Tesla C2070 wants to be default graphics card

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

We recently bought a Tesla C2070 for CUDA development, and we have a hardware issue that we can’t solve. If anyone here is able to help, that would be very appreciated.

The new C2070, unlike the old Teslas, has a DVI port to connect to a monitor. As I understand it, this can be disadvantageous as it limits the time a kernel can run. In any event, we have a second graphics card in our computer which is specifically meant to be the display card, but our ASUS motherboard always selects the Tesla as the default graphics card at boot. There is no option in BIOS to change that.

If the Tesla is removed from the system, the other graphics card runs fine.

Does anyone have ideas on how this can be resolved? If it matters, it is a Ubuntu Linux machine.

Thanks a lot!

2

The watchdog timer on Linux only applies to CUDA devices running X.org. If you configure X to use your other card for the display, you should be fine. (I don’t know how to pick the device for the display in X.org these days, but I’m sure there must be a tutorial somewhere.)

If you don’t run X at all, then you don’t need to do anything.

3

Are you using the latest BIOS ? Check ASUS site for an update…

After you boot to your OS, select the other card as primary display.
Sometimes the OS may writeback something in NVRAM or whatever that the BIOS can understand.

4

Try to use nvidia-smi and specify main video card look at doc nvidia-smi.
nvidia-smi -L (1 2 3 4) if you do not want 1 type nvidia-smi 2 3 4
-Mike