The science apps don't know how to run a gpu task in SLI configuration. Normally it is recommended to remove the SLI connector when doing distributed computing as it typically slows down computation.
Nvidia cards do not do well with multiple tasks running concurrently on the same card. Best to run only a single task.
AMD cards on the other hand run exceedingly well with 2,3 or 4 tasks running concurrently.
Difference is in the hardware and mainly in the drivers for the different platforms.
That is why AMD cards are always recommended for both Einstein and Milkyway and also for their better FP64 ratings.
Two projects per GPU for RTX2060 and up is acceptable.
2070Super and higher can run 3 (but if this is your main display GPU, the image will become laggy).
However, crunching 1 task at 5 minutes, 2 at 7, or 3 at 9 minutes, Nvidia GPUs seem to run multiple projects pretty good in one GPU.
The only thing I've noticed is, with Moo or Milkyway (I always confuse them), one or the other, the GPU only runs at ~105W. Even with multiple projects (1 project runs at 80W).
HI Keith, Just used the Web
)
HI Keith, Just used the Web site config...
Looks like its a hit and miss affair now...
Seen some the SLI rigs running 2 work units per GPU then later on back to 1 per GPU
Least they are all running though so that's a good thing and no BSOD's so far.
The science apps don't know
)
The science apps don't know how to run a gpu task in SLI configuration. Normally it is recommended to remove the SLI connector when doing distributed computing as it typically slows down computation.
Nvidia cards do not do well with multiple tasks running concurrently on the same card. Best to run only a single task.
AMD cards on the other hand run exceedingly well with 2,3 or 4 tasks running concurrently.
Difference is in the hardware and mainly in the drivers for the different platforms.
That is why AMD cards are always recommended for both Einstein and Milkyway and also for their better FP64 ratings.
Wow really, one up for the
)
Wow really, one up for the ole SETI then lol...
Thanks for the heads up. Not about to take off the Bridge just to fold as I use my PCs so I'll revert back to 1 task per GPU..
Thanks for the tip.
Two projects per GPU for
)
Two projects per GPU for RTX2060 and up is acceptable.
2070Super and higher can run 3 (but if this is your main display GPU, the image will become laggy).
However, crunching 1 task at 5 minutes, 2 at 7, or 3 at 9 minutes, Nvidia GPUs seem to run multiple projects pretty good in one GPU.
The only thing I've noticed is, with Moo or Milkyway (I always confuse them), one or the other, the GPU only runs at ~105W. Even with multiple projects (1 project runs at 80W).