ImportError: Could not find 'cudart64_90.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Download and install CUDA 9.0 from this URL: https://developer.nvidia.com/cuda-90-download-archive

주피터 노트북에서 케라스로 뭘 좀 해보려는데 위와 같은 문제가 생겼다.
새로운 가상환경에 Tensorflow-gpu 버전을 설치해서 일어난 문제였다.
아무튼!
인터넷에서 검색해보니 아래와 같은 해결 방법이 나왔다.

I got this problem when I try to do something with Keras on Jupyter notebook.
because I made a new virtual environment and I installed Tensorflow-gpu version on there.
Anyway!
I found something after I surfed a little bit on the internet world.

Alright, I have a workaround here that allows you to install CUDA 9.1 with most VS integration.

1. Use the VS installer to install VC++ 2015.3 Tool Set (v140)
2. Use 7zip to extract the local CUDA installer to a folder.
3. Run setup.exe from that folder
   • At Options, select "Custom (Advanced)"
   • Under CUDA, UNCHECK Visual Studio Integration
   • Proceed with the rest of the installation
4. Once that's installed, go back to the folder you extracted everything to
5. Go to 'CUDAVisualStudioIntegration' within that folder
6. Use the two msi installers to install Nsight and NVTX
7. Copy the files from 'extras\visual_studio_integration\MSBuildExtensions' into 
   'C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\VC\VCTargets\BuildCustomizations'
   AND
   'C:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\V140\BuildCustomizations'

That should get you to the point where projects will compile and run from VS.
the files in 'CudaProjectVsWizards' look like the templates that let you start new CUDA projects and make new CUDA files within VS. It probably wouldn't take too much research to figure out where those go.

I'm not sure where 'CUDAVisualStudioIntegrationExt.dll' and 'CUDAVisualStudioIntegration.nvi' are meant to go.

이 문서에 따르면 Keras를 Tensorflow-gpu와 함께 사용하려면 해야할 것이 엄청 많다.

1. 사용하고 있는 그래픽 카드가 CUDA를 사용할 수 있는 것인지 알아볼 것.

2. 'Visual Studio'를 설치 (뭐라구요?)

3. 'CUDA' 설치 (저는 9.0 사용 중)

4. 'cuDNN' 설치 (저는 7.2.1 사용 중)





This document said for using Keras with Tensorflow-gpu, you need to do so a lot.

1. Check your graphic card is good to use CUDA(parallel computing platform on GPU)

2. Install 'Visual Studio' (What!?)

3. Install 'CUDA' (mine is 9.0)

4. Install 'cuDNN' which is fit to your CUDA version' (mine is 7.2.1)


더 많은 정보가 필요하시다면 여기.

If you need more information check this link.




그리고 다 잘되는지 아래 명령어로 확인해 볼 수 있어요!

And you can check with this command everything is working well

import tensorflow as tfsess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

이런 식으로

like this


이게 잘 돌아가면 이제 더 깊은 딥러닝의 세계로 갈 준비가 된 겁니다.

If it is working, you are good to go to the deeper side of deep learning.





https://devtalk.nvidia.com/default/topic/1024458/nvidia-cuda-version-9-0-installer-failed/?offset=39