System speeds brain analysis

The faster technology is helping scientists to provide quicker results

The cluster lets scientists process 100 brain images in 16 minutes

A supercomputer that analyses images of the brain is providing scientists with more detailed data 24 times faster than previously possible.

Scientists can more quickly examine how the brain works since the launch of the UK’s largest high-performance compute cluster (HPCC) for neuroimaging data analysis at Cardiff University Brain Research Imaging Centre (Cubric).

Different types of scans are combined to generate a single view of the brain.

Scientists are using the computer cluster to simultaneously process 100 complete brain images in just 16 minutes.

This enables them to receive results, analyse findings, adjust research parameters and complete projects faster.

‘The more data you have, the greater the validity of the results as more volume increases the quality of analysis from a statistical point of view,’ said Spiro Stathakis, IT systems manager at Cubric.

‘We can look at images generated in near real time and identify any problems. Research will be a lot more accurate.’

Scientists visiting the facility are also reducing the number of repeat visits they need to make because the HPCC has increased productivity.

‘Scientists from different areas establish research partnerships and the HPCC makes forging alliances with us more appealing,’ said Stathakis.

High-performance computing integrator OCF installed the supercomputer and is on hand to assist Cubric with any problems that arise. Cubric opted to use Linux software to conduct the analysis.

‘Using proprietary software has cost implications and we are au fait with Linux,’ said Stathakis.The system consists of a 75-node cluster of IBM e326m dual processor, dual-core AMD Opteron servers, providing 300 processors. It also includes 40 IBM IntelliStation A Pro graphical workstations to view and manipulate the images. These can be standalone processing engines, or combined to give a total of 460 processing cores.

The Cubric cluster performs an estimated 530 billion operations per second.