cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version

Question

I get the following error when l run tensorflow in GPU.

2018-09-15 18:56:51.011724: E tensorflow/core/common_runtime/direct_session.cc:158] Internal: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version
Traceback (most recent call last):
  File "evaluate_sample.py", line 160, in <module>
    tf.app.run(main)
  File "/anaconda3/envs/tf/lib/python2.7/site-packages/tensorflow/python/platform/app.py", line 125, in run
    _sys.exit(main(argv))
  File "evaluate_sample.py", line 123, in main
    with tf.Session() as sess:
  File "/anaconda3/envs/tf/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 1494, in __init__
    super(Session, self).__init__(target, graph, config=config)
  File "/anaconda3/envs/tf/lib/python2.7/site-packages/tensorflow/python/client/session.py", line 626, in __init__
    self._session = tf_session.TF_NewSession(self._graph._c_graph, opts)
tensorflow.python.framework.errors_impl.InternalError: Failed to create session.

Where do the following errors come from ?

E tensorflow/core/common_runtime/direct_session.cc:158] Internal: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version

and tensorflow.python.framework.errors_impl.InternalError: Failed to create session.

Such tha my version of :

tensorflow is : 1.10

cat /proc/driver/nvidia/version

NVRM version: NVIDIA UNIX x86_64 Kernel Module 390.77 Tue Jul 10 18:28:52 PDT 2018

GCC version: gcc version 7.3.0 (Debian 7.3.0-28)

nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver

Copyright (c) 2005-2016 NVIDIA Corporation

Built on Sun_Sep__4_22:14:01_CDT_2016

Cuda compilation tools, release 8.0, V8.0.44

Solution 1

Updating nvidia driver solved this issue.

You can check your cuda toolkit compatiblity here. Then update your nvidia driver by downloading it from here.

Solution 2

The reason for this error is the mismatch of your installed Cuda Toolkit version and the version of the python package cudatoolkit, which is usually installed as dependency of tensorflow-gpu.

In order to fix this you have to first match your tensorflow version with your installed Cuda Toolkit version like shown here

Then you have to check the version of your cudatoolkit package. This have to match major and minor version, so e.g. if you have Cuda Toolkit 9.0 installed and cudatoolkit9_1 is installed you need to downgrade to cudatoolkit9 via your python.

Solution 3

In the case I just solved, it was updating the GPU driver to the latest and installing the cuda toolkit. Your error is telling you your CUDA driver version is too old. I believe the nvcc version we were seeing was 7.5, and you have 7.3.

I think all you will have to do is: sudo apt install nvidia-cuda-toolkit then reboot.

Below are the steps I took for the problem where the libcuda.so.1 file could not be found.

First, the ppa was added and a newer GPU driver installed:

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update
sudo apt install nvidia-390

After adding the ppa, it showed options for driver versions, and 390 was the latest 'stable' version that was shown.

Then install the cuda toolkit:

sudo apt install nvidia-cuda-toolkit

Then reboot:

sudo reboot

It updated the drivers to a newer version than the 390 originally installed in the first step (it was 410; this was a p2.xlarge instance on AWS).

Solution 4

Just update your nvidia drivers and it will solve the issue

View more solutions

Comments

[James Adams]

Thanks, this was helpful. BTW I followed the above, also on a p2.xlarge instance, but I didn't need to reboot.

[pumpkin_cat]

Provide the steps to update the driver along with the specific driver version if possible