The Nvidia DGX-1 was announced in April this year at the company's GPU Technology Conference. The system is designed for deep learning applications.
The first units shipped in the Asia Pacific region have gone to CSIRO.
CSIRO deputy chief information officer Angus Macoustra said there is interest and expertise in artificial intelligence across the organisation, especially within Data61.
In that last category, it is being used to re-analyse the data collected over the last 40 years by the Parkes radiotelescope in the hope of detecting more pulsars.
Deep learning systems are particularly useful for identifying patterns within very large data sets.
"Offering unprecedented performance in a compact package, the DGX-1 gives CSIRO researchers the power and capabilities needed to solve complex problems – and we can't wait to see the scientific discoveries that result," said Nvidia APAC vice-president of sales and marketing Raymond Teh.
Moving existing applications to a GPU-based platform "does require a level of expertise," Macoustra said, so every six months groups within CSIRO are invited to submit proposals to collaboratively adapt their existing software to take advantage of GPU performance.
This involves a mix of porting, optimising and re-architecting.
CSIRO has historically operated on a four-year hardware refresh cycle for such systems. "Obsolescence is a challenge," Macoustra said, as it makes it harder for CSIRO to compete with other research organisations.
The DGX-1 came at the right time for CSIRO as there were growing demands from various research groups. Macoustra said the organisation wanted a readymade platform rather than having to configure its own systems, and the DGX-1s provide it with early access to Nvidia's Pascal platform ahead of the impending replacement of large-scale clusters.