I'm running GPUGRID on my GT240 using nvidia-drivers version 260.19.29 CUDA version 3.2 boinc version 6.10.58-r1 and kernel 2.6.36-gentoo-r5. I'm running WU 749-GIANNI_DHFR1000-11-99-RND0211_1 and the GPU seems to be doing very little work per nvidia-smi. I'm seeing anywhere between 0% and
20% GPU utilization while the WU is running and the GPU temperature is not rising very much. I've tried renicing the process to give it more CPU but that does not seem to affect the GPU utilization #'s. Also it seems like this should take 24 hours max to complete on a GT240 but it's reporting an ETA of 4 days. Any suggestions?(I'm thinking I can run two or three WU's at once to get around this issue but I'm not sure if that would be optimal) FYI folding@home always used somewhere between 97-99 percent of the GPU.


Thanks,

Chad

I saw that post but I didn't see anything that appeared to apply.

To run GPUGrid tasks on a Fermi GPU find the app_info.xml file and include the following,

<app_info>
<coproc>
<type>CUDA</type>
<count>0.5</count>
</coproc>
</app_info>

but my card is CC1.3 so I believe it isn't Fermi correct?

As to the swan_sync thing where would you put that in /etc/env.d/99boinc? I don't see the process for the project bouncing around CPU's so that doesn't seem like it would help.

Your card is CC1.2.

The first thing to check is that your card is running at maximim performance; the 260.x drivers have a tendancy to reduce the clocks, especially on GT240's. You can use NVidia control panel on Win to do this. Can't give exact instructions for gentoo.

Create, or edit an existing, cc_config.xml file in the Program Data folder for Boinc, using a text editor,

Add the following lines:

<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
</options>
</cc_config>

This post might help

Leave one of your CPU cores free; this speeds up the GPU task (99% equates to use 5 out of 6 cpu cores), especially if you use swan_sync=0, which you should also do.

Set the nice value to zero.

Sorry, I'm not a Linux expert, or I would give more precise instructions.

I saw that post but I didn't see anything that appeared to apply.

To run GPUGrid tasks on a Fermi GPU find the app_info.xml file and include the following,

<app_info>
<coproc>
<type>CUDA</type>
<count>0.5</count>
</coproc>
</app_info>

but my card is CC1.3 so I believe it isn't Fermi correct?

As to the swan_sync thing where would you put that in /etc/env.d/99boinc? I don't see the process for the project bouncing around CPU's so that doesn't seem like it would help.

I don't think that this is a good idea: setting it to 0.5 will push 2 WUs on the card but the video memory will be too small to support it and it will crash.

<count>0.5</count> is only feasible on Fermi’s and when trying to run multiple slow tasks; do not use it with a GT240.
FAQ: Best configurations for GPUGRID