Guide to installing two graphic cards ?

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello

i have a computer running X on a quadro 5800 . I have installed a second Tesla card that i want to use ONLY for cuda and there will be no display connected it. Is there a guide or an how to on installing two nvidia graphics cards on the same computer? . I ve done lots of googling but havent found anything definitive . I installed the second card and re installed the cuda 3.1 driver . i also ran nvidia -xconfig but have not had any success . In fact after installing the second card, i can no longer run X as it says no screen found. lscpi returns:

02:00.0 VGA compatible controller: nVidia Corporation Device 05fd (rev a1)
03:00.0 VGA compatible controller: nVidia Corporation Device 06d1 (rev a3)

Any help would be appreciated.Thanks

2

I ran into this same issue. It is documented in the NVIDIA driver README, but this file is only mentioned on the download page for the production NVIDIA drivers, not the developer drivers (that I could find).

In short, run ‘nvidia-xconfig --enable-all-gpus’

Alternatively, one can explicitly specify the BusID for the display GPU in xorg.conf.

See: http://us.download.nvidia.com/XFree86/Linu…ADME/index.html (section 9, “Some X servers have trouble with multiple GPUs”)

  • Matt
3

I ran into this same issue. It is documented in the NVIDIA driver README, but this file is only mentioned on the download page for the production NVIDIA drivers, not the developer drivers (that I could find).

In short, run ‘nvidia-xconfig --enable-all-gpus’

Alternatively, one can explicitly specify the BusID for the display GPU in xorg.conf.

See: http://us.download.nvidia.com/XFree86/Linu…ADME/index.html (section 9, “Some X servers have trouble with multiple GPUs”)

  • Matt
4

thanks matt, i ran

[codebox]update-pcciids

[/codebox]

to recognize the cards and then

specified the bus id in xorg.conf. That did the job

5

thanks matt, i ran

[codebox]update-pcciids

[/codebox]

to recognize the cards and then

specified the bus id in xorg.conf. That did the job

6

just curious, did you install ‘NVIDIA’ driver or devdriver one?

I’ve struggled to get the setting done but devDriver one seems working. So download the driver from the cuda site not nvidia download site.

So my setting is gt8600 for my display and I use gtx 470 for cuda computing.

7

just curious, did you install ‘NVIDIA’ driver or devdriver one?

I’ve struggled to get the setting done but devDriver one seems working. So download the driver from the cuda site not nvidia download site.

So my setting is gt8600 for my display and I use gtx 470 for cuda computing.

8

Hello,

I am trying to enabling two CUDA devices(GTX480 and C2070). Both cards works fine individually.
I tried nvidia-xconfig --enable-all-gpus --no-sli (should I include --multigpu)
After this step, I can see two devices /dev/nvida0/1 and in the /ect/X11/XF86Config file.
But as long as a run a cuda program, it gives this error: CUDA error return 10 : invalid device ordinal

Here is part of the XF86Config file:
Section “Device”
Identifier “Device0”
Driver “nvidia”
VendorName “NVIDIA Corporation”
BoardName “Tesla C2070”
BusID “PCI:4:0:0”
EndSection

Section “Device”
Identifier “Device1”
Driver “nvidia”
VendorName “NVIDIA Corporation”
BoardName “GeForce GTX 480”
BusID “PCI:5:0:0”
EndSection

And the lspci result:
02:00.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)
03:00.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)
03:02.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)
04:00.0 VGA compatible controller: nVidia Corporation Device 06d1 (rev a3)
04:00.1 Audio device: nVidia Corporation Device 0be5 (rev a1)
05:00.0 VGA compatible controller: nVidia Corporation Device 06c0 (rev a3)
05:00.1 Audio device: nVidia Corporation Device 0be5 (rev a1)
06:00.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)
07:00.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)
07:02.0 PCI bridge: nVidia Corporation Device 05b1 (rev a3)

Does anyone know what is the problem?