the problem that I have understanding this is that there are already 2 cores supposedly available for cuda use. These 2 cores are not used by CPU task processing for this express purpose. What is the default setting for avg_ncpus & max_ncpus, since removing two sub-parms from the app_info.xml does not increase the number of active CPU tasks from 5 to 6? If the default is as I suspect 1.0 & 1.0, reducing another core for CPU processing, would it behove me to quit HT processing altogether so that I get back to 6 cores processing tasks?


I don't buy computers, I build them!!
ID: 1329495 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1329534 - Posted: 20 Jan 2013, 19:20:06 UTC - in response to Message 1329495.  
Last modified: 20 Jan 2013, 19:22:46 UTC

I started successfully running the nVidia version of cuda_opencl_100 the other day today I noticed something that I don't like and need help understanding what is happening to decide if there is a bug or if it is just me.

I am running BOINC 7.0.44, _x41zc cuda 5.0, under Win7(64-bit) with 1x EVGA GTX660SC/2Gb & 1x EVGA GTX460SE v1/1Gb. Normally I run at 80% CPU (6 cores/6 CPU tasks) and 3 tasks each on the graphic devices. When running Einstein or Milkyway opencl the number of CPU tasks running do not decrease, running on 6 cores. However when running SETI opencl, the number of tasks reduced by 1 core/1 task. At first I thought that it was because of
<avg_ncpus>1.0</avg_ncpus>
<max_ncpus>1.0</max_ncpus>


I removed the suspected sub-parms and after restarting BOINC and it is still running on 5 cores/5 tasks with 2 opencl tasks. How do I increase the number of CPU tasks back to 6 and still have 2 opencl tasks running without increasing the percentage of processors, or is this a bug that needs to be reported?

I set it like that in the app_info so the OpenCL AP app isn't slowed because of lack of CPU time, both the Nvidia and ATI OpenCL AP apps perform better with a free core,
change it to the following if you want:

<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>

Claggy


I will try your suggestion. But the problem that I have understanding this is that there are already 2 cores supposedly available for cuda use. These 2 cores are not used by CPU task processing for this express purpose. What is the default setting for avg_ncpus & max_ncpus, since removing two sub-parms from the app_info.xml does not increase the number of active CPU tasks from 5 to 6? If the default is as I suspect 1.0 & 1.0, reducing another core for CPU processing, would it behove me to quit HT processing altogether so that I get back to 6 cores processing tasks?

There are two ways of reserving CPU cores, one is the Blunt method of changing the percentage of 'On multiprocessors, use at most' to 75%,

the other method is to reserve them in your app_info (while leaving the CPU % figure set to 100%), say you're want to run two Cuda tasks on each GPU, and you want to reserve two cores for that,
as well as setting the count value to 0.5, you'd also set the <avg_ncpus> and <max_ncpus> to both to 0.5, Boinc will use those 0.5 figures and when it starts the four GPU tasks (two on each GPU),
it'll stop doing any extra CPU tasks, so giving the GPU it's two free cores, (4 times 0.5 is 2), the advantage of this method is that once the Wu's that need cores reserving end, the cores are available for CPU apps to use.

Claggy
ID: 1329534 · Report as offensive

Message boards : Number crunching : CUDA_Opencl_100


 
©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.