Using Linux with the onboard videoboard and CUDA with the PCIe

Accelerated Computing CUDA CUDA Programming and Performance
1

Basically, how can I use Linux with the onboard videoboard of my system, and have my GeForce 8500 GT exclusively for CUDA?

Thanks in advance.

2

Yes.
When you install the Nvidia driver, do not modify the X settings. Use the script in the release note to create the /dev/nvidia* entries and load the driver.

This is the standard way of operation on cluster nodes.

3

I have to reinstall the driver, or simple configuring back the X to use the onboard GPU is enough?

4

Configuring X to the use on board should be enough.

5

Thank you, I’ll try it asap.

6

Surprisingly, I’ve never had to do this in any of my multiple Linux installs. I just install and point X at my display card, writing the xorg.conf file and restart X and all GPUs are recognized for CUDA.

This is even true for the extreme cases like the funky MSI-G65 board with onboard 8200 embedded display, and I have 2 GTX295s and 1 GTX480 in it. It Just Works.

Perhaps the /dev/nvidia hackery is only needed when your display card is NOT an NVidia GPU?

Or maybe because I use Ubuntu and it’s different for some distributions?