The University of Massachusetts Dartmouth’s Center for Scientific Computing & Visualization Research (CSCVR) has contributed 120 GPU servers to the UMass shared computational cluster installed at the MGHPCC.
The UMass shared cluster at the MGHPCC is a large (~15,000) processor-core supercomputer system that serves the entire UMass system’s computational scientists. From astrophysicists at the Amherst and Dartmouth campuses, to biomedical researchers at Worcester, this key central facility provides the computational resources needed to advance world-class research programs across the system. While a large number of processors is the most common way to build powerful supercomputers today; lately advances in graphics processors (GPUs) has allowed them to be used as “accelerators” for scientific calculations often delivering a speed-up of over an order-of-magnitude when used in conjunction with standard CPUs in specialized servers. Until recently, the UMass shared cluster lacked such GPU-based servers in quantity, but recently the UMass Dartmouth CSCVR has significantly enhanced the UMass shared facility by contributing part of a GPU cluster that was donated to it by a bitcoin mining investor.
The UMass Dartmouth CSCVR provides undergraduate and graduate students with educational experiences in supercomputing and fosters collaborative research in the computational sciences within the university and with researchers at other universities, national labs, and industry. Faculty at the CSCVR have long been involved in pioneering efforts to use GPUs in scientific computing. In particular, Professor Gaurav Khanna, who serves as associate director of the Center, pioneered the use of PlayStations for computational black hole astrophysics.
In December 2014, the New York Times’ Laura Parker covered Prof. Gaurav Khanna’s novel use of PlayStations for his computational research in black hole astrophysics in a full length article. A bitcoin mining investor read the story and decided to donate a large supercomputer to the university campus for further enabling research productivity in the CSCVR. The donated system was built in 2012 at cost nearing a million dollars and consists of 180 GPU servers installed in 10 racks, integrated tightly over a fast network. Its replacement cost today could be in ballpark of $400,000.
The donor, Daniel Driscoll of San Francisco shares his thoughts on his donation: “This past winter during my morning commute, I stumbled across an article about a research project that recently used a networked super computer modeling the physics of black holes. Having grown up in the 80’s next to the shuttle launch pad in Cape Canaveral, Florida, I remain an ardent fan of all things space. What struck me as even more interesting was how Professor Gaurav Khanna had created this super computer from now derelict PlayStation 3 consoles. Being now part of the game industry, I was all too familiar with the transition of PlayStation fans to the new flagship, the PlayStation 4, and had myself done a few experiments to understand the potential uses for the no-doubt countless number of old PlayStation 3’s likely going to the trash. One such experiment involved a private venture to explore and understand the Bitcoin craze of 2013, a venture that ultimately proved fruitless but left me in possession of not just several PlayStations, but all also a large collection of what I had deemed a more effective array of server computers. Now that bitcoin has come and gone for the hobbyist farmer, Professor Khanna and the University of Massachusetts Dartmouth has provided me an opportunity to join those explorers I once grew up with, if only in my own small way.”
The UMassD CSCVR is glad to share part of this GPU cluster with other UMass researchers to expand the impact of Mr. Driscoll’s donation. Prof. Sigal Gottlieb, the Director of the CSCVR remarks on how its donation would impact the research productivity at the MGHPCC: “These computational resources will be a significant addition to the resources of the UMass System. Through this shared resource all UMass computational researchers, and their students, will benefit from this valuable computational tool as we continue to expand the scope and range of the computational problems we are attacking. We are hopeful that sharing our GPU cluster with the wider UMass community will inspire and encourage other groups in all the UMass campuses to share their research clusters as well, thus enabling more computational research across the system”
View a simulation of the effects of a black hole: