Hi everyone,

BOINC reports 0.26 CPU for a gpugrid task. However, my hardware monitor reports that the task consumes 1 cpu + 1 gpu.

Is there anyway to tell BOINC that a gpugrid task uses 1 CPU + 1 GPU so that BOINC won't run too many jobs? All jobs a significantly slowed down due to excessive resource contention and context switches.

Is there anyway to tell BOINC that a gpugrid task uses 1 CPU + 1 GPU so that BOINC won't run too many jobs? All jobs a significantly slowed down due to excessive resource contention and context switches.

Thanks. Where can I get or generate a default app_info.xml? I searched forum, but didn't find any.

Create a new file.

Oh, I did not check the operating system. My mistake.
To make that clear, my app_info is for windows, I'm running win7x64.

So could someone with Linux experience please post the necessary changes? THX.

I looked around a little bit.
Saenger posted this Message 19291 - Posted 6 Nov 2010 9:36:13 UTC

<app_info>
<app>
<name>GPUGrid</name>
<user_friendly_name>GPUGrid</user_friendly_name>
</app>
<file_info>
<name>acemd2_6.04_x86_64-pc-linux-gnu__cuda</name>
<executable/>
</file_info>
<file_info>
<name>libcudart.so.2</name>
<executable/>
</file_info>
<app_version>
<app_name>GPUGrid</app_name>
<version_num>604</version_num>
<flops>1.0e11</flops>
<avg_ncpus>0.05</avg_ncpus>
<max_ncpus>0.05</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1.0</count>
</coproc>
<file_ref>
<file_name>acemd2_6.04_x86_64-pc-linux-gnu__cuda</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libcudart.so.2</file_name>
</file_ref>
</app_version>
</app_info>

maybe it's outdated, but could be a start for your experiments.

Thank you very much, Werkstatt and skgiven. I will give this a shot tonight and post the outcome here with my Linux app_info.xml.

PS: I detached several times due to some mysterious hang-ups when my gpu is heavily loaded for hours, along with some blue screens when gaming under Windows. I finally fixed the overclocking settings that I misunderstood and I am coming back crunching numbers. :)

The xml didn't work right away. Looks like some version/file mismatch.
I changed the <name> field to acemd2 and the <version_num> to 613. Then I got the job, but had computation error in 1 minute. Btw, i only want to run acemd standard and beta. Is acemd2 the right app? libcudart.so.2 doesn't exist on server as well. It was libcudart.so.1.xx in gpugrid folder before I use app_info.xml, but it's libcudart.so.3 in my cuda library directory. Which one will gpugrid use? The one in www.gpugrid.net folder or the one in my cuda library path?

I will run a couple of default jobs to see what executable/library acemd uses. Should be easy to fix once I figured out the version and library.

Thank you again for your help. Good to know a solution exists. I will let you know if I make any progress.

Looks like the workload changed since I detached last time. The units I got recently really uses only 4% of CPU, which is excellent! Now I can easily keep GPU busy, without hurting the CPU jobs.

Thank you for all your help. I will wait until next application to see if I need the app_info.xml.