Tuesday, 21 September 2010 18:30

Nvidia powers MATLAB and Amber GPU acceleration

Written by Jon Worrel

GTC 2010:
Massive advances in computational biology
During the Nvidia GTC 2010 Keynote, CEO Jen-Hsun Huang announced that MATLAB and Amber (PMEMD) are capable of receiving CUDA acceleration using the power of Fermi-based GPUs.

“MATLAB is used by millions, as a general purpose numerical computation package,” said Huang. “It has stats tool kits, image-processing toolkits, data-acquisition toolkits, informatic toolkits. MATLAB will support CUDA-accelerated GPUs with the parallel computing toolkit. 

In the field of computational biology, some of the simulations are enormously complex. Many computational simulations include walking and gene sequencing. Researchers have begun to simulate molecules at the atomic level using quantum chemistry. Amber (PMEMD) is among the most popular molecular simulation packages available.

“Amber is one of the leading nanomolecular dynamic simulators.” said Huang The team at University of California, San Diego recently took the Amber simulator and made it Multi-GPU capable. The speed-up is really, really impressive.”

He showed off the JAC Benchmark, where 192 quad-core CPUs can simulate 46ns/day. In contrast, just 8 Fermi-based GPUs can run the same simulation at 52n/s per day.

“But graphs don’t tell the whole picture. The KRAKEN is a massive supercomputer. With an 8-GPU cluster running Amber for GPUs they’re able to achieve the same level of performance with something that is just a fraction of its size.”

GPU-accelerated Amber (PMEMD) has been implemented using CUDA and can only run on Nvidia GPUs at present. Due to accuracy concerns in the scientific field with single-precision computations, the code makes use of double-precision computation and places high hardware requirements on GPUs that can compute double-precision with high efficiency. As a result, Nvidia is emphasizing its Tesla S2050/S2060/C2050/C2070 and GTX 470/480 graphics cards as the perfect candidates for this computationally intense simulation software.

In perspective, Jen-Hsun is highlighting the computational power differences between CUDA-accelerated GPU versus massive clusters of CPUs. “Now GPU computing is instantaneously available to millions of users around the world doing in work in high productivity, analysis and development.”

Last modified on Wednesday, 22 September 2010 18:03
blog comments powered by Disqus


Facebook activity

Latest Commented Articles

Recent Comments