Forum > GPU crunching
V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use
Archangel999:
sorry for that i said but most of the people don't know how to install the applications that is why they don't work
PS: Thanks Raistmer for the mod :)
Raistmer:
--- Quote from: Jason G on 28 Feb 2009, 06:29:49 am ---So.... not advised to run with current Astropulse only work climate then, lest you get NumGPUs over subscription... right?
--- End quote ---
Anything else than SETI MB could result in overloaded CPU and idle GPU if leaved w/o user intervention. Lack of BOINC ability for project pairing make it impossible to be sure at least one task allocated for SETI MB and CPU->GPU rescheduling is possible only inside teamed SETI MB programs.
Jason G:
--- Quote from: Raistmer on 28 Feb 2009, 01:52:31 pm ---
--- Quote from: Jason G on 28 Feb 2009, 06:29:49 am ---So.... not advised to run with current Astropulse only work climate then, lest you get NumGPUs over subscription... right?
--- End quote ---
Anything else than SETI MB could result in overloaded CPU and idle GPU if leaved w/o user intervention. Lack of BOINC ability for project pairing make it impossible to be sure at least one task allocated for SETI MB and CPU->GPU rescheduling is possible only inside teamed SETI MB programs.
--- End quote ---
Cheers for confirming that for me.
Raistmer:
--- Quote from: SoNic on 28 Feb 2009, 08:52:24 am ---Thank you for the new release, it is working great on my system.
I have one complaint thou. Throwing out the VLAR WU... It is a shame, I was just looking to the Boinc client and I saw 12 of them beeing rejected one after another. I don't think it is fair to the other wingman.
Is there a way that this can be solved without autokill? Like force them to the CPU instead of GPU? What happens if there are not killed - under V10 app?
--- End quote ---
It much more fair to send VLAR for CPU processing than leave it to CUDA processing cause CPU will do it faster while CUDA will spend exessive time. All should do its own work. Topic was discussed already.
CPU rescheduling maybe possible but seems complicated in current conditions (especially because VLARs come in packs - how will you keep GPU busy and CPU not overloaded too much in this situation?
IF someone wanna fiddle with it I can provide sources of my mods for start.
Raistmer:
About "child processes" issue:
Most probably it means that modded AK_v8b was unable to start CUDA_MB app. So it can't wait finish of non-started process hence it exited with that message.
Why it can't run CUDA app - there are many possibilities - one of them - access rights issue.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version