Phone chips point to 4 teraflops for $10k

Beverley Head
Tuesday, 15 March 2011 16:44

Business IT - Networking

Commodity graphics processor chips used in games machines and mobile phones are revolutionising the face of supercomputing - and should lead to the delivery of 4 teraflop desktop machines running Windows HPC or Linux for just north of $10,000 by 2012.

Once the province of boffins in universities or research laboratories it is now economically feasible for commercial entities to install desktop supercomputers delivering processing speeds of 1 teraflop. These systems cost $10,000-15,000 and consume 1100 Kw of power.

According to Dragan Dimitrovici, managing director of Xenon, which delivers supercomputers based on the NVidia Tesla GPU platform in Australia, it is this new power and economy that is driving organisations such as Rio Tinto and Chevron to use high performance computing clusters to perform seismic data analysis in Australia.

At a briefing in Sydney, Andy Keane, the general manager of NVidia's Tesla computing group said that in 2011 the company's Kepler GPU would be released underpinning the next generation machines offering 4-6 teraflops for the same cost as the current Tesla entry level systems. Beyond that NVidia has roadmapped the even faster Maxwell series which is scheduled for delivery in 2013.

Demand for large scale computer systems locally is on the rise according to Mr Dimitrovici.  While the CSIRO was the first organisation to take on a Xenon developed high performance computing cluster based on the Tesla GPU in 2009, he said that the company expected to roll out another five this year, taking Australia's clustered Tesla GPU count to more than 1000.

This sort of technology is ultimately expected to provide the computational underpinnings for projects such as climate modelling, virus modelling and genomics.

Dr Luke Domanski, a high performance computing consultant in CSIRO's recently formed advanced scientific computing group, said that the system installed at the national research agency was gradually being taken up by scientists in areas such as biogeochemical modelling, computational fluid dynamics and medical imaging.