PC with two GPUs

Message boards : GPUs : PC with two GPUs
Message board moderation

To post messages, you must log in.

AuthorMessage
magdaddy

Send message
Joined: 5 Jul 20
Posts: 1
Message 99637 - Posted: 5 Jul 2020, 22:12:27 UTC

I have a PC with two GPUs. Boinc has no problem using both of them with Einstein@home. I would like to use the GPUs for Einstein, but use my CPUs for Rosetta and World Community Grid. However, when I use Einstein with the other projects, only one GPU will be used. What setting is needed to ensure that the two GPUs are always running for Einstein?

thanks!
-Mag
ID: 99637 · Report as offensive
Nick Name

Send message
Joined: 14 Aug 19
Posts: 55
United States
Message 99669 - Posted: 7 Jul 2020, 4:23:37 UTC - in response to Message 99637.  

Rosetta work is probably taking priority due to shorter deadlines. Try reducing the number of CPU threads BOINC can use.

Options -> Computing Preferences -> Usage Limits (Use at Most % of the CPUs)

Lower this so that BOINC uses at least one less thread than it is now, if you don't get the desired result keep reducing it until you do. BOINC will over-commit the CPU to run GPU work. If you have eight threads available, it will run eight CPU tasks plus as many GPU tasks as it can. Einstein GPU work likes a full thread too, so by lowering the Use at Most % you are - counter intuitively - freeing up resources for GPU work.
Team USA forum
Follow us on Twitter
Help us #crunchforcures!
ID: 99669 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 99826 - Posted: 9 Jul 2020, 18:43:18 UTC - in response to Message 99669.  

It is my understanding that GPUs always get preference.
Sometimes it takes boinc a while before it realizes there's a GPU that's not being utilized.
First I would check your settings, to see if it's set to pause GPU when the PC is in use.
I also would make sure I have at least 0.2 days worth of work. I personally set mine to 1 day of work.

Then I would go on Einstein@home website, to set your project to only accept GPU.

Then I would update the project manually in Boinc manager. This should trigger new downloads.

Then it should automatically start them.

If Boinc stops Einstein CPU project, this is possible sometimes, even when you've set enough cores to crunch on.
It is possible, that your settings don't allow enough RAM to be utilized.
By default this setting is 50% and 90% I believe.
I up it to 85% and 90%.
ID: 99826 · Report as offensive
robsmith
Volunteer tester
Help desk expert

Send message
Joined: 25 May 09
Posts: 1283
United Kingdom
Message 99832 - Posted: 9 Jul 2020, 21:34:49 UTC

You can select which class of processor each project uses, and this is done on the project website.

WCG:
go to https://www.worldcommunitygrid.org/ms/device/viewBoincProfileConfiguration.do?name=Default
then scroll down and make sure you've only selected the CPU projects by unticking the "If my computer can process work on my graphics card, then please send me work to run on my graphics card for the projects that I have selected above." option
and possibly the "Please opt me in to new projects as they become available."
Then save the changes

Rosetta:
go to https://boinc.bakerlab.org/rosetta/prefs.php?subset=project
Click on the first "edit preferences" button and make sure that "Use CPU" is ticked

Save the changes

Einstein
Go to https://einsteinathome.org/account/prefs/project
Tick which ever GPU family you have, and untick the "use CPU" option
There's a whole load more options lower down the screen - most have some description against them, but two you may want to say "no" to are "Run CPU versions of applications for which GPU versions are available"
and "Allow non-preferred apps"

Save the changes.

In each case the changes will probably take a few hours to work their way through.

One thing to be aware of is that if you have two different GPUs from the same manufacturer (say a GTX980 and an RTX2080) in the same computer only the more powerful one will be used by default - overcoming this means editing a file, but if that's needed we can go through that another day...
Sometime located elsewhere in the (un)known Universe
But most often found somewhere near the middle of the UK
ID: 99832 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 718
United States
Message 99840 - Posted: 10 Jul 2020, 5:39:52 UTC - in response to Message 99832.  
Last modified: 10 Jul 2020, 5:42:25 UTC


One thing to be aware of is that if you have two different GPUs from the same manufacturer (say a GTX980 and an RTX2080) in the same computer only the more powerful one will be used by default - overcoming this means editing a file, but if that's needed we can go through that another day...

From my experience, sharing an RTX2060 with an RTX2080Ti does allow both GPUs to be used by default.
I think a slow GPU needs to be below a certain performance (eg: IGP, or very slow GPU) for Boinc not to use that one.
Another thing I've observed, is that if you share projects like einstein with 2 different speed GPUs, that you'll get 2 different types of WUs.
A small Wu for the slow GPU, and a big WU for the fast ones.
The 2080Ti is about twice as fast as a 2060.
If I mix the two GPUs in one system, I could be running the 2080 Ti, well below it's intended power target, as WUs still are randomly assigned by the Boinc Client.

For that reason, I prefer to use one system with GPUs of about equal performance,
like:
2080Ti + 2080 Super; or,
2070 Super + 2080 or 2080 Super; or,
2060 + 2060 Super or 2070.

If the jump is too high (2060 + 2080 Super), the WUs are randomly distributed to each GPU, sometimes resulting in less than optimal WUs for the GPU.
Einstein on average uses like 75-120W WUs, so for my 2080 Ti, I have to triple them. (gpu = 0.333 in settings). If I share my 2080Ti system with a 2060, even tripling them would result in a max of 120-150W (out of 300).
ID: 99840 · Report as offensive

Message boards : GPUs : PC with two GPUs

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.