New cluster gives Purdue researchers easy access to a large GPU-based system
Born to sling intricate graphics onscreen and raised by computer game makers, graphics processing units, GPUs for short, have grown up to become all the rage in high-performance computing, for science and engineering research and many other purposes.
“They're the future,” says Charles Bouman, Purdue's Showalter Professor of Electrical and Computer Engineering and Biomedical Engineering.
That future has arrived at Purdue with the first large-scale, purpose-built GPU supercomputing cluster offered for use by faculty and their labs through ITaP Research Computing.
A dedicated portion of the new Brown cluster, the Brown-GPU cluster offers the same advantages of the other clusters in Purdue's Community Cluster Program. ITaP builds, operates and maintains the clusters, including security and expert support. All researchers have to do is their research, which Bouman, for one, says he prefers over the burden of running his own research computing system.
The community clusters also offer faculty and their labs access to much larger systems than they could afford individually.
For information on how to access the Brown-GPU cluster, see ITaP Research Computing’s cluster orders website. To make the GPUs more affordable, ITaP is offering users a new subscription-based purchase option.
For $2,400 per year, researchers can buy into a shared queue available only to those who have purchased access. Faculty can add their students to their queues the same way as on the other community clusters. Those who want a dedicated queue still have the option to buy nodes, with a five-year term of service.
The Brown-GPU cluster dovetails with Purdue's new data science initiative and it already is set for an expansion to facilitate machine learning research, thanks in part to an investment from the Office of the Executive Vice President for Research and Partnerships, says Preston Smith, director of research services and support for ITaP.
With an architecture that’s naturally parallel and boasting extremely powerful hardware, GPUs have become essential to solving the large image reconstruction problems and developing the deep learning algorithms that are Bouman’s research focus.
“It's so much faster that it would have been previously, for all intents and purposes, impossible,” Bouman says. “We're both increasing the speed and increasing the quality of the results.”
An issue with GPUs has been the difficulty in coding programs to run on them, but improvements in standard software libraries are easing that challenge, Bouman says.
Bouman’s lab focuses on new and improved ways to create better images faster from data captured by instruments ranging from CT, or computed tomography, scanners to synchrotron X-rays. Constructing those images — in four dimensions, three of space and one of time — requires processing huge amounts of data generated by the instruments, hence the need for high-performance computing.
Among other things, GPUs also are relevant to a wide range of sensing problems where inferences may be drawn from indirect, sparse and noisy sensor data, such as non-destructive evaluation of materials and navigation by autonomous vehicles.
To learn more about the Brown-GPU cluster and Purdue’s Community Cluster Program, as well as research data storage and other resources available from ITaP Research Computing, contact Preston Smith, email@example.com or 49-49729.