TensorFlow : failed call to cuInit: CUDA_ERROR_NO_DEVICE

Question

My test :

import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()`

Error :

c:\l\work\tensorflow-1.1.0\tensorflow\stream_executor\cuda\cuda_driver.cc:405]

failed call to cuInit: CUDA_ERROR_NO_DEVICE

-> but "/cpu:0" works fine

Config :

nvidia-smi :

CUDA Version 9.1
tensorflow-1.1.0
Windows 10
cudnn64_7.dll (installed in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.1\bin)
Only tensorflow-gpu is installed in my Conda environment

Why can't Tensorflow detect my GPU ?

If you're using Tensorflow 1.1 from the prebuild binaries, you need CUDA 8 and cuDNN 5.1. Did you build this yourself? — GPhilo, Feb 07 '18 at 09:24
I have updated Tensorflow to v1.5. The error message disappeared but it is still using my CPU instead of my GPU. Do you what could be the reason? — Vincent Roye, Feb 08 '18 at 07:52
You installed via pip "tensorflow" instead of "tensorflow-gpu" perhaps? — GPhilo, Feb 08 '18 at 07:58
Now I'm on : - Tensorflow-gpu 1.6 - CUDA 9.0 - Cudnn 7.0 Still : failed call to cuInit: CUDA_ERROR_NO_DEVICE — Vincent Roye, Feb 08 '18 at 11:04
I have given more details there : https://github.com/tensorflow/tensorflow/issues/16860 — Vincent Roye, Feb 08 '18 at 11:27
Wait, in the issue you have `os.environ["CUDA_VISIBLE_DEVICES"]="1"`. ow many GPUs do you have on your system? The screenshot only shows one, with device id `0`... — GPhilo, Feb 08 '18 at 16:11

score 36 · Accepted Answer · answered Feb 09 '18 at 23:08

36

The issue was solved on GitHub. This error message will be shown if you set an invalid value for the CUDA_VISIBLE_DEVICES environment variable, e.g. when you only have a single GPU (which has ID 0) and set CUDA_VISIBLE_DEVICES=1 or CUDA_VISIBLE_DEVICES=2.

answered Feb 09 '18 at 23:08

mrry

125,488
26
399
400

6

Seems to be also be happening with CUDA_VISIBLE_DEVICES=-1, which tells it to not use GPU. Tensorflow 2.4.1. – Kenneth Evans Mar 20 '21 at 23:52

score 35 · Answer 2 · edited Aug 09 '18 at 12:45

35

The true answer is to add following line of code:

import os
os.environ['CUDA_VISIBLE_DEVICES'] = "0"

edited Aug 09 '18 at 12:45

Muhammad Saqlain

2,112
4
33
48

answered Aug 09 '18 at 12:35

fly

383
3
2

4

Thank you. Funny enough this solution helped even when explicit "CUDA_VISIBLE_DEVICES=0 python ..." did not work. – Ivan Kovtun Feb 01 '20 at 01:09
1

So many places say to use "-1" to ignore GPU, but that just causes these error messages. Setting to zero makes sense, and it works (go figure) – DAG Dec 15 '21 at 17:38

score 0 · Answer 3 · answered Apr 15 '22 at 22:20

0

If you're using Google Colab, it could be that your runtime is not GPU. Click the “Runtime” dropdown menu. Select “Change runtime type”. Now select GPU.

answered Apr 15 '22 at 22:20

Claude COULOMBE

3,434
2
36
39

TensorFlow : failed call to cuInit: CUDA_ERROR_NO_DEVICE

3 Answers3

Linked