PDA

View Full Version : 2 threads, only want 1



MindCrime
04-24-14, 01:20 AM
Backstory: In the past I made an app_config to run 2 WUs at once on this machine, I've setup another machine that now runs on the SETI.USA Hydra account. Since it was a single account I decided I'll just uninstall BOINC and reinstall and add projects fresh. It's got a GeForce 8800GTS (single precision)with a dualcore opteron 170; not very quick and minimal threads. It can do almost 40k in a day if it has a lot of primegrid, or collatz and the like.

Well I like to spread the love around and noticed einstein is running 2 threads of: Binary Radio Pulsar Search (Arecibo, GPU) 1.39 (BRP4g-cuda32-nv301) The Hydra (opteron/8800gts) machine never had an appconfig file for einstein and when I go to einstein preferences for the SETI.USA Hydra it has all the GPU utilization factors set to 1. It's neat and all I like the project but the extra thread really eats up what little CPU thread(s) i have for any single CPU project. There is no xml files in the einstein folder ProgramData>Boinc>projects>einstein.phys.uwm.edu

Is there a way to manage project settings like this in case they're changed by someone else?

Bryan
04-24-14, 01:33 AM
Put this in your Einstein folder and name it app_config.xml

<app_config>
<app>
<name>einsteinbinary_BRP4G</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>

Then in the BOINC ADVANCED tab hit "read config files". That should drop you to a single WU that reserves a single CPU thread.

Shandia
04-24-14, 02:17 AM
I am wondering, what is the difference between cpu = 1.00 threads and cpu = 0.98 threads. From where I sit, there is no performance improvement between 0.98 and 1.00. That extra 0.02 does get used in cpu tasks. So when I configure it to cpu = 1 I end up shorting my cpu tasks overall. When I want a little more power to gpu tasks, I'll configure it to cpu = 1.98 or however many cores I am not needing for cpu tasks. At least that's my experience with my main computer. My computer with a damaged nvidia 9500gt there doesn't seem to be any advantage to push past cpu = 0.98 ever. I could configure all the cpu threads to work with the gpu and not see any change at all.

Bryan
04-24-14, 09:47 AM
I am wondering, what is the difference between cpu = 1.00 threads and cpu = 0.98 threads. From where I sit, there is no performance improvement between 0.98 and 1.00. That extra 0.02 does get used in cpu tasks. So when I configure it to cpu = 1 I end up shorting my cpu tasks overall. When I want a little more power to gpu tasks, I'll configure it to cpu = 1.98 or however many cores I am not needing for cpu tasks. At least that's my experience with my main computer. My computer with a damaged nvidia 9500gt there doesn't seem to be any advantage to push past cpu = 0.98 ever. I could configure all the cpu threads to work with the gpu and not see any change at all.

That number has nothing to do with how much CPU is actually used by the GPU. The GPU program determines how much is needed. What that number does is tell BOINC how many tasks to schedule at a time. For example, if I have a quad core and I set the GPU to "1" then BOINC will run 3 CPU tasks and reserve 1 core for the GPU. If I set it at "2" then it will only run 2 CPU tasks and reserve 2 cores for the GPU. The GPU won't actually use any more processor than it did at a setting of 1. If you set it at ".1" then it will run 4 CPU tasks and the GPU will have to fight to get serviced and that really kills its throughput.

There are some GPU projects that require more than 1 core/thread of CPU usage ... like the old POEM app.