Stampede2, a $30 million number-crunching machine, is now up and running at the Texas Advanced Computing Center (TACC) at the University of Texas at Austin. A university announcement said that when Stampede2 is fully completed this summer, it will be able to perform 18 quadrillion mathematical operations per second, making it one of the most powerful supercomputers in the world. The supercomputer was built with funding from the National Science Foundation and will serve as a resource for researchers across the country.
Phase one of the supercomputer, which is now complete, has already been ranked as the 12th most powerful supercomputer worldwide, according to the latest Top 500 list, which ranks computers by their ability to solve linear equations. Though impressive, Stampede2’s computing power still lags some way behind the fastest supercomputer in the world — China’s Sunway TaihuLight, which can perform 93 quadrillion calculations per second.
Stampede2’s processing power, expected to be roughly equivalent to 100,000 desktop computers when complete, will be shared by thousands of researchers. Applications can be made through the NSF’s XSEDE program, and time will be allotted through a competitive peer-review process.
Since April, researchers have been using Stampede2 to conduct large-scale studies of gravitational waves, earthquakes, cancer proteins and nanoparticles. UT Austin said that neither the university nor its partner institutions in managing the supercomputer — Clemson, Cornell, Indiana and Ohio State Universities and the University of Colorado — had preferential access to Stampede2.
Dan Stanzione, TACC’s executive director, said that Stampede2 represents a “new horizon” for academic researchers, and will serve as “a workhorse for our nation’s scientists and engineers.” The supercomputer will support innovation in almost every area of research and development, the university said, from providing insights to fundamental theory to applied work with near-term impact on society.
Stampede2 replaces Stampede1, which began operations at UT Austin in 2013 and will be fully decommissioned by this fall. Stampede2 is said to occupy half the physical space and consume half the power of its predecessor, but it has double the memory, storage capacity, bandwidth and peak performance.
A spokeswoman for TACC said that a key difference between Stampede1 and Stampede2 is that the new supercomputer will use Intel’s latest Xeon Phi accelerator as its main processor — giving greater power and cost efficiency when running multiple calculations simultaneously. “Most codes are going to run better on the next generation of Xeon Phi as the main processor,” said the spokeswoman. “Given the diversity of science that we need to support, we need a diversity of architectures to go with that.”
David Schnyer, a professor of cognitive neuroscience at UT, used Stampede1 last year to run machine-learning algorithms that may one day be able to spot early indications of mental illness in brain scans. He said that although the original Stampede supercomputer had impressive capability, access wasn’t always prompt due to high demand.
“There were bottlenecks,” said Schnyer. “A lot of times we’d be expecting things to be by done by Monday, and we’d come in Monday morning to find they hadn’t run yet.” With Stampede2, Schnyer says, researchers have been told these delays will be alleviated — helping researchers stick to their schedules and work more efficiently.
Rommie Amaro, a professor of biochemistry at the University of California, San Diego, was granted early access to Stampede2. She described it as a “tremendous machine” and agreed with Schnyer that its increased processing power should mean that “lines should turn over faster” than they did for Stampede1.
However, speed isn’t the most important difference between the two machines, said Amaro. “More than speed, it’s the scope of calculations we can now run,” she said. Amaro’s research involves building model biological systems to study what happens to targets such as the tumor-suppressor protein p53, under different conditions. “With Stampede2 we can create many different versions of the model and explore what happens. In the past, we were much more limited.”