How to diagnose CUDA 10.1 installation problem on 20.04, “No CUDA-capable devices detected”

I have a ZOTAC ZBOX EN1070K-BE and for the first time I am trying to make use of the GTX. The machine is running Ubuntu 20.04 and I installed nvidia-cuda-toolkit. Test of installation using command "nvcc --version" returns "Cuda compilation tools, release 10.1, V10.1.243". I compiled and executed a test program saxpy.cu from devblogs.nvidia.com; it worked. My real target application is the neural network chess program Leela Zero (lc0). I installed this, but when I execute it in command-line mode it gives me "Creating backend [cudnn-auto]... Switching to [cudnn]... No CUDA-capable devices detected." Can someone advise me what I may do in order to diagnose this problem further?

0
задан 14 June 2020 в 13:24

1 ответ

The problem was diagnosed through the command

sudo nvidia-settings

which told me "ERROR: NVIDIA driver is not loaded." I followed up with

sudo apt install nvidia-prime nvidia-driver-430

and then I rebooted the machine. Nothing else needed.

0
ответ дан 19 June 2020 в 21:25

Другие вопросы по тегам:

Похожие вопросы: