Advanced search

Message boards : Wish list : Help other protein groups implement GPU processing

Author Message
Scott A. Howard
Send message
Joined: 29 Dec 08
Posts: 9
Credit: 9,407
RAC: 0

Scientific publications
Message 16993 - Posted: 13 May 2010 | 12:18:23 UTC

Based on the number of threads concerning the lack of ability to process GPUGrid WUs in a timely fashion on their machines, it would benefit the world community if GPUGrip would help the other protein processing groups implement GPU processing.

The only other such group is The Lattice Project but they have not had WUs recently. There are many other protein projects that routinely have WUs but for the CPU only.

GPUGrid helping the other groups would allow us to spread our GPU time around on projects that are more accommodating and community friendly.

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 482
Credit: 554,467,553
RAC: 7,235
Scientific publications
Message 17311 - Posted: 25 May 2010 | 21:26:50 UTC - in response to Message 16993.

Do you feel like donating enough to GPUGRID that they can hire another person in order to free enough time to allow this?

The Lattice Project currently has some rather large and rather long workunits going - the one that's already run 291 hours on my machine says that the 1.2 GB I'm giving it is barely enough that it can run at all. If it takes that much memory to run on just one processor, I'd expect it to take a major rewrite to be able to run much faster with more processors but not much more memory.

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Scientific publications
Message 17366 - Posted: 27 May 2010 | 15:43:18 UTC - in response to Message 17311.

Yeah, people need to watch out for those long Lattice tasks!

I had several trying to run on my i7, and they took up over 5GB RAM. The system only has 2GB; fine for most projects but not for the Lattice Project - Net result, the CPU was sitting at about 30% and the HDD was churning away.
Also, because it uses HT the tasks take longer than on real cores.
I have to manually limit how many run, and stop picking up new lattice tasks to keep that system going. It will take a week to finish them.
I also did not like the fact that after 17h it was still at 1% and a restart sent it back to 0%. I guessed that my system had started some sort of runaway RAM eating thread, so I restarted.

Anyway, they are CPU tasks, though Lattice does occasionally have GPU tasks.

Post to thread

Message boards : Wish list : Help other protein groups implement GPU processing