Arch Linux

hardrock

Member

Registered: 2016-12-20

Posts: 28

CUDA driver version is insufficient for CUDA runtime version

I'm trying to run some pytorch script that uses CUDA.

However I'm running into this error, I'm not sure whats wrong.

RuntimeError: cuda runtime error (35) : CUDA driver version is insufficient for CUDA runtime version at /opt/conda/conda-bld/pytorch_1524584710464/work/aten/src/THC/THCGeneral.cpp:70

I have installed CUDA 10.0.130 and nvidia drivers 415.18-4 and according to the CUDA 10 release notes, it should be enough: https://docs.nvidia.com/cuda/cuda-toolk … index.html

root@hardrock [1] hit # pacman -Ss nvidia | grep installed
extra/libvdpau 1.1.1+3+ga21bf7a-1 [installed]
extra/nvidia 415.18-4 [installed]
extra/nvidia-utils 415.18-1 [installed]
extra/opencl-nvidia 415.18-1 [installed]
community/cuda 10.0.130-2 [installed]

Lone_Wolf

Member

From: Netherlands, Europe

Registered: 2005-10-04

Posts: 11,920

Re: CUDA driver version is insufficient for CUDA runtime version

/opt/conda/conda-bld/pytorch_1524584710464/work/aten/src/THC/THCGeneral.cpp:70

are you using this conda ?

Package, dependency and environment management for any language—Python, R, Ruby, Lua, Scala, Java, JavaScript, C/ C++, FORTRAN

It seems that provides the conda part of the infamous anaconda (not the snake).

$ pacman -Ss torch
community/python-pytorch 1.0rc1-5
    Tensors and Dynamic neural networks in Python with strong GPU acceleration
community/python-pytorch-cuda 1.0rc1-5
    Tensors and Dynamic neural networks in Python with strong GPU acceleration
$

We have 2 pytorch packages in community repo and the name of the second one includes cuda.
Native archlinux packages compiled from source tend to work better then things run through some framework system, try python-pytorch-cuda.

Last edited by Lone_Wolf (2018-11-28 16:14:17)

---

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

(A works at time B)  && (time C > time B ) ≠  (A works at time C)