I installed nvidia beta driver 266.35 on a system that I just installed another 580 on. The GPU usage seems to be a bit higher with this driver versus my other system with 262.99. The 262.99 system is at 39-45% and the 266.35 system is at 49-58%. Both systems are otherwise identically configured. This is running one work unit at a time via both systems.
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.
Thank you, that works better now: GPU usage 75%, Memory usage 29% (of 512MB), crunching time 5.71 +/- 0.19[h] from 7 files.
I would like know where and how to adjust the priority of the CUDA files permanently from "Normal - Low" to "High". If I do it in the Task-Manager by hand, it´s valid only for the just running task. If the file is done and the next became active, I have to adjust this afresh by hand. Is there such an app-file needed as shown before here?
The latest (1.05) version, released today or yesterday, seems to fix the thread priority problem, see the technical news section item. So with this new version of the CUDA BRP3 app, it should no longer be necessary to either reserve a whole CPU or set priorities manually. It should crunch along nicely with all available cores doing CPU BOINC tasks in parallel.
Primegrid has driver issues with Linux CUDA for me. Einstein is my priority but my GPU is sitting idle if no-one else writes Linux apps for it.
I figured out that the Nvidia driver 256.53 allows me to use the 32bit Linux PrimeGrid proth CUDA app. I am not sure if it is of any interest to the developers here, but the 260 series dirvers for Linux all error out almost instantly.
For comparison, using a simple GeForce GT220, the Collatz CUDA app still allows me to use the computer fairly normally (slight slowdown). The PrimeGrid app hits the GPU so hard that my system is effectively unusable for anything else. Both report 100% GPU usage, though. However, I am not sure I have a good GPU monitor program in Linux.
The higher the priority of a process is, the less it can be interrupted by other processes in the background. So the process with higher priority ends up more early. By default processes from BOINC have the priority “Less than Normalâ€. This is to make the computer reacting more spontaneous on demand and you have the feeling, it´s my computer that´s doing my job and the BOINC files are working in the background, if there is time for it. With the advent of multi core processors and the hyper threading technique this became less important, as there is always one processor shortly available to react on demand.
In cooperation of a cpu with a gpu the job of the cpu is to communicate with the operational system and feeding the gpu with fresh data and taking away processed data, so the gpu can crunch uninterrupted in a constant stream. The gpu is crunching while the cpu is doing all the administration work. This is important because of the 7 to 50 times higher crunching power of the gpu compared to the cpu.
For adjusting the process priority in Win 7, Win XP and I think also in Win Vista go with the mouse into the free field in the lower task bar and click on the right mouse button. In the upcoming field click on “Task Manager†and in the now upcoming new window on the tab “Processesâ€. The processes are sorted in alphabetical order. Click on “einsteinbinary_BRP3_1.05_.....†and then with the right mouse button on “Adjust Priorityâ€. The change of the priority in the upcoming field takes place only after confirmation.
If you start the freeware tool MSI Afterburner you can observe the effect of changing the priority from “Less than Normal†to “Highâ€. The graph becomes less cluttered and the mean gpu load increases. This is also after the last changes in the software, but the effect is remarkably less pronounced.
If you are farther interested in this topic look for this thread.
I hope this is sufficiently clear.
Kind regards
Martin
P.S.: The names of buttons, tabs etc. you find may be a little different from the names I showed, as I´m using the German version and have to translate this.
RE: RE: It`s look like
)
Try setting BOINC to use 75% CPU resource. This should leave one core open for the GPU.
I installed nvidia beta
)
I installed nvidia beta driver 266.35 on a system that I just installed another 580 on. The GPU usage seems to be a bit higher with this driver versus my other system with 262.99. The 262.99 system is at 39-45% and the 266.35 system is at 49-58%. Both systems are otherwise identically configured. This is running one work unit at a time via both systems.
RE: You may be able to
)
Seems to be right. Without PL, running 3 WU's, I now have a gpu-usage of 72% - 80% and an average finishing time of 1:35 hours.
RE: Set the CPU usage to
)
Thank you, that works better now: GPU usage 75%, Memory usage 29% (of 512MB), crunching time 5.71 +/- 0.19[h] from 7 files.
I would like know where and how to adjust the priority of the CUDA files permanently from "Normal - Low" to "High". If I do it in the Task-Manager by hand, it´s valid only for the just running task. If the file is done and the next became active, I have to adjust this afresh by hand. Is there such an app-file needed as shown before here?
Thanks for your assistance in forehand.
Martin
Hi! The latest (1.05)
)
Hi!
The latest (1.05) version, released today or yesterday, seems to fix the thread priority problem, see the technical news section item. So with this new version of the CUDA BRP3 app, it should no longer be necessary to either reserve a whole CPU or set priorities manually. It should crunch along nicely with all available cores doing CPU BOINC tasks in parallel.
CU
HB
RE: The latest (1.05)
)
That would have been Thursday 6 January, around midday. http://einstein.phys.uwm.edu/apps.php
RE: Primegrid has driver
)
I figured out that the Nvidia driver 256.53 allows me to use the 32bit Linux PrimeGrid proth CUDA app. I am not sure if it is of any interest to the developers here, but the 260 series dirvers for Linux all error out almost instantly.
For comparison, using a simple GeForce GT220, the Collatz CUDA app still allows me to use the computer fairly normally (slight slowdown). The PrimeGrid app hits the GPU so hard that my system is effectively unusable for anything else. Both report 100% GPU usage, though. However, I am not sure I have a good GPU monitor program in Linux.
RE: RE: I would like
)
Under Win you don't have to
)
Under Win you don't have to do anything. Version 1.05 of the Einstein BRP3 App works just as it should.
MrS
Scanning for our furry friends since Jan 2002
The higher the priority of a
)
The higher the priority of a process is, the less it can be interrupted by other processes in the background. So the process with higher priority ends up more early. By default processes from BOINC have the priority “Less than Normalâ€. This is to make the computer reacting more spontaneous on demand and you have the feeling, it´s my computer that´s doing my job and the BOINC files are working in the background, if there is time for it. With the advent of multi core processors and the hyper threading technique this became less important, as there is always one processor shortly available to react on demand.
In cooperation of a cpu with a gpu the job of the cpu is to communicate with the operational system and feeding the gpu with fresh data and taking away processed data, so the gpu can crunch uninterrupted in a constant stream. The gpu is crunching while the cpu is doing all the administration work. This is important because of the 7 to 50 times higher crunching power of the gpu compared to the cpu.
For adjusting the process priority in Win 7, Win XP and I think also in Win Vista go with the mouse into the free field in the lower task bar and click on the right mouse button. In the upcoming field click on “Task Manager†and in the now upcoming new window on the tab “Processesâ€. The processes are sorted in alphabetical order. Click on “einsteinbinary_BRP3_1.05_.....†and then with the right mouse button on “Adjust Priorityâ€. The change of the priority in the upcoming field takes place only after confirmation.
If you start the freeware tool MSI Afterburner you can observe the effect of changing the priority from “Less than Normal†to “Highâ€. The graph becomes less cluttered and the mean gpu load increases. This is also after the last changes in the software, but the effect is remarkably less pronounced.
If you are farther interested in this topic look for this thread.
I hope this is sufficiently clear.
Kind regards
Martin
P.S.: The names of buttons, tabs etc. you find may be a little different from the names I showed, as I´m using the German version and have to translate this.