Archive notice: This article was originally published on March 1, 2011. Links and embedded videos are preserved as part of the historical record.
NVIDIA presents the latest version of the CUDA toolkit, which enables software to access the computing power of graphics processors. Its three essential components are “NVIDIA GPUDirect™ 2.0 Technology”, which organises communication between CUDA cores within a single computer and thus fully exploits the system’s performance. There is also “Unified Virtual Addressing “(UVA), which creatse single access to RAM and graphics memory, thus significantly reducing programming effort. The third part is the “Thrust C++ Template Performance Primitives Libraries”, which provides a collection of ready-made algorithms for C++ programmers that are up to five times faster than the Standard Template Library (STL) and Threading Building Blocks (TBB).
Version four of the CUDA programming environment also adds MPI integration of CUDA applications, in which a modified MPI implementation automatically transports data to and from the GPU via InfiniBand if the application activates it. “Multi-GPU Sharing by Single CPU Thread” enables a single CPU to address all the GPUs in the system, allowing developers to access resources easily across multiple GPU platforms.
A first release candidate of CUDA Toolkit 4.0 will be available free of charge from 4.3. for participants in the “CUDA Registered Developer Program” at nvidia.com/paralleldeveloper. Incidentally, you can learn more about GPU processing in the focus section of the latest DIGITAL PRODUCTION issue 2/2011.
