I haven't seen any options for controlling individual cards, as yet. However, I have an ATI 7950 in my wife's 3770K, it's going full tilt with an app_info file, producing 10 workunits every 8 minutes, and there is no lag for 'general windows stuff' - not even a small lag of any kind.

I also had two more 7950s in my 2600K, same deal, same result, for 'general windows stuff'. I even use AutoCAD at the same time, and only notice occasional momentary lags that do not interfere with my work.

As you may already know, the ATI cards are better suited than the nVidias, for doing GPU work on this particular HCC project, so your numbers will be lower.

Best of luck to you, and thanks for your efforts.

I am going to give an experience response here. I am not an I/T or tech person, nor a BOINC genius. But I have been through the GPU mill a couple of times.

Everything I say is just my own experience and anything I say may be disputed by others based upon their experience.

I started with twin GTX580's with and then without and then with SLI. I now have at the computer builders option twin GTX670's.

The original desire for two GPU cards was to crunch two work units at a time. Crunching on two cards was a huge mistake, at least for me.

BOINC wants to be on device 0. My builder wants also to be on device 0.

The considered decision was to enable SLI -although it is much disliked outside of the gamer community. This was to let the twin units act as one big GPU to accomodate both BOINC and the builder. Further, in cc_config, BOINC was told to "ignore dev 1". So, with BOINC and the builder wanting to be on the same card, twin cards with SLI turned out to be a blessing in disguise.

GPU projects are a strange breed. They are all individually compiled. They may or may not "get along" with other GPU projects or with various computers and cards.

My original impetus to run GPU was GPUGrid, a project with exceptional people and values doing great work.

I needed to dismiss GPUGrid for suspicions of its creating computing difficulties.

I drifted toward projects based in Astronomy because my blog has been heavily tilting toward that subject.

I am now running EINSTEIN and SETI on GPU. They seem to get along fine with each other and with my computer, a Maingear Shift Super Stock purchased expressly to crunch GPU work and with no other purpose.

I tried to run MILKY WAY and failed. According to a very major BOINC person, this project has an internal "Resources" problem with OpenCL, which has been forwarded to the project to no avail.

So, don't look for too many settled rules. Try what you want, see what you can get away with, just hope to be successful.

If this post is a bit messy, so has been my experience.

You would create a file called cc_config.xml in the BOINC directory. This would be the same directory containing the file client_state.xml. In the cc_config.xml file, something like this may work:


<cc_config>
<options>
<exclude_gpu>
<url>http://www.worldcommunitygrid.org</url>
<device_num>0</device_num>
<type>nvidia</type>
<app>hcc1</app>
</exclude_gpu>
</options>
</cc_config>

I don't know anything about that. But Milky Way is special. It's the only project that require GPUs that support double precision math. They also had a NV optimized OpenCL app before one for AMD (since they had a CAL app for AMD that didn't work with the 7xxx series). But the 7970 ran the NV optimized OpenCL app 3x faster than any NV card.