CUDA

• tensorflow對應版本：

https://www.tensorflow.org/install/source_windows

• pytorch對應版本：

https://pytorch.org/get-started/previous-versions/

• GPU算力表：

https://developer.nvidia.com/cuda-gpus

• 顯卡驅動程式與 CUDA 的版本對應：

https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

• Nvidia驅動程式

https://www.nvidia.com/zh-tw/drivers/

• CUDA 官方的安裝流程

https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html

• CUDA 下載

https://developer.nvidia.com/cuda-toolkit-archive

• 測試 CUDA 是否安裝成功

1. nvcc -V

2. C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\extras\demo_suite>deviceQuery.exe

3. C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4\extras\demo_suite>bandwidthTest.exe

• 安裝 cuDNN，cuDNN 下載：

https://developer.nvidia.com/rdp/cudnn-archive

•  cuDNN 官方流程：

https://docs.nvidia.com/deeplearning/cudnn/latest/installation/windows.html

• 測試tensorflow-gpu是否成功

python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

• 若出現找不到dll錯誤，將下列檔案複製到C:\windows\system32：

cublas64_11.dll

cublasLt64_11.dll

cudart64_110.dll

cudnn64_8.dll

cufft64_10.dll

curand64_10.dll

cusolver64_11.dll

cusparse64_11.dll