NVIDIA Officially Releases CUDA 8

While it has been available in early form for a few months, today NVIDIA announced the general availability of CUDA 8.

CUDA 8.0 comes with full support for the Pascal GPU architecture including the Tesla P100/P40/P4 accelerators, new unified memory capabilities, native FP16 and INT8 computation support, nvGRAPH, new profiling capabilities, improved compiler performance, and expanded developer platform support. The new nvGRAPH addition is a GPU-accelerated graph analytics library. The expanded platform coverage now includes Ubuntu 16.04 LTS with GCC 5.4.

With the compilation improvements in CUDA 8, NVIDIA claims the compilation time can be 2x or faster for a range of codes, particularly those using C++ templates.

CUDA developers wishing to learn more about CUDA 8.0 can visit this NVIDIA blog post for an extensive look at the new features.