Titan will be a Cray machine with 18,000 Nvidia GPUs expected to make it the fastest system on the planet with peak performance of 20 petaflops. CSIRO's system is more modest.

According to Dragan Dimitrovici, managing director of Xenon, CSIRO's latest system will have 134 nodes with three GPU cards per node.

The CSIRO deal, which came in at around the $1 million mark, was formally announced today and saw Xenon beating off competition from Dell, SGI and HP. The CSIRO's Xenon cluster will be powered by 268 Intel Xeon E5-2650 processors (2,144 CPU cores) and feature 390 Nvidia Tesla GPU cards (174,720 GPU cores).

The next generation of Nvidia GPUs will be the Kepler series slated to deliver 4-6 teraflops for the same price as the current Tesla GPUs in the second half of this year. (Titan will use the Kepler generation of GPUs).

The following generation of even faster Nvidia chips is the Maxwell series - and the Xenon systems selected will be able to upgrade to these future chips.

The CSIRO was the first local organisation to take on a Xenon developed high performance computing cluster based on the Tesla GPU back in 2009 and went out to tender for a new system at the end of last year.

According to CSIRO's head of computational and simulation sciences, John Taylor, who leads  CSIRO's research team developing new methodologies for analysing and exploiting large and complex data sets and data streams, Xenon has built up something of head start in this area, having been one of the first vendors to market with GPU based clusters.

---

Dr Taylor said that the current system upgrade began before Easter and was expected to be complete by the end of April. He anticipated a 50 per cent boost in terms of performance, to around 75 teraflops, and forecast that could increase further to 120-130 teraflops.

The Xenon system can also be upgraded once the Nvidia Kepler chips are available which he predicted could deliver a further two or three fold boost in performance.

In terms of the applications for the new platform Dr Taylor said it would be a broad range across the CSIRO including materials science, bioscience, environmental science and medical imaging. In a media release issued by Xenon Dr Taylor said; 'We're in the business of turning information into knowledge and quantifying uncertainty to help people make better decisions, whether that's in bio-chemical marine models, climate modelling or for medical imaging.

'The faster we can process data, the faster we can get solutions to market - and that's in Australia's best interests.'

Dr Taylor said one of the features which was attractive was the energy efficiency of the Xenon system. 'We want to try to get to exascale computing where we have a 1,000 fold increase in speed - but we can't do that if there is a 1,000 fold increase in power.'

Xenon's approach decouples processor speed from power consumption