tensorflow: CUDA_ERROR_NO_DEVICE
I just installed TensorFlow for Ubuntu using the instructions on the website.
If I open a session in python:
import tensorflow as tf
hello = tf.constant('Hello, TensorFlow!')
sess = tf.Session()
I get these warnings:
I tensorflow/core/common_runtime/local_device.cc:25] Local device intra op parallelism threads: 8
E tensorflow/stream_executor/cuda/cuda_driver.cc:466] failed call to cuInit: CUDA_ERROR_NO_DEVICE
I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:86] kernel driver does not appear to be running on this host (joao): /proc/driver/nvidia/version does not exist
I tensorflow/core/common_runtime/gpu/gpu_init.cc:112] DMA:
I tensorflow/core/common_runtime/local_session.cc:45] Local session inter op parallelism threads: 8
What have I done wrong?
About this issue
- Original URL
- State: closed
- Created 9 years ago
- Comments: 19 (1 by maintainers)
Commits related to this issue
- Merge pull request #255 from ROCmSoftwarePlatform/develop-upstream-update-r1.12-doc Develop upstream update r1.12 doc — committed to Cerebras/tensorflow by whchung 6 years ago
- Add initial liveness analysis including test cases. Closes #255 PiperOrigin-RevId: 284935454 Change-Id: Ib8e85cae6b07e8eaa6e94359646f86942fa26011 — committed to tensorflow/tensorflow by dfki-mako 5 years ago
- Add initial liveness analysis including test cases. Closes #255 PiperOrigin-RevId: 284979684 Change-Id: I934378b3e1107a179d12bc995777e871aacd7508 — committed to tensorflow/tensorflow by pifon2a 5 years ago
Do you have a GPU in your machine that you want to use?
If not:
I had a problem: failed call to cuInit: CUDA_ERROR_NO_DEVICE. And root of problem was enviroment varible CUDA_VISIBLE_DEVICES = 1. One is ID, not count. Switch it to 0.
Just to expand on what @Alek-dr mentioned, if on your system the environment variable CUDA_VISIBLE_DEVICES is empty or set to any value other than IDs of GPUs on your system then you might encounter this issue. On my system
echo $CUDA_VISIBLE_DEVICES=NoDevFilesfor some reason. Bigger problem? May be.I have two GPUs installed so setting
export CUDA_VISIBLE_DEVICES=0,1fixes the issue and TensorFlow is able to use both of them.I think nvidia-driver is not working.
You have to check command
nvidia-smiIf these command is not working, you have reboot your computer.
Maybe
nvidia-smiworks correctly.Just fixed the issue. You must check if the script you are running doesn’t set the variable to “”.
so 2.1 capability devices won’t be much faster than CPU?
This doesn’t solve my issue. I ran
tf.config.list_physical_devices()and I still can’t see the GPU displayed, despite that error message is gone.