InternalError: cudaGetDevice() failed. Status: cudaGetErrorString symbol not found

cudagpupython

I run this code:

tf.test.is_gpu_available( cuda_only=False, min_cuda_compute_capability=None )

I get the following error:

2019-10-25 18:25:20.855191: I
tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports
instructions that this TensorFlow binary was not compiled to use: AVX2
2019-10-25 18:25:20.879831: I
tensorflow/stream_executor/platform/default/dso_loader.cc:44]
Successfully opened dynamic library nvcuda.dll 2019-10-25
18:25:21.461924: I
tensorflow/core/common_runtime/gpu/gpu_device.cc:1618] Found device 0
with properties: name: GeForce MX130 major: 5 minor: 0
memoryClockRate(GHz): 1.189 pciBusID: 0000:01:00.0 2019-10-25
18:25:21.470775: I
tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25]
GPU libraries are statically linked, skip dlopen check. 2019-10-25
18:25:21.503654: I
tensorflow/core/common_runtime/gpu/gpu_device.cc:1746] Adding visible
gpu devices: 0
————————————————————————— InternalError Traceback (most recent call
last) in
—-> 1 tf.test.is_gpu_available( cuda_only=False, min_cuda_compute_capability=None )

~\Anaconda3\envs\deep_env\lib\site-packages\tensorflow_core\python\framework\test_util.py
in is_gpu_available(cuda_only, min_cuda_compute_capability) 1430
1431 try:
-> 1432 for local_device in device_lib.list_local_devices(): 1433 if local_device.device_type == "GPU": 1434 if
(min_cuda_compute_capability is None or

~\Anaconda3\envs\deep_env\lib\site-packages\tensorflow_core\python\client\device_lib.py
in list_local_devices(session_config)
39 return [
40 _convert(s)
—> 41 for s in pywrap_tensorflow.list_devices(session_config=session_config)
42 ]

~\Anaconda3\envs\deep_env\lib\site-packages\tensorflow_core\python\pywrap_tensorflow_internal.py
in list_devices(session_config) 2247 return
ListDevicesWithSessionConfig(session_config.SerializeToString())
2248 else:
-> 2249 return ListDevices() 2250 2251

InternalError: cudaGetDevice() failed. Status: cudaGetErrorString
symbol not found.

After crearing the following venv:

conda create -n Deep_learning_env python=3.6

pip install -U numpy matplotlib pandas ipython
git clone https://github.com/scipy.git scipy
pip install https://cntk.ai/PythonWheel/CPU-Only/cntk-2.7.post1-cp36-cp36m-win_amd64.whl
pip install https://cntk.ai/PythonWheel/GPU/cntk_gpu-2.7.post1-cp36-cp36m-win_amd64.whl
pip install tensorflow
pip install tensorflow-gpu
pip install gensim
pip install keras
pip install  --upgrade --no-deps cntk
pip install --upgrade --no-deps cntk-gpu

conda install theano pygpu
conda install -c peterjc123 pytorch
conda install -c anaconda cudatoolkit
conda install -c anaconda cudnn

conda list cudnn
# Name   Version  Build      Channel
cudnn    7.6.0    cuda10.1_0    anaconda
nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:12:52_Pacific_Daylight_Time_2019
Cuda compilation tools, release 10.1, V10.1.243
python --version

Python 3.6.7
which nvcc
/c/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v10.1/bin/nvcc

NVIDIA GeForce MX130

more info:
cntk 2.7
cntk-gpu 2.7
tensorflow 2.0.0
tensorflow-estimator 2.0.1
tensorflow-gpu 2.0.0
ipython 7.8.0

also when I import tensorflow I get the following warning:

tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could
not load dynamic library 'cudart64_100.dll'; dlerror: cudart64_100.dll
not found

Best Answer

You should checkout this: Github tensorflow Currently Cuda 10.1 is compatible with TF

Related Solutions

The suggested way to handle pip(easy_install) with homebrew

Maybe it is time to consider using virtualenv. Virtualenv creates self-contained python environments using the python version you specify. After activating the new virtual environment, everything you install using pip goes under that environment. This helps avoid situations like the one you described.

E.g. create and activate a new python environment using the default python:

# create environment
$ virtualenv --distribute myproject
New python executable in myproject/bin/python
Installing distribute...done.
Installing pip...done.

# activate environment
$ . ./myproject/bin/activate

# check default python
(myproject)$ which python
/Users/me/myproject/bin/python

It is suggested to use the --distribute options to indicate that distribute should be used for installing packages in the new environment instead of (the older) setuptools. After activation your command prompt changes to indicate which python environment is active.

Now install some package. The files will go into myproject directory:

# install django
(myproject)$ pip install django
...

# search for django dir
(myproject)$ find myproject -iname django
myproject/lib/python2.7/site-packages/django

Finally, deactivate:

# deactivate and check for default python
(myproject)$ deactivate
$ which python
/usr/bin/python

To create an environment using a non-default version of python:

$ virtualenv --distribute -p /path/to/custom/python mynewproject

By default virtualenv will copy to the new environment any packages installed for the python version you use to bootstrap it. To prevent this and create an empty environment use the --no-site-packages option. This is especially useful to create environments which can be exactly replicated e.g. from development to production.

Update: As of version 1.7 --no-site-packages has become the default behaviour of virtualenv.

If you want more details, there are plenty of tutorials and blog posts online. E.g.:

Notes on using pip and virtualenv with Django. (most of the post is not django-specific)
Working with virtualenv.

Give it a try and I'm sure you'll stick with it.

Note: Make sure that your executable scripts do not have the python interpreter hardcoded. I.e. their first line should be #!/usr/bin/env python and not something like #!/usr/bin/python.

Linux – Test tensorflow-gpu failed with Status: CUDA driver version is insufficient for CUDA runtime version (which is not true)

I'm sorry, I've already found the solution myself. My mistake was that I created the Anaconda environment with a python version >3.5. Under these circumstances tensorflow-gpu=1.13 will be installed, if you execute the following command:

conda install -c anaconda tensorflow-gpu

However, if you create an environment with python=3.5, tensorflow-gpu=1.10 will be installed, which works for this CUDA version.

Best Answer

Related Solutions

The suggested way to handle pip(easy_install) with homebrew

Linux – Test tensorflow-gpu failed with Status: CUDA driver version is insufficient for CUDA runtime version (which is not true)

Related Question