Select Slot of Main Graphic Card

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everybody,

I have a system with three graphic cards, two of them are Fermi TESLA C2070, and the other one is a QUADRO 295. The Fermi ones are connected to the PCI-Express 16x slots, and the Quadro is connected to the 1x slot. The model of the server is a FUJITSU CELSIUS R670 and the operating system is Ubuntu Server 10.04.

I would like to use the Fermi cards for high performance computing with CUDA and the Quadro for video display, but it seems not possible to use a PCI-E 1x graphic card if a 16x is connected (the Quadro’s slot is not detected by the BIOS). I don’t care if the BIOS just detects the Fermi cards, the main point is to “redirect” the output to the Quadro after booting the operating system.

Is there anything I could configure in BIOS or in X.org?
Thanks in advance.

2

In your xorg.conf file, you can specify the display card by PCI bus ID. My development box has a GTX 275 as a display card and a Fermi card for compute. I have this in xorg.conf:

Section "Device"

    Identifier     "Device0"

    Driver         "nvidia"

    VendorName     "NVIDIA Corporation"

    BoardName      "GeForce GTX 275"

    BusID          "PCI:1:0:0"

    Option         "Coolbits" "1"

EndSection

to force the X11 server to run on the GTX275 and leave the Fermi alone. You only need to set the PCI ID for you setup and you should get X on the right card. There isn’t much you can do about post and bios output at boot, but once X11 starts it should work fine.