Supercomputers enter their exaflop era

When movies, jokes, or athletes flop, it’s a bad thing. For supercomputers, on the other hand…the more flops the better.

Wait, what? Flops, or floating-point operations per second are a key measure of computational performance