NIU scientists have acquired a new high-performance computer cluster that will exponentially ramp up on-campus capabilities to sort and analyze large quantities of research data.
The hybrid GPU/CPU supercomputer has a capacity of more than 30 teraflops, meaning it can do more than 30 trillion calculations per second. It is expected to usher in a new era of high performance computing at NIU, opening up opportunities for faculty researchers in a wide of array disciplines and for students interested in supercomputing.
“This provides a quantum leap in computing power for NIU,” said Department of Computer Science Chair Nicholas Karonis, who led a team of scientists in acquiring the cluster. “I think it’s going to be a game-changer for our university.”
In terms of speed, memory and storage capacity, the computer cluster would likely rank among the top 600 supercomputers worldwide, Karonis said. It was powered up this month and is currently undergoing testing.
“The term supercomputer is loosely defined,” Karonis said. “Our cluster is not of the magnitude of supercomputers used at federal labs such as Argonne National Laboratory, but it will achieve high-performance computing.
“The NIU cluster also will open new doors for students interested in supercomputing and for faculty researchers who need to crunch very large amounts of data.”
The supercomputer initiative grew out of a medical-imaging project led by John Lewis of the Northern Illinois Proton Treatment and Research Center. That project aims to advance Proton Computed Tomography, or pCT, which uses proton beams to produce medical images for proton therapy, a precise form of radiation treatment for cancer.
NIU physicists are designing and building a detector for the project, while NIU computer scientists are developing software needed to convert the enormous amounts of data collected by the detector into medical images.
“As we were working to develop pCT capability, we realized one of the problems was the time it takes to reconstruct an image,” Lewis said. “This computer system will speed up image reproduction, from four or five hours to minutes. Our hope is to have all of the equipment for this project ready for testing in about eight months.”
The cost of the computer cluster was about $843,000, with most of the funding coming from the proton-imaging project. Physics Professor Emeritus Clyde Kimball also contributed a portion of the funding he received for nanotechnology-related research that also requires high-performance computing.
By combining resources, “we went from a computer that would have been revolutionary for NIU to something even bigger, better and more powerful,” Kimball said.
“It was really Clyde as much as anyone else who came up with this vision of building a computing resource for the entire campus,” Karonis added. He and Kimball have dubbed the cluster “Gaea” (pronounced GUY-uh), for the mythological Greek goddess who was the mother of all.
“We wanted something emblematic of origins or beginnings,” Karonis said.
The supercomputing initiative also included assistance from Argonne National Laboratory scientists and NIU Information Technology Services, the College of Liberal Arts and Sciences and the Division of Research and Graduate Studies.
“This is a wonderful tool that will enable faculty to answer problems that seem intractable now,” said NIU Vice President for Research Lisa Freeman, who helped facilitate collaboration across campus on the project. “We also expect this incredible resource to help NIU attract new research faculty and students.”
While the Proton Computed Tomography project will have priority use of the computer, Freeman said she foresees the tool as being an invaluable resource for faculty researchers in areas ranging from physics, chemistry and biology to geography, economics, engineering and finance.
“Because computing has applications in many fields, this also is going to give students a much broader view of science and technology and better prepare them for work at universities and in industry,” Freeman said.
Users will be able to access the high-performance computer remotely. The cluster consists of 60 nodes, each of which could be likened to a high-end desktop computer, connected by a specialized high-speed network. It also boasts large disk-storage capacity – about 300 times more than would be found in an average computer.
The cluster’s use of both Central Processing Units (CPUs) and Graphics Processing Units (GPUs) is an important feature. Working in tandem, they pack a powerful computational punch.
Popularly associated with video gaming, GPUs are mass produced and relatively inexpensive. Within the last decade, computer scientists have begun tapping into the remarkable power of GPUs for imaging and for many types of mathematical computation. Hybrid CPU/GPU supercomputers provide a relatively low-cost means to achieving high performance.
The use of the hybrid technology comes with a learning curve, however, for both faculty and students. Karonis envisions a small team of students eventually operating the supercomputing equipment and assisting campus researchers in harnessing the cluster’s power. Already a number of graduate students are beginning to learn the programming.
“This is going to have an impact on our curriculum in computer science, no doubt about it,” Karonis said. “The type of computing performed on this cluster requires knowledge in such areas as parallel and distributed processing, message passaging and GPU programming. These are all new things that are going to hit the computer science department.”