GNOME Desktop will cause CUDA environment does not work.

Dear all,

Before I installed Gnome Desktop, nvidia-smi can show following information.

[root@vm10-0-6-21 ~]# nvidia-smi
Sun Nov 12 17:48:02 2017       
| NVIDIA-SMI 375.26                 Driver Version: 375.26                    |
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|   0  Tesla P40           On   | 0000:00:07.0     Off |                    0 |
| N/A   14C    P8    10W / 250W |      0MiB / 22912MiB |      0%      Default |
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|  No running processes found                                                 |

But after install gnome desktop with "yum groups install “GNOME Desktop” and run nvidia-smi again, I get following error information.

[root@vm10-0-6-21 ~]# nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

My environment is CentOS 7.3 64bit.

How to solve this problem?

Thanks a lot.


The install of gnome desktop seems to have disrupted the GPU driver, perhaps by pulling in a nouveau driver or something else disruptive via a package dependency.

You might want to study exactly what yum groups install “GNOME Desktop” did in terms of packages installed/updated.

To fix it, you probably need to reinstall the GPU driver. Depending on how you installed the driver initially, this may be complicated by whatever your yum install did. The GPU driver can be installed either via a package manager method (yum) or by a runfile installer. Depending on how you installed the GPU driver initially, and what actually transpired during your yum desktop install, you may need to clean up the mess by uninstalling the previous driver.

You can find some instructions about handling conflicting install methods in the linux install guide for CUDA, and the readme distributed with NVIDIA GPU drivers contains information as well.