Linux – How to select which graphics card to use

graphicslinux

I have a desktop machine (running Ubuntu 13.10) with two video outputs (both nVidia):

an integrated graphics card, which I want to use for my desktop display, and
a PCI graphics card, which I would like to use for CUDA based tasks.

Is there a way to select which one to use for the X server and which for CUDA, and make it permanent?

Best Answer

I believe you can follow the directions outlined here in this post titled: Multiple Nvidia Video Cards & Monitors on Linux Mint 15 MATE. Even though the details are targeting Linux Mint they should still be applicable, since the bulk of the tweaking is being done through the NVIDIA drivers.

I would assume that you can do so either through the NVIDIA dialogs or through your xorg.conf file.

dialogs

ss #1

xorg.conf files

Section "Device"
    Identifier     "Device2"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "GeForce GT 610"
    BusID          "PCI:1:0:0"
    Screen          1
EndSection

Section "Device"
    Identifier     "Device1"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "GeForce 6200"
    BusID          "PCI:5:4:0"
EndSection

Section "Device"
    Identifier     "Device0"
    Driver         "nvidia"
    VendorName     "NVIDIA Corporation"
    BoardName      "GeForce GT 610"
    BusID          "PCI:1:0:0"
    Screen          0
EndSection

Then just only assign the displays to the internal "device":

Section "Screen"
    Identifier     "Screen1"
    Device         "Device1"
    Monitor        "Monitor1"
    DefaultDepth    24
    Option         "Stereo" "0"
    Option         "nvidiaXineramaInfoOrder" "CRT-0"
    Option         "metamodes" "nvidia-auto-select +0+0"
    SubSection     "Display"
        Depth       24
    EndSubSection
EndSection

This is just to give you the general approach, see the article for further details.

Related Solutions

Display sleep: “sleep” graphics card

Preface: Whether this works is highly dependent on your hardware.

Since you're using a Radeon card and all more recent cards(starting with GCN 1.0 circa 2011) support something called ZeroPower, the first step would be to check if it's maybe just a DPM issue. Try forcing your card into a low power state. I assume that you're using the open source drivers since fglrx pretty much useless on Arch. Try

    echo low > /sys/class/drm/card0/device/power_dpm_force_performance_level

to force the card into it's lowest power state, turn of your screen and see if the fan turns off. For more information about DPM you can look here.

If the above fails or you simply want to try it out, you can also try removing/disabling the card. A word of caution: Playing with PCIe hotplug can be a very entertaining way to crash your system. To do so, stop Xorg and unload the kernel module your GPU driver uses(probably radeon in your case). Afterwards, find out how your CPU is called (lspci, some line will contain somthing like 01:00.0 VGA compatible controller). Using this number, you can remove the GPU from the bus by doing(adjust the numbers)

    echo 1 > /sys/bus/pci/devices/0000\:01\:00.0/remove

This will not turn off the power but hopefully cause the GPU to power down since it's no longer attached. To reattach it you can try to redetect it using

    echo 1 > /sys/bus/pci/rescan

via SSH or rebooting the machine (probably also via SSH).

Best Answer

Related Solutions

Display sleep: “sleep” graphics card

Related Question