NVIDIA nvImageCodec CUDA 12

0.8.0.22 verified Fri May 01 auth: no python

NVIDIA nvImageCodec provides accelerated image encoding and decoding using GPU hardware on CUDA 12 systems. Version 0.8.0.22 is the latest beta release, with a focus on JPEG, JPEG2000, and other formats. The library is under active development with beta releases every few months.

pip install nvidia-nvimgcodec-cu12

Common errors

error ModuleNotFoundError: No module named 'nvimgcodec' ↓

cause Installed the wrong CUDA variant or package not installed.

fix

Ensure you installed 'nvidia-nvimgcodec-cu12' for CUDA 12.

error RuntimeError: CUDA error: no kernel image is available for execution on the device ↓

cause GPU compute capability is too old for the installed version.

fix

Check GPU compatibility (CC >= 5.0 required for CUDA 12). Use a compatible GPU or downgrade to nvImageCodec for CUDA 11.

error AttributeError: module 'nvimgcodec' has no attribute 'Decoder' ↓

cause Using old API before v0.6.0 where classes had different names.

fix

Update to v0.6.0+ and use 'nvimgcodec.Decoder'.

Warnings

deprecated The package is beta (v0.8.0.22). API may change without notice in future releases. ↓

fix Pin version and test upgrades.

gotcha Only CUDA 12 is supported. Do not install on CUDA 11 systems or older. ↓

fix Use the appropriate CUDA variant package (e.g., nvidia-nvimgcodec-cu11 for CUDA 11).

breaking The library name changed from 'nvImageCodecs' to 'nvimgcodec' in v0.6.0. Import paths changed accordingly. ↓

fix Use 'import nvimgcodec' instead of old 'import nvImageCodecs'.

Imports

nvimgcodec
```
import nvimgcodec
```

Quickstart

Load and decode an image using nvImageCodec.

import nvimgcodec
import numpy as np

# Create encoder/decoder instance
decoder = nvimgcodec.Decoder()

# Open an image file
img = decoder.read('example.jpg')
print(f"Image shape: {img.shape}")