Today Nvidia announced that CUDA Toolkit 4.0 Release Candidate will be available to registered developers starting March 4th. Designed to help increase developer adoption of CUDA, the new release centers around three new features:
- Faster Multi-GPU Programming with GPUDirect 2.0: New facility provides developers with peer-to-peer memory access, transfers & synchronization. MPI implementations natively support GPU data transfers, which entails less coding and higher programmer productivity.
- Rapid Application Porting with Unified Virtual Addressing: UVA provides a single address space comprising system memory and GPU memory.
- Easier Parallel Programming in C++: Thrust is a new C++ Templatized Algorithms & Data Structure set similar to C++ Standard Template Library (STL) that automatically chooses the fastest code path at compile time.
In other CUDA news this week, Fixstars released it’s Eclipse CUDA plugin, the first integrated development environment for CUDA on Linux. Using Fixstars’ CUDA plugin for Eclipse, developers can write, build, and debug *1 CUDA C/C++ applications using a GUI on their favorite CUDA-supported Linux distribution. Fixstars previously included the plugin as part of it’s Yellow Dog Linux for CUDA, but with this release the plugin will be available for many popular Linux distributions including Ubuntu, Fedora, and Red Hat Enterprise Linux.