How do I configure GPUs on machine by machine basis?
3 Apr 2013 4:12:43 UTC
Topic 196880
(moderation:
I have 3 machines running E@H, all with GPUs. One GPU has 3GB, 1 has 2GB and 1 has 1GB. Setting .33 for GPU utilization factor affects all three machines and when I use that, the 1GB GPU runs opencl-ati tasks very slowly. Setting to .5 is OK for 1GB GPU but 3 tasks run nicely on other 2 machines and I think I might be able to run 4 or more on the 3GB GPU (at least I'd like to try).
Any suggestions on how I could configure each box seperately?
RE: I have 3 machines
)
This is the easy part... lol ok so first go to Einstein@Home preferences... pick something besides default such as work... home or school. Add separate preferences for home... Add separate preferences for school... Separate preferences for work.. I already have work setup so that's why it does not say add in front of it... Pick 1 and change the setting GPU utilization factor of BRP apps DANGEROUS! Only touch this if you are absolutely sure of what you are doing! to the .5 save it. Now go to Computers on this account View "from your account page. Click on the Details link of the computer you want to change. Down at the bottom you will see Location... change that to whatever you pick from the Einstein@Home preferences "home, work, school" then save.
Now go to the computer that setting is for and do update. You are now done. Do note however it may take awhile for the settings to kick in. It has been said when you change this setting for how many units to do at 1 time will take time until the work cache catches up to the setting change. So if you change it and it takes a day to clear the work don't worry about it not changing to run only 2 task. It will take time so don't worry it will change.
PC setup MSI-970A-G46 AMD FX-8350 8 core OC'd 4.45GHz 16GB ram PC3-10700 Geforce GTX 650Ti Windows 7 x64 Einstein@Home
RE: RE: I have 3 machines
)
I agree with the above as long as you have 4 pc's or less, but if you have more then an app_config.xml file is easier. An example of one is:
milkyway
0.5
0.05
The gpu usage line current says 0.5 and will run 2 units at once, but to run 3 units at once you would just change it to 0.33. The 0.5 in the cpu usage line tells the machine to use no more then 1/2 of a cpu core to keep the gpu busy. In some cases this is NOT enough and a setting of 1.0 makes for faster crunching. Save the file in the hidden directory c:\Program Data\Boinc\Projects\Einstein as a text type file in NOTEPAD!
To use an app_config.xml file you MUST be using Boinc version 7.0.40 or higher, with your current version of 7.0.28 it will NOT work.
Thanks for all the info.
)
Thanks for all the info. Decided to go with the first method since I didn't want to upgrade to the beta. Here are a few results
Determined that the 3 GB GPU could handle 3 simultaneous Binary Pulsar Searches without destroying my normal screen operation. Before each task ran in about 50 minutes. With 3 simultaneous, I get 3 in ~82 minutes which is equivalent to about 1 every 27.33 minutes. I could have tried 4 but decided I might overload the card (GPU-Z was reporting over 80% GPU usage even though GPU memory was only a little more than half full).
On the 1GB card I could only run 1 at a time since it was my main computer and more than that caused flicker. In order to move from 3 to 1, I had to abort all GPU tasks since with 3 simultaneous was taking 6 hours. Then I downloaded new tasks and they now take about 55 minutes each. I tried two but there was no real gain and my screen response suffered even playing the occasional FreeCell game.
On the 2GB card I run 2 at a time. I got some benefit from this (2 in about 80 minutes rather than 1 in 60 minutes) which is ok for now.
One note: when you setup different profiles for default, home, work, and school, its useful to know which version is currently working on each machine. Spent a week thinking I was using one profile only to realize I was actually using another. I solved this by making the resource shares different for each profile. I started with 200 for default, I use 201 for work, 202 for home, and 203 for school. That way when you update the profile in BOINC manager, you know immediately what GPU resource number you're using because you can see immediately what resource share value is current in the projects tab. If the GPU doesn't adapt with 1-2 days, you might be well off to abort tasks to see if you can't get things working as configured.
I may fiddle a bit more to try to optimize but if I stay with the 7.0.28 Windows installation, I don't expect much more improvement before I can start using app_config.xml files for the next version.
Again, thanks for all the info.