CUDA App not Utilizing Quadro Plex


We have an nVidia Quadro Plex hooked up to a computer with 3 other nVidia cards in it (All nVidia Quadro 5600’s i believe)

The CUDA app I am working on is supposed to divide it’s payload up between all of the cards, but currently it seems to be ignoring the Quadro Plex.
Looking at the devices, we see nVidia 0, 1, and 2, along with vcs0, which is the Quadro Plex.

Any Ideas on why this is happening? My first thought was a bad toolkit install, but I’d rather here what people here have to say.


Does anyone have any Ideas?

Try some of the CUDA SDK examples to see if they work properly. devicequery and bandwidth test should be enough and they both use every GPU.

If those work, then your problem is likely with your app and not the hardware.

I have run the sample applications, and they are not locating the Quadro Plex either.

What OS is this?

CentOS (Linux) x64 Kernel Version 2.6.18-164.9.1.e15 I’m pretty sure it’s the newest version of CentOS