I have a linux box running CentOS 7.9 (fully updated). I have installed the latest Cuda Toolkit (460.32.03). Everything work just fine for my primary VGA card, a GeForce GT 640, but not for two additional GT 730 cards installed in the system.
My intention is to use the GT 730 cards for data processing only (i.e., for some GPU API) and the GT 640 card for both video output and data processing.
‘lspci | grep VGA’ gives
01:00.0 VGA compatible controller: NVIDIA Corporation GK107 [GeForce GT 640] (rev a1)
03:00.0 VGA compatible controller: NVIDIA Corporation GF108 [GeForce GT 730] (rev a1)
83:00.0 VGA compatible controller: NVIDIA Corporation GF108 [GeForce GT 730] (rev a1)
but ‘nvidia-smi’ shows only the GT 640 card:
Sat Feb 13 15:20:47 2021
±----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 GeForce GT 640 On | 00000000:01:00.0 N/A | N/A |
| 30% 37C P8 N/A / N/A | 140MiB / 1996MiB | N/A Default |
| | | N/A |
±------------------------------±---------------------±---------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
I have added “Device” sections for the 3 VGA cards in /etc/xorg.conf and added “GPUDevice” keywords accordingly in the “Screen” section:
Section “Device”
Identifier “Device0”
Driver “nvidia”
BoardName “GeForce GT 640”
BusID “PCI:1:0:0”
VendorName “NVIDIA Corporation GK107 [GeForce GT 640] (rev a1)”
EndSection
Section “Device”
Identifier “Device1”
Driver “nvidia”
BoardName “GeForce GT 730”
BusID “PCI:3:0:0”
VendorName “NVIDIA Corporation GF108 [GeForce GT 730] (rev a1)”
EndSection
Section “Device”
Identifier “Device2”
Driver “nvidia”
BoardName “GeForce GT 730”
BusID “PCI:83:0:0”
VendorName “NVIDIA Corporation GF108 [GeForce GT 730] (rev a1)”
EndSection
Section “Screen”
Identifier “Screen0”
Monitor “Monitor0”
Device “Device0”
GPUDevice “Device0”
GPUDevice “Device1”
GPUDevice “Device2”
DefaultDepth 24
SubSection “Display”
Viewport 0 0
Depth 24
Modes “1920x1080”
EndSubSection
EndSection
But upon starting, the X server replies with
[ 30.737] () |–>Screen “Screen0” (0)
[ 30.737] () | |–>Monitor “Monitor0”
[ 30.737] () | |–>Device “Device0”
[ 30.737] () | |–>GPUDevice “Device0”
[ 30.737] () | |–>GPUDevice “Device1”
[ 30.737] () | |–>GPUDevice “Device2”
[ 30.738] (II) Platform probe for /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card0
[ 30.753] (–) PCI:*(1@0:0:0) 10de:0fc1:3842:2645 rev 161, Mem @ 0xde000000/16777216, 0xc0000000/268435456, 0xd0000000/33554432, I/O @ 0x00008000/128, BIOS @ 0x???/524288
[ 30.754] (–) PCI: (3@0:0:0) 10de:0f02:10de:0825 rev 161, Mem @ 0xdc000000/16777216, 0x381ff0000000/134217728, 0x381ff8000000/33554432, I/O @ 0x00007000/128, BIOS @ 0x???/524288
[ 30.754] (–) PCI: (131@0:0:0) 10de:0f02:0000:0000 rev 161, Mem @ 0xf8000000/33554432, 0x383ff0000000/134217728, 0x383ff8000000/67108864, I/O @ 0x0000f000/128, BIOS @ 0x???/524288
and then
[ 31.041] (WW) NVIDIA(0): The NVIDIA GeForce GT 730 GPU installed in this system is
[ 31.041] (WW) NVIDIA(0): supported through the NVIDIA 390.xx Legacy drivers. Please
[ 31.041] (WW) NVIDIA(0): visit Unix Drivers | NVIDIA for more
[ 31.041] (WW) NVIDIA(0): information. The 460.32.03 NVIDIA driver will ignore this
[ 31.041] (WW) NVIDIA(0): GPU. Continuing probe…
I am clueless. The web page at “developer . nvidia . com / cuda-gpus” states that both GT 640 and GT 730 are CUDA-enabled and the nvidia driver web page (at “www . nvidia . com / en-us / geforce / drivers”) offers the nvidia drive 460.32 for both GT 640 and GT 730 boards.
Am I missing some necessary setup to make my 3 VGA cards accessible to CUDA 11.2?
The output of nvidia-bug-report.sh is:
nvidia-bug-report.log.gz (1.1 MB)