NVIDIA today has announced that the release candidate of NVIDIA CUDA Toolkit 4.0 will be making its way to developers this month. According to the company, their latest edition of CUDA coder was designed to assist developers in getting the best out of GPU computing through three of its main features: NVIDIA GPUDirect 2.0 Technology, single merged-memory address space for both CPU and GPU through unified virtual addressing (UVA) and Thrust C++ template primitives library.
How these features are going to help developers tap the power out of GPUs for the benefit of their applications? NVIDIA stated that since GPUDirect 2.0 offers direct GPU-to-GPU communication compared to previous method that requires data from one GPU memory to be transferred to the system’s memory before it can be transported to the other GPU, it goes together with UVA for faster multi-GPU programming. The toolkit also focused on making parallel programming easier for developers with the help of Thrust's open source C++ library.
The NVIDIA CUDA Tookit 4.0 will be available free of charge to developers under CUDA Registered Developer Program starting from 4th March. You can further learn about the new coder at www.nvidia.com/cuda.
Article From: Lowyat
No comments:
Post a Comment