2.1 KiB
2.1 KiB
CUDA 11.8 and cuDNN 8.x Installation Guide for PaddleOCR GPU Support
Current Issue
PaddlePaddle 2.6.0 is compiled for CUDA 11.8, but the system has CUDA 12.9 installed, causing cuDNN version detection failures.
Installation Steps
1. Download CUDA 11.8
Download Link: https://developer.nvidia.com/cuda-11-8-0-download-archive
Windows Installation:
- Select: Windows → x86_64 → 10/11 → exe (local)
- File:
cuda_11.8.0_522.06_windows.exe
2. Download cuDNN 8.x for CUDA 11.8
Download Link: https://developer.nvidia.com/rdp/cudnn-archive
Requirements:
- cuDNN v8.9.7 for CUDA 11.x (or latest 8.x version)
- NVIDIA Developer Account required (free)
3. Installation Steps
Install CUDA 11.8:
- Run the CUDA 11.8 installer
- Choose "Custom" installation
- Keep default components selected
- Install to default location:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
Install cuDNN:
- Extract the cuDNN zip file
- Copy these files to CUDA 11.8 directory:
bin\cudnn64_8.dll→C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin\include\cudnn*.h→C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\include\lib\x64\cudnn*.lib→C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\lib\x64\
4. Update Environment Variables
System Environment Variables:
CUDA_PATH:C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8CUDA_HOME:C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
Update PATH:
- Add:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin - Add:
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\libnvvp
5. Verification
After installation, run:
python check_cuda_cudnn.py
Alternative: Use CPU-Only Mode (Temporary)
If CUDA 11.8 installation is not possible immediately, use CPU mode:
set CUDA_VISIBLE_DEVICES=-1
set PADDLE_OCR_USE_GPU=False
Notes
- Multiple CUDA versions can coexist on the same system
- Environment variables determine which CUDA version is used
- Restart system after CUDA/cuDNN installation
- PaddlePaddle 2.6.0 requires CUDA 11.8 specifically