Steps for GPU installation :
- please make sure the version compatability. choose the relevan version https://www.tensorflow.org/install/source_windows, for my case I will install tensorflow-gpu under Windows OS
CUDA 11.2 , cuDNN 8.1, Python 3.8, tensorflow_gpu-2.10.0
- download and install download CUDA via the official website https://developer.nvidia.com/cuda-11.0-download-archive install CUDA and follow the instruction
download cuDNN via the official website https://developer.nvidia.com/cudnn create account first to get access to download page Download cuDNN v8.9.3 (July 11th, 2023), for CUDA 11.x choose the dowloaded file according your OS. Local Installer for Windows (Zip) unzip cudnn-windows-x86_64-8.9.3.28_cuda11-archive.zip, then copy and paste to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0
or you can follow this guideline https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-zlib-windows
-
Install anaconda dan create environment py38 Use anaconda to manage python library create environtment using conda , follow this instruction https://conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html conda create -n py38 python=3.8 pip install keras pip install tensorflow-gpu==2.10.0
-
Troubleshooting if you have an issue with "cusolver64_11.dll not found", then the solution
-
Move to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
-
Rename file cusolver64_10.dll To cusolver64_11.dll it works for me
if you have an issue with ZLIB or zlibwapi.dll Save zlibwapi.dll, follow this instruction https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#install-zlib-windows save as link ZLIB DLL unzip zlib123dllx64.zip. then copy all files in ..\zlib123dllx64\dll_x64 to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\bin
other python libraries (optional):
pip install matplotlib
pip install scikit-learn
pip install pandas
pip install seaborn
pip install opencv-python