Is there a possibility to run 2 WU's on 1 GPU ?
I have a usage of 44% (max) so the GPU has the room for another WU.
Perhaps a setting in BOINC ?
I tested it yesterday and got a usage of 80% on my GTX260 on winxp32. On vist64 in only get it to 63%. I ran only the two CUDA-tasks and no CPU-task. You have to change the value of in the client_state.xml in your boinc data folder (I changed it from 1.000 to 0.49). But the value will reset after a download of new CUDA-tasks. So no real solution. I don´t know if there is an app_info.xml file. More information here http://einsteinathome.org/node/195281.
"Funny" is maybe not the right thing to have said. Yet each time these wishes for preferences for the GPU come up, the user expects that they're 'easy' to add, without having any knowledge of the complexity behind programming one line for BOINC so it runs and compiles the same on Windows, OS X and any Linux distro you throw it at.
Bernd is right. The statement was written from the users perspective. And there was no word, that this meight be easy. And there was also no word, that it might be quick. It was a suggestion for a longterm development strategy, to provide more easy to use service to the users. In best case everything is automated and in the same time use dynamically as much ressources as are available.
This is really hard stuff. But the management of user requirements and user acceptance are daily bread in the development of Enterprise software, e.g. for ERP. If you don't win the users, you'll not have the project payback. That's the idea behind the suggestion
I´m running a small Laptop consisting of CPU: Intel I3M350 (2,27GHz) and GPU: Nvidia GeForce 310M (474MB) running under Win7HP.
After installation of the new graphics driver 260.99 I can crunch also BRP3-cuda files. (Thank you for the hint in the BOINC messages.) While crunching this, the mean cpu load for S5GC1HF files is 22.5%, for the BRP3-cuda file 1,7% as a fifth file, taken from the resource monitor. Before installation of the new graphics driver the S5GC1HF files where running about 10,5[h], where as they are running now about 13[h], astonishing 25% longer. Probably a cooling problem of the cpu? The BRP3-cuda files are running about 13[h] also, the non cuda files about 3 times longer. From 25 measurements taken with the tool GPU-Z there is a gpu load of 28,7+/-12%, with maximums of 43% and minimums of 3% and the memory usage of 11,3+/-4,3%, with maximums of 17% and minimums of 1%. I find this combination of relative low gpu and cpu load and long runningtime amazing. What´s the bottle neck? As the gpu core temperature is 84°C, I believe it´s not advisable to look for a second cuda file at the same time. Up to now all BRP3 files became validated, which is not naturally, as the relative amount of invalid files are increasing continuously since a week and reached 4.4% today.
By the way: HAPPY NEW YEAR to all out there!!!
Martin
Set the CPU usage to n-1 CPU for the non CUDA tasks. CUDA will then use the remaining CPU and your GPU will have a permanent load of ~80% and run in ~ 3.5 hours.
You have to change the value of in the client_state.xml in your boinc data folder (I changed it from 1.000 to 0.49). But the value will reset after a download of new CUDA-tasks. So no real solution. I don´t know if there is an app_info.xml file.
You have to change the value of in the client_state.xml in your boinc data folder (I changed it from 1.000 to 0.49). But the value will reset after a download of new CUDA-tasks. So no real solution. I don´t know if there is an app_info.xml file. More information here http://einsteinathome.org/node/195281.
Done that (.2cpu/.5gpu), but before 1 WU took about 53 minutes to finish and now, running two WU's at the same time, they take 103 and 106 minutes so no gain.
My systems are AM2+ mobo/2G DDR2/PhenomII X4 945/GTX460SC and FTW.
CPU does Malaria
You have to change the value of in the client_state.xml in your boinc data folder (I changed it from 1.000 to 0.49). But the value will reset after a download of new CUDA-tasks. So no real solution. I don´t know if there is an app_info.xml file. More information here http://einsteinathome.org/node/195281.
Done that (.2cpu/.5gpu), but before 1 WU took about 53 minutes to finish and now, running two WU's at the same time, they take 103 and 106 minutes so no gain.
My systems are AM2+ mobo/2G DDR2/PhenomII X4 945/GTX460SC and FTW.
CPU does Malaria
With the app.info.xml from here http://einsteinathome.org/node/195553 (thank you leks!!!) I run now 3 CUDA-tasks on my GTX260. The GPU load goes now up to 85%, but only if there are no CPU-tasks running! In the first 10 minutes each task has reached 6.xx %. So adding together to 18,5% in 10 min. they will run longer but if I run only one task it only gets me to 13% in 10 min.
