Tuesday, March 1, 2011

CUDA 4.0 celebrates the parallelism

On this first day of CeBIT, NIVIDA announced the release of a new version of its CUDA development tools optimizing parallel applications. The firm has tried to make it easier for developers to take advantage of highly parallel architecture of GPUs. The avowed aim is to encourage more publishers to go out versions of GPGPU applications.

To achieve his ends, NVIDIA initially worked on the communication between graphics processors to optimize applications running on servers or workstations by combining several. He then designed a virtual unified addressing system that gathers system memory and GPU in a single address space.

Finally, the Toolkit provides algorithms and models Thrust C + + supposed to speed up development time and performance software. CUDA 4.0 The architecture also provides management of multiple GPUs in a single CPU thread and vice versa, that is to say that multiple threads can share a single CPU GPU.

Finally, the library cuda-gdb now supports Mac OS X. The CUDA Toolkit 4.0 is available for free download from the March 4, 2011 in Release Candidate. The final edition is expected to arrive within two months.

No comments:

Post a Comment