tensorflow: tensorflow-gpu-1.8: failed call to cuInit: CUDA_ERROR_UNKNOWN

System information

  • No custom code, Validate you installation test from tensorflow website: Validate your installation test code
  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04.4
  • TensorFlow installed from (source or binary): with pip install --upgrade tensorflow-gpu
  • TensorFlow version (use command below): 1.8 gpu
  • Python version: 2.7
  • CUDA/cuDNN version: 9.0 / 7
  • GPU model and memory: Nvidia GeForce GTX 1080 Ti 11GB
  • Exact command to reproduce: # Python import tensorflow as tf hello = tf.constant('Hello, TensorFlow!') sess = tf.Session() print(sess.run(hello))

Describe the problem

I installed tensorflow and then tried to run the “validate your installation” program from the website. # Python import tensorflow as tf hello = tf.constant('Hello, TensorFlow!') sess = tf.Session() print(sess.run(hello)) After sess = tf.Session() im recieving the following error: 2018-05-14 13:19:32.098811: E tensorflow/stream_executor/cuda/cuda_driver.cc:406] failed call to cuInit: CUDA_ERROR_UNKNOWN 2018-05-14 13:19:32.098874: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:158] retrieving CUDA diagnostic information for host: censored 2018-05-14 13:19:32.098889: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:165] hostname: censored 2018-05-14 13:19:32.098935: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:189] libcuda reported version is: 390.48.0 2018-05-14 13:19:32.098992: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:193] kernel reported version is: 390.48.0 2018-05-14 13:19:32.099006: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:300] kernel version seems to match DSO: 390.48.0

This Issue https://github.com/tensorflow/tensorflow/issues/16860 did not resolve my problem.

Hopefully you can help me with this issue.

About this issue

  • Original URL
  • State: closed
  • Created 6 years ago
  • Reactions: 5
  • Comments: 15 (5 by maintainers)

Most upvoted comments

installing nvidia-modprobe solved it for me. sudo apt install nvidia-modprobe Ubuntu 18.04LTS Cuda 9.0

Just to clarify, your card wasn’t showing ECC errors, people just appear to be reading the table wrong:

There are four headings in that table: Volatile, Uncorrectable ECC Errors, GPU Utilisation and Compute Mode. They correspond to the table below it in the following order - top left, top right, bottom left, bottom right: (blank), N/A, Default, 0%.

I assume the card doesn’t have ECC RAM which is why the ECC readout is N/A. But in short, nvidia-smi wasn’t showing any errors.

For me install nvidia-modprobe solved this problem. The command is sudo apt install nvidia-modprobe. If that doesn’t work, you can try reboot. If you are working on remote/shared server that cannot be rebooted, you can also try to reinstall the driver.