MATLAB and more
NVIDIA's GPU Technology Conference (GTC) kicked-off this week, and the talk is all about GPGPU and high-performance computing. CEO Jen-Hsun Huang took to the stage on day one to deliver some interesting details on the direction that the company will be taking its CUDA technology over the coming years.
The first announcement - and arguably the biggest development of the presentation - was a partnership with PGI to develop an x86 compiler for CUDA. This will allow programs written for NVIDIA GPUs to run natively on CPUs, meaning that a company could deploy an application safe in the knowledge that any computer will be able to run it - whether it has a GPU installed or not.
The second partnership was with MATLAB developer MathWorks, who has been working with NVIDIA to enable CUDA support within the software's parallel-computing toolkit. According to Huang, calculations using the technology will see a ten to 40 times speed increase over using a CPU alone.
Similar announcements were also made regarding the Amber 11 computational-biology package and the upcoming release of ANSYS Mechanical r13, both of which will feature CUDA support. The former demonstrated particularly impressive improvements, with eight NVIDIA GPUs shown to outperform 192 quad-core CPUs running on the Kraken supercomputer.
