NVIDIA Pushes CUDA 7 RC With C++11 Features, Runtime Compilation

NVIDIA announced the release today of the CUDA 7.0 Release Candidate that brings new features to their popular but proprietary parallel programming toolkit.

CUDA 7.0 adds support for C++11 functionality, integrates new CUDA libraries, greater performance out of the Thrust library with the new Thrust 1.8 version, cuFFT performance improvements, support for GPU core dumps, new CUDA Memcheck tools, and various compiler improvements.

CUDA 7 is NVIDIA's first release that adds C++11 feature support to the nvcc compiler for supporting the new language features in the host and device code.

CUDA 7 introduces the new cuSolver library for dense and sparse direct linear solvers and a new Runtime Compilation library. The nvrtc library allows for CUDA C++ device code to be compiled at run-time and for that resulting PTX code to be run immediately on the GPU. This library and new API can be used for run-time code generation and dynamically modifying CUDA kernel code with lower overhead than calling the nvcc compiler from the program. With CUDA 7, the nvrtc library is considered a preview feature.

More details on the CUDA 7 features can be found via the announcement of the CUDA 7 Release Candidate on NVIDIA's developer blog.