The Stampede2 project procured, deployed, and operated a petascale supercomputer for science and engineering at the Texas Advanced Computing Center (TACC), part of the University of Texas at Austin. Deployed in 2017, Stampede 2 is still operating today, and will continue to do so through June of 2023, when it will retire after six years of production operations. To date, the machine has run more than nine million simulation, AI, and data analysis jobs for more than fifty thousand users. These jobs took more than 10 billion hours of processor time.
Stampede2 consists of more than six thousand compute servers made by Dell Technologies, with processors from Intel and storage from the Cray division of Hewlett-Packard Enterprise. The network consisted of a 100Gb high-speed Omnipath fabric, originally provided by Intel, and now supported by Cornelis Networks. The compute servers included 4,204 Intel Xeon Phi "Knights Landing" processors, and 1,736 servers with dual Intel Xeon Scalable "Sky Lake" processors, for a total of more than 400,000 processor cores, and nearly 20 petaflops of peak performance.
Since the on-time, on-budget start of production operations in mid-2017, time on Stampede 2 has been allocated quarterly via the XSEDE peer-reveiw process. More than 3,000 projects received allocations on the machine, across virtually all fields of science. Typically, requests for the machine were three to four times the available time. Throughout all four years of operation of the machine, usage has been well over 90% of full capacity, limited only by the complications of scheduling very large simulations of varying size. In the most recent year of allocations, highlights included multiple projects that improve the endurance of batteries, or reduce the need for scarce materials in constructing batteries. Other users identified target materials that would allow solar panels to be constructed without the use of lead. Catalogs of human RNA were produced for the first time through the use of computation on Stampede 2. Stampede 2 also provided computation for drug discovery and epidemiological models in response to the COVID-19 pandemic. Other users made use of Stampede 2 to forecast and respond to natural disasters, including wildfires, hurricanes, and tornadoes. Research in AI has been growing steadily on the machine.
Operations also included signficant training and outreach components. There have been hundreds of thousands of visits to the online "Virtual Workshop" for Stampede 2 hosted by project partner Cornell University. Thousands of people have attended online or in-person training, ranging from week long courses and all day conference workshops to shorter webinars. Hundreds of students have participated in outreach events, including the summer "Code@TACC" week-long residential events for high school. Students participating in these events were overwhelmingly first generation college students, and tracking data shows attendees enter college in STEM fields at more than 10x the national average.
Stampede 2 has been reliable, accessible, and overwhelmingly impactful to countless NSF researchers.
Last Modified: 10/13/2021
Modified by: Daniel Stanzione