Ask Your Question
1

Is it possible to install CUDA toolkit 10 without nVidia proprietary drivers?

asked 2019-03-04 15:34:25 -0600

why2 gravatar image

updated 2019-03-04 15:36:51 -0600

Hi,

I installed the RPMFusion NVidia driver and the CUDA toolkit 10 based on this guide and did the package manager installation. So far so good, until I run the built binary deviceQuery, as described in the post installation section. It returns

./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 100
-> no CUDA-capable device is detected
Result = FAIL

Even after disabling the SELinux with sudo setenforce 0. Under the /dev/ folder I found out /dev/nvidiactl /dev/nvidia-uvm /dev/nvidia-uvm-tools. Any hint?

Thanks.

edit retag flag offensive close merge delete

Comments

Does your GPU actually support CUDA? Not all NVIDIA cards do.

revolucion09 gravatar imagerevolucion09 ( 2019-03-04 19:00:49 -0600 )edit

yes, I have GTX 1060

why2 gravatar imagewhy2 ( 2019-03-07 04:01:46 -0600 )edit

1 Answer

Sort by » oldest newest most voted
0

answered 2019-03-07 04:10:12 -0600

why2 gravatar image

updated 2019-03-07 04:11:07 -0600

I solved the issue. After a reboot, I cannot see the login screen. By typing sudo journalctl -b I can see that there is an incompatibility with the Nvidia kernel module:

Mär 07 10:34:41  /usr/libexec/gdm-x-session[1322]: (EE) NVIDIA: Failed to initialize the NVIDIA kernel module. Please see the
Mär 07 10:34:41  /usr/libexec/gdm-x-session[1322]: (EE) NVIDIA:     system's kernel log for additional error messages and
Mär 07 10:34:41  /usr/libexec/gdm-x-session[1322]: (EE) NVIDIA:     consult the NVIDIA README for details.
Mär 07 10:34:41  kernel: NVRM: API mismatch: the client has the version 418.39, but
                                NVRM: this kernel module has the version 418.43.  Please
                                NVRM: make sure that this kernel module and all NVIDIA driver
                                NVRM: components have the same version.

It was because the RPMFusion-Non-Free package install the newer version of the kernel module 418.43 and the CUDA driver support the version 418.39. Thus I check the package and I can see that due to my last installation the kmod-nvidia couldn't not be removed automatically. So I did:

sudo dnf remove kmod-nvidia-4.20.13-200.fc29.x86_64-3:418.43-1.fc29.x86_64

and then install the correct version:

sudo dnf install kmod-nvidia-418.39

Now, the CUDA check deviceQuery returns a good result:

Compute Mode:
    < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.1, CUDA Runtime Version = 10.1, NumDevs = 1
Result = PASS
edit flag offensive delete link more

Question Tools

1 follower

Stats

Asked: 2019-03-04 15:34:25 -0600

Seen: 428 times

Last updated: Mar 07 '19