CUDA 4.0 が 3/4 にリリースされます (CUDA 4.0 will be released at 3/4 )

NVidia から送られてくる "CUDA: Week in Review"というタイトルのメールで, CUDA 4.0 が今週の金曜日にリリースされると書いてありました.

In the online news summary titled "CUDA: Week in Review" which is an E-Mail delivered from NVidia, it describes that CUDA 4.0 will be released on Friday this week.

そのメールには, CUDA 4.0 にはより多くの開発者にとって並列プログラミングがしやすくなるような 3 つの新しい重要な特徴があると書いてありました.

The letter says that there are three critical new features at CUDA 4.0, all of which make parallel programming easier and accessible to more developers.

  • (a) NVIDIA GPUDirect 2.0, which enables peer-to-peer memory access and thus faster multi-GPU programming;
  • (b) Unified Virtual Addressing, which provides a single flat memory address space for CPU and GPU resources, enabling quicker and easier parallel programming;
  • (c) More C++ capabilities and an easier way to program parallel apps in C++ with the addition of the Thrust library of template performance primitives.

I guess (a) will be useful for narrow developers who uses multiple GPU for HPC currently.

I think (b) is good because it will be easy for programmers to manage the memory resource between CPU and GPU.

And (c) will be choice for developers to use it instead of writing template containters, algorithms and iterators from scratch.