recently multi GPU machine will only use one GPU

Don't tell me, i can guess, the Nvidia driver is a 306.xx or later driver, the Boinc client is a 7.0.27 or later client.

recent drivers combined with Boinc 7.0.27 and later introduces a Wacky Nvidia GPU memory reporting Bug, where eithier the Total or the Available GPU memory is reported wrong,

since Boinc will only use the most capable by default, and since the reported Available GPU memory is wrong, Boinc will only use the most similar GPUs,

Boinc 7.0.36 has a partial fix (the Available GPU memory is fixed, but not the Total GPU memory), while Boinc 7.0.38 has the full fix, Links are in the Boinc 7 changelog thread on the Boinc Forum.

But Boinc 7.0.32 and later introduce a times 10 increase in internally reported flops for the GPU, and that puts existing GPU tasks on the verge of going Maximum Time Exceeded, (new tasks are O.K)

so the GPU cache should be eithier run down and reported (or just completed) prior to upgrade, if you can't do that you can always use Boinc Recsheduler, and use it to protect against -177 errors,

Or you can just use the <use_all_gpus> option in a cc_config.xml to force Boinc to use all the GPUs:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

Just make the above with Notepad, make sure you save it as cc_config.xml (and not as cc_config.xml.txt), drop it in your Boinc Data directory (location is the Boinc startup messages, and the directory is hidden by default), then restart Boinc,

If you're done it right, there should be a line saying so in the Boinc startup messages:

30/11/2012 18:21:37 Config: use all coprocessors

Claggy