IBM has developed Summit, recognized as the planet's most powerful supercomputer, specifically engineered to handle the calculation-intensive models that drive today's artificial intelligence (AI) innovations. The Massachusetts Institute of Technology (MIT) is now set to benefit from this groundbreaking technology.
Earlier this year, IBM committed to donating a state-of-the-art $11.6 million computer system to MIT, designed with the same architecture as Summit—the supercomputer constructed for the U.S. Department of Energy at Oak Ridge National Laboratory. This advanced computing cluster is scheduled to become operational this autumn, coinciding with the inauguration of the MIT Stephen A. Schwarzman College of Computing. Researchers will leverage this enhanced capability to develop more sophisticated AI models addressing diverse challenges, from creating advanced hearing assistance devices to engineering more durable lithium-ion batteries.
"We're thrilled to witness numerous AI initiatives at MIT receive this substantial computational enhancement, and we eagerly anticipate the groundbreaking discoveries that will follow," remarks John E. Kelly III, IBM's executive vice president, who unveiled this contribution during MIT's launch celebration for the Schwarzman College of Computing in February.
IBM has designated this computing system as Satori, a term derived from Zen Buddhism representing "sudden enlightenment." While Satori occupies a physical space comparable to a shipping container, its intellectual capabilities more closely resemble a high-performance vehicle, executing an astonishing 2 quadrillion calculations every second. To put this in perspective, this computational capacity equals every person on Earth simultaneously solving over 10 million multiplication problems each second for an entire year. This remarkable performance positions Satori within the middle tier of the world's 500 fastest supercomputers.
The rapid advancement in AI technologies has generated an insatiable demand for computational resources necessary to train increasingly complex models on expanding datasets. Concurrently, federal funding for academic computing infrastructure has experienced a steady decline over three decades. According to Christopher Hill, who directs MIT's Research Computing Project, the current demand for computing power at MIT exceeds available resources by a factor of five.
"IBM's contribution arrives at an opportune moment," states Maria Zuber, a geophysics professor serving as MIT's vice president for research. "The establishment of the new college will undoubtedly intensify the demand for computational capabilities. Satori will significantly alleviate this pressing challenge."
The computing shortfall became immediately evident to John Cohn, chief scientist at the MIT-IBM Watson AI Lab, upon the lab's inception last year. "Cloud computing resources alone proved insufficient for our demanding AI training requirements," he explains. "The prohibitive costs and extended processing times prompted us to explore whether we could bring enhanced computational capabilities directly to MIT."
Satori was specifically engineered to address this need, featuring IBM Power9 processors, a high-speed internal network, extensive memory, and 256 graphics processing units (GPUs). Originally developed for rapid processing of video game graphics, GPUs have emerged as the primary workhorses for contemporary AI applications. Similar to Summit, Satori has been optimized to extract maximum performance from each GPU.
This donation continues IBM's longstanding tradition of collaboration with MIT, which has historically catalyzed numerous computing breakthroughs. In 1956, IBM facilitated the establishment of the MIT Computation Center through the donation of an IBM 704, the first mass-produced computer capable of handling complex mathematical operations. Nearly thirty years later, IBM provided funding for Project Athena, an initiative that introduced networked computing throughout the campus. These collaborative efforts yielded numerous technological innovations, including time-share operating systems, foundational programming languages, instant messaging, and the network security protocol Kerberos, among others.
More recently, IBM committed to investing $240 million over a decade to establish the MIT-IBM Watson AI Lab, a founding supporter of MIT's Quest for Intelligence initiative. Beyond addressing MIT's computing limitations, Satori will be configured to enable researchers to exchange data with all major commercial cloud providers while preparing their code for execution on IBM's Summit supercomputer.
Josh McDermott, an associate professor in MIT's Department of Brain and Cognitive Sciences, currently utilizes Summit to develop advanced hearing aid technology. However, prior to executing their models, he and his students invested countless hours preparing the code. McDermott notes that Satori will streamline this process in the future and enable more ambitious projects over the long term.
"We're currently constructing computer systems to model a single sensory system, but our goal is to develop models capable of seeing, hearing, and touching simultaneously," he explains. "This objective requires computational resources on a much grander scale."
Richard Braatz, the Edwin R. Gilliland Professor in MIT's Department of Chemical Engineering, employs AI to enhance lithium-ion battery technologies. He and his team recently created a machine learning algorithm to predict battery lifespan based on historical charging patterns. Currently, they're developing multiscale simulations to evaluate novel materials and designs for extending battery longevity. With the enhanced capabilities provided by a system like Satori, these simulations could capture critical physical and chemical processes, thereby accelerating discovery. "Improved predictions enable us to bring innovative concepts to market more rapidly," Braatz notes.
Satori will be installed at a converted silk mill now serving as a data center—the Massachusetts Green High Performance Computing Center (MGHPCC) in Holyoke, Massachusetts—and will connect to MIT via dedicated, high-speed fiber optic cables. Operating at 150 kilowatts, Satori will consume energy comparable to a mid-sized building at MIT, yet its carbon footprint will be almost entirely neutralized through the use of hydroelectric and nuclear power at the Holyoke facility. Featuring energy-efficient cooling, lighting, and power distribution systems, the MGHPCC became the first academic data center to receive LEED-platinum status—the highest green building certification—in 2011.
"Locating Satori at Holyoke minimizes its carbon emissions and environmental impact without diminishing its scientific contributions," explains John Goodhue, executive director of the MGHPCC.
For additional details, visit the Satori website.
