When CUDA_ERROR_UNKNOWN comes up to call cuInit,

The detailed error messge is

(venv) user@LifeNTech:~/Workspace/deep_neural_network$ python test_keras.py Using TensorFlow backend. Epoch 1/5 2019-02-11 17:45:26.303774: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA 2019-02-11 17:45:26.315447: E tensorflow/stream_executor/cuda/cuda_driver.cc:397] failed call to cuInit: CUDA_ERROR_UNKNOWN 2019-02-11 17:45:26.315530: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:163] retrieving CUDA diagnostic information for host: LifeNTech 2019-02-11 17:45:26.315552: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:170] hostname: LifeNTech 2019-02-11 17:45:26.315635: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:194] libcuda reported version is: 390.87.0 2019-02-11 17:45:26.315687: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:198] kernel reported version is: 390.87.0 2019-02-11 17:45:26.315695: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:305] kernel version seems to match DSO: 390.87.0 

This thing  might be helpful.
# nvidia-modprobe -u