Its' been a long time!
5 Feb 2019 14:01:52 UTC
Topic 218110
(moderation:
It's been a long time since I've participated in Einstein and I want to use it as a backup, running GPU tasks for when SETI is down. I'm currently running on an i7/7700K w/ twin Nvidia GTX1070 FE GPUs, and unfortunately Win$doze 10 pro. I have an app_config set for 1 CPU / .5 GPU. Is there any fine tuning that I should employ to get the most productive work out of the GPUs?
I don't buy computers, I build them!
Cliff - I've moved your post
)
Cliff - I've moved your post to the Cruncher's Corner where other users can help you with your question of GPU tuning.
Einstein@Home Project
Cliff Harding wrote:I... I
)
Your computer last made contact about a week ago and it currently shows as having no tasks. Have you previously crunched the FGRPB1G GPU tasks and are you sure your app_config.xml file is correctly set up for the current GPU work?
With the current work, nVidia GPUs will not perform optimally if you don't allow a full CPU core to support each GPU task instance. I don't run nVidia GPUs so I'm just repeating what those who do have observed.
Also, these days, there doesn't seem to be very much benefit from running concurrent GPU tasks, particularly if you want to use CPU cores for crunching at other projects. If you were crunching 4 GPU tasks, you would need 4 CPU cores for support. If you were not wanting to crunch any CPU tasks at all, there would be no problem crunching 2 GPU tasks on each GPU and you wouldn't need an app_config.xml file either. You could just use the GPU utilization factor in your project preferences.
It would still be worthwhile starting with just a single task on each GPU to establish a performance baseline and then change to 2 per GPU to see how much improvement you achieved. There should be some but it probably won't be all that great. A lot has changed since the days when running multiple tasks gave big improvements.
Cheers,
Gary.