SAN JOSE, Calif. – MathWorks announced the first support for graphics processors in its parallel computing products, choosing Nvidia's processors and their CUDA version 1.3 software.
MathWorks picked the CUDA environment for its support for double precision math and IEEE-compliant operations. The company will consider supporting other general-purpose GPU environment such as OpenCL and Microsoft's DirectCompute when they adopt those features.
The announcement comes at Nvidia's second annual conference for developers using its graphics chips in high performance computing applications that range from oil and gas exploration to financial modeling. A wide variety of companies are announcing at the event new systems, boards, appliances or other products for general-purpose GPU computing using Nvidia components.
Graphics chips have slowly been gaining traction for a range of demanding vertical applications. At the Hot Chips conference in August one high-performance computing developer detailed advances using FPGAs, but said he also has promising graphics-based systems under test in the lab.
MathWorks' Parallel Computing Toolbox and Distributed Computing Server products now support CUDA 1.3-complaint Nvidia Tesla-10 and -20 chips. A full list of the graphics chips supported is available online.
The new software will deliver five- to fifty-fold performance benefits when run on GPUs compared to quad-core Intel processors. However, users will probably not see more than five-fold increases due to system bottlenecks in Ethernet PCI Express.
The company has had the products in beta testing since late last year. "We were surprised to find hundreds and hundreds of people signing up for the beta from every vertical market we support," said Silvina Grad-Freilich, senior marketing manager of parallel computing at MathWorks.
“Enabling Matlab users to accelerate their applications using GPUs provides the foundation for ground-breaking innovations across engineering and science applications,” said Sumit Gupta, senior manager of Tesla products at Nvidia, speaking in a press statement.