v0.14.0-beta
Release Highlights
Python x86_64 wheels are now available on PyPi, cvcuda-cu11 and cvcuda-cu12 for CUDA 11 and CUDA 12, respectively
pip install cvcuda-cu<CUDA_VERSION>
CV-CUDA v0.14.0 includes the following changes:
New Features:
Added support for SBSA ARM/Grace cuda 12, including ManyLinux2014-compliant Python wheel generation
We do not provide SBSA-compatible aarch64_cu11 packages yet, this will be addressed in an upcoming release
aarch64_cu12 packages distributed on Github and Pypi are the SBSA-compatible ones. Jetson builds can be found in explicitly named “Jetson” archives in Github release assets.
Added support for compiling NVCV on QNX
Added support for VYUY and YUV8p formats in NVCV
Improved test coverage for NVCV and operators
Minor corrections to documentation
Bug Fixes:
Made Python cache thread-local to avoid race conditions and potential crashes in Python gilless multithreaded setups
Compatibility and Known Limitations
Starting with v0.14, aarch64_cu12 packages (deb, tar.xz or wheels) distributed on Github (release “assets”) and Pypi are the SBSA-compatible ones. Jetson builds (deb, tar.xz, whl) can be found in explicitly named “Jetson” archives in Github release assets.
We do not provide SBSA-compatible aarch64_cu11 packages yet, this will be addressed in an upcoming release.
Only x86_64 wheels are available on PyPi as of 02/28/25. SBSA/Grace CUDA 12 wheels will be added shortly.
For the full list, see main README on CV-CUDA GitHub.
License
CV-CUDA is licensed under the Apache 2.0 license.
Resources
Acknowledgements
CV-CUDA is developed jointly by NVIDIA and the ByteDance Machine Learning team.