I thought I'd try crunching this again after a long absence. I found that the app ignores BOINC settings and uses every GPU in the system. Apparently this is how it is supposed to work. This is not desirable for a number of reasons. Is there a way to limit it to one?

You can create an app_config.xml that should give you the desired results that you want.

Thanks for the response. Each GPU has its own client, with GPU exclusions defined in cc_config so that shouldn't be necessary. I went ahead and tried this:

<app_config>

<app_version>
<app_name>dnetc</app_name>
<plan_class>cuda31</plan_class>
<avg_ncpus>1</avg_ncpus>
<ngpus>1</ngpus>
</app_version>

</app_config>

(Code tags seem broken, using it put the above all on one line.)

It had no effect.

I also tried a project-specific exclusion in cc_config and that didn't work either.

I have found I can suspend BOINC and edit the dnetc-gpu-1.3.ini file in the slot, changing run-work-limit from -1 to 1. Obviously that's not a practical solution. Edits to that file in the project folder do not hold, I suppose because changes cause it to no longer match the file signature and it downloads a new one.

Is there anything else I might try?

Thanks for the response. Each GPU has its own client, with GPU exclusions defined in cc_config so that shouldn't be necessary. I went ahead and tried this:

<app_config>

<app_version>
<app_name>dnetc</app_name>
<plan_class>cuda31</plan_class>
<avg_ncpus>1</avg_ncpus>
<ngpus>1</ngpus>
</app_version>

</app_config>

(Code tags seem broken, using it put the above all on one line.)

It had no effect.

I also tried a project-specific exclusion in cc_config and that didn't work either.

I have found I can suspend BOINC and edit the dnetc-gpu-1.3.ini file in the slot, changing run-work-limit from -1 to 1. Obviously that's not a practical solution. Edits to that file in the project folder do not hold, I suppose because changes cause it to no longer match the file signature and it downloads a new one.

Is there anything else I might try?

Do this:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<exclude_gpu>
<url>http://moowrap.net/</url>
<device_num>1</device_num>
</exclude_gpu>
</options>
</cc_config>

That tells Boinc to use all the gpu's in your system but to EXCLUDE gpu #1 at Moo. Put the file in the Boinc folder in c:\projectdata\boinc and stop and restart Boinc and it should stop using gpu #1. If you want it to crunch with gpu #1 but not gpu #2 just edit the file and change the 1 to a 2 and exit and restart Boinc. Be sure to save the file as a txt type file in NOTEPAD if you use Windows as a word processing program adds stuff that Boinc can't read so messes everything up.

I also tried a project-specific exclusion in cc_config and that didn't work either.

Do this:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<exclude_gpu>
<url>http://moowrap.net/</url>
<device_num>1</device_num>
</exclude_gpu>
</options>
</cc_config>

Thanks for the response. I had already tried that before posting. As I said, a project-specific exclusion in cc_config does not work. That includes this:

<exclude_gpu>
<url>http://moowrap.net/</url>
<device_num>1</device_num>
<type>NVIDIA</type>
</exclude_gpu>

as well as what you suggested.

I've been using this since I set up the client and it doesn't work either:

<ignore_nvidia_dev>1</ignore_nvidia_dev>

Is anyone successfully using cc_config or app_config to limit the CUDA app?

Hi,
same problem here. Two cards #0 and #1.
At first I got tasks with "..(0.2 CPUs + 2 NVIDIA GPU)"
I try to exclude #1.
cc_config:
...
<exclude_gpu>
<url>http://moowrap.net/</url>
<device_num>1</device_num>
</exclude_gpu>
...
This doesn't help. Next I add:

<ignore_nvidia_dev>1</ignore_nvidia_dev>

Restart Boinc, but still no tasks starts, because there is only one available for the project and WUs with two GPUs in the queue.
I abort all tasks, reload new tasks, same problem.
I restart Boinc, nothing change.
So I reset the project, restart the PC and now I got new tasks for only one GPU.

But now I can't use GPU #1 for other projects. It works only with:
<ignore_nvidia_dev>1</ignore_nvidia_dev>

Conclusions:
To solve the problem you must:
- reset the project in BOINC and
- reboot the PC. Only restarting BOINC doesn't work.

Another problem:
At first Moowrap needs additional two full cores, not 0.2 CPUs.
Some days later, after some changes, the WU needs now one full CPU, but the GPU is only running 20 minutes and 60 minutes only with CPU.
It‘s very confusion.

I'm not having the multiple CPU on one work unit problem, but clearly this project doesn't work reliably with BOINC. Projects running on VirtualBox are better behaved than this one, as far as respecting BOINC restrictions. I am crunching on an AMD card for now, but that is not optimal. Hopefully someone has a solution.

I'll be ending participation here shortly and I just wanted to say that I never saw or found any direct solution to the problem on the boards here or anywhere else. Using a (lesser) single AMD card was the only solution I found. For some, you have to either take a GPU out or live with the app using them all.