US Poised to Take the Lead in Exascale Computing
From almost the first moment that computing began, the quest has been to have faster, more powerful computers, with developments coming at a rapid pace.
Simple operations that took room-sized computers gave way to personal computers that performed many of the tasks that helped shape our modern world. In turn, supercomputers arose that were capable of delivering 100 quadrillion (1017) operations per second.
Ironically, the improvement of performance to such levels led back to room-sized machines with several different computer processors working together to conduct operations.
Innovation within the scientific community led to a sort of “arms race” between competing countries, each vying to have the highest-performing computer in the world according to the Top500 list, released bi-annually since 1993 thanks in large part to input from University of Tennessee Distinguished Professor Jack Dongarra.
“There used to be more volatility in the rankings, in terms of new machines and benchmarks” said Dongarra, who is also Director of the Innovative Computing Laboratory (ICL) at UT. “When June 2021 and then November 2021 lists, only one new computer entered the top 10, and it was number 10. Japan’s Fugaku is currently No. 1 and has been for since June 2020.”
That could soon change as the newest supercomputer at Oak Ridge National Laboratory, Frontier, comes online.
Fugaku performs 442 quadrillion floating point operations per second (petaflops), which is three times the rate of the No. 2 computer, ORNL’s Summit, which comes in at 148.6 petaflops.
That’s something Frontier is designed to surpass. The goal is for it to be capable of sustained exaflop performance.
The path to get there is part of the US Department of Energy’s (DOE) Exascale Computing Project, which is counting on advancements made through the ICL and other labs.
“We are developing algorithms, software, and tools for the DOE Exascale computers, including Frontier,” said Dongarra. “These efforts include everything from linear algebra methods, to mathematical libraries, to coding, and testing the hardware.”
If Frontier achieves its goals, crowning it the new No. 1 would be a fitting culmination to his career, as he announced his intent to retire in an interview with HPCWire in November.
“I’ve been here 32 years, and we’ve built the ICL into a world-class institution that has played a role in several key advances in computing,” said Dongarra. “My retirement doesn’t mean an end to that, just a new chapter. Really, it just means I won’t be teaching, but I’ll still be helping conduct research.”
Research that will, someday, help build a computer system even more powerful than Frontier.