NVIDIA CUDA 11.2 Released For Further Enhancing Its Proprietary Compute Stack
In addition to the NVIDIA 460 series Linux beta driver being released this week, CUDA 11.2 has also made its debut for Windows and Linux.
There are the usual assortment of small to medium sized updates with the CUDA 11.2 compute stack update. Some of the CUDA 11.2 highlights include:
- Support for importing Direct3D 11/12 textures.
- The external semaphore interface can now work with the Vulkan Timeline Semaphores to import, wait, and signal on Vulkan Timeline Semaphores.
- Performance improvements for multi-GPU systems with the cuFFT library.
- New cudaMallocAsync() and cudaFreeAsync() calls for the stream ordered memory allocator that can provide significant performance improvements over the non-Async versions.
- Support for Red Hat Enterprise Linux 7.9, Red Hat Enterprise Linux 8.3, Fedora 33, and Debian 10.6.
More details on the CUDA 11.2 changes can be found via the release notes. The proprietary, vendor-specific compute stack can be downloaded from developer.nvidia.com.
There are the usual assortment of small to medium sized updates with the CUDA 11.2 compute stack update. Some of the CUDA 11.2 highlights include:
- Support for importing Direct3D 11/12 textures.
- The external semaphore interface can now work with the Vulkan Timeline Semaphores to import, wait, and signal on Vulkan Timeline Semaphores.
- Performance improvements for multi-GPU systems with the cuFFT library.
- New cudaMallocAsync() and cudaFreeAsync() calls for the stream ordered memory allocator that can provide significant performance improvements over the non-Async versions.
- Support for Red Hat Enterprise Linux 7.9, Red Hat Enterprise Linux 8.3, Fedora 33, and Debian 10.6.
More details on the CUDA 11.2 changes can be found via the release notes. The proprietary, vendor-specific compute stack can be downloaded from developer.nvidia.com.
8 Comments