The University of Vermont’s supercomputer is about to get faster. A lot faster.
The National Science Foundation has awarded the university a $1 million grant to significantly upgrade its Vermont Advanced Computing Core (VACC).
Over the next few months, 72 high performance graphics processing units, or GPUs, will be added to the current VACC, housed in UVM’s primary datacenter in South Burlington, to create a new high performance cluster.
Working together in a “massively parallel” system, the new cluster – dubbed DeepGreen – will be up to 3,000 times faster than the current VACC. At its peak, DeepGreen will be able to achieve a speed of over 1 petaflop, or one thousand million million computations per second, the equivalent of 20,000 laptop computers working in tandem.
Research advances today are increasingly dependent on finding meaning and patterns within massive amounts of information, or big data, Del Maestro said.
“The new processing power will allow us to take all that data and find the things that are important in it – the needles in the haystack,” Del Maestro said.
“We’re in a new scientific era that’s mainly about two things: data and computational power,” said Safwan Wshah, an assistant professor in UVM’s Computer Science department who studies computer vision, a research area that will benefit greatly from the VACC’s enhanced capability.
The upgraded VACC is a “major leap and a necessary leap,” he said.
The increase in processing speed, combined with a “big pipe,” a hyper-fast connection the university will install from campus to the VACC, will enable faculty to take on new types of research projects they did not have the computational power to explore in the past, Del Maestro said.
He cited three examples. “Josh Bongard in Computer Science will use DeepGreen to analyze a gigantic crowd sourced data set to produce safer human-robot interactions,” he said. “Hugh Garavan in Psychiatry will use the new machine learning cores on the cluster to determine the impact of substance use on developing adolescent brains using brain imaging. And Yolanda Chen in Plant and Soil Science can massively speed up the genome re-sequencing of the Colorado Potato Beetle to better combat emerging threats to our food supply in a changing climate.”
A competitive process
UVM won the competitive grant – NSF awards between 10 and 15 like it every two years – because the grant application demonstrated that faculty use of the upgraded VACC would be widespread and multi-disciplinary and showed that the center could be a regional resource for schools like Middlebury College and Norwich University.
NSF was also swayed by the UVM administration’s support of the project.
“UVM faculty scholars across the university, in all of our colleges and schools, are engaged in research and discovery that today requires extremely powerful computational tools,” said David V. Rosowsky, UVM provost and senior vice president. “Immense data sets, computationally-intensive modeling and analysis, predictive analytics, and advanced visualization all require our faculty and students to have access to state-of-the-art, high-performance computing. The Office of the Provost was very happy to support this important initiative for our campus and our state.”
The new graphics processing units in the VACC upgrade will augment more conventional CPU’s, or central processing units, in the current cluster.
Known by computer gamers for their ability to quickly render data-heavy graphics, GPU’s are also “excellent for AI and machine learning,” whose algorithms need high performance computing to function and “learn,” Del Maestro said.
The upgraded VACC will also be a great asset in undergraduate education, said Wshah, who teaches machine learning and deep learning to about 80 students a year.
“The upgrade will enable them to take on more and bigger projects,” he said. “They are very excited to be entering this new age of discovery.”
The DeepGreen cluster at the VACC will be in place by early 2019. Training sessions, geared to faculty with varying levels of experience with high performance computing, will be held in the spring.
UVM’s Enterprise Technology Services group will be doing the design, construction and support for the new cluster.
Source: UVM News