But I have other problems. I also run vista64 on the same computer and only get 70% GPU load. But the work still gets done quicker than with only one task. 1.1%/min to 1.4%/min.
I also used Process Lasso to assign a core to a WU.
This morning it took 105 minutes to finish a WU, with or without cpu running, and went back to running single WU.
Is there a possibility to run
)
Is there a possibility to run 2 WU's on 1 GPU ?
I have a usage of 44% (max) so the GPU has the room for another WU.
Perhaps a setting in BOINC ?
RE: Is there a possibility
)
Yes. possible.
no boinc. application settings.
RE: Is there a possibility
)
I tested it yesterday and got a usage of 80% on my GTX260 on winxp32. On vist64 in only get it to 63%. I ran only the two CUDA-tasks and no CPU-task. You have to change the value of in the client_state.xml in your boinc data folder (I changed it from 1.000 to 0.49). But the value will reset after a download of new CUDA-tasks. So no real solution. I don´t know if there is an app_info.xml file. More information here
http://einsteinathome.org/node/195281.
RE: "Funny" is maybe not
)
Bernd is right. The statement was written from the users perspective. And there was no word, that this meight be easy. And there was also no word, that it might be quick. It was a suggestion for a longterm development strategy, to provide more easy to use service to the users. In best case everything is automated and in the same time use dynamically as much ressources as are available.
This is really hard stuff. But the management of user requirements and user acceptance are daily bread in the development of Enterprise software, e.g. for ERP. If you don't win the users, you'll not have the project payback. That's the idea behind the suggestion
I´m running a small Laptop
)
I´m running a small Laptop consisting of CPU: Intel I3M350 (2,27GHz) and GPU: Nvidia GeForce 310M (474MB) running under Win7HP.
After installation of the new graphics driver 260.99 I can crunch also BRP3-cuda files. (Thank you for the hint in the BOINC messages.) While crunching this, the mean cpu load for S5GC1HF files is 22.5%, for the BRP3-cuda file 1,7% as a fifth file, taken from the resource monitor. Before installation of the new graphics driver the S5GC1HF files where running about 10,5[h], where as they are running now about 13[h], astonishing 25% longer. Probably a cooling problem of the cpu? The BRP3-cuda files are running about 13[h] also, the non cuda files about 3 times longer. From 25 measurements taken with the tool GPU-Z there is a gpu load of 28,7+/-12%, with maximums of 43% and minimums of 3% and the memory usage of 11,3+/-4,3%, with maximums of 17% and minimums of 1%. I find this combination of relative low gpu and cpu load and long runningtime amazing. What´s the bottle neck? As the gpu core temperature is 84°C, I believe it´s not advisable to look for a second cuda file at the same time. Up to now all BRP3 files became validated, which is not naturally, as the relative amount of invalid files are increasing continuously since a week and reached 4.4% today.
By the way: HAPPY NEW YEAR to all out there!!!
Martin
Set the CPU usage to n-1 CPU
)
Set the CPU usage to n-1 CPU for the non CUDA tasks. CUDA will then use the remaining CPU and your GPU will have a permanent load of ~80% and run in ~ 3.5 hours.
RE: You have to change the
)
This is done in app_info.xml
http://einsteinathome.org/node/195553
RE: You have to change the
)
Done that (.2cpu/.5gpu), but before 1 WU took about 53 minutes to finish and now, running two WU's at the same time, they take 103 and 106 minutes so no gain.
My systems are AM2+ mobo/2G DDR2/PhenomII X4 945/GTX460SC and FTW.
CPU does Malaria
RE: RE: You have to
)
With the app.info.xml from here http://einsteinathome.org/node/195553 (thank you leks!!!) I run now 3 CUDA-tasks on my GTX260. The GPU load goes now up to 85%, but only if there are no CPU-tasks running! In the first 10 minutes each task has reached 6.xx %. So adding together to 18,5% in 10 min. they will run longer but if I run only one task it only gets me to 13% in 10 min.
But I have other problems. I also run vista64 on the same computer and only get 70% GPU load. But the work still gets done quicker than with only one task. 1.1%/min to 1.4%/min.
I also used Process Lasso to
)
I also used Process Lasso to assign a core to a WU.
This morning it took 105 minutes to finish a WU, with or without cpu running, and went back to running single WU.