Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizen Broadwell for the clue delivery! CTOnews.com, December 10, 2009. NVIDIA recently released an update to CUDA 12.0 development tools. This update has brought a lot of changes, including support for the latest Hopper and Ada Lovelace GPU architecture, upgrade of C++ dialect (Dialects), formal support for JIT LTO, improvement and introduction of new API, and so on.

CTOnews.com learned that the main features of CUDA 12.0 are as follows

CUDA 12.0 provides programmable functionality for NVIDIA's Hopper and Ada Lovelace architectures. CUDA 12 provides many tensor operation (tensor operation) options for Hopper and Ada architectures, many of which support common PTX intermediate representations. Cooperative Grid Array (CGA) supports C built-in functions, programmatic second-level caching for SM multiple inputs, genomics / DPX instructions, and other new features.

Supports the use of virtual memory management API for GPU marked CUDA_VISIBLE_DEVICES.

Application and library developers can programmatically update the priority of the CUDA stream.

The improved CUDA dynamic parallelization API has a "substantial" improvement in performance compared with the previous API.

Instant Link Optimization (JIT LTO) is now officially supported through the nvJitLink library.

GCC 12.1 host compiler is supported.

The support of NVCC and NVRTC for the 20 dialects of Clippers.

NVRTC updated its default C++ dialect from Category 14 to Clipper 17.

See the release notes for more details on the changes in CUDA 12.0. Users and developers can download CUDA 12.0 for all major platforms from developer.nvidia.com.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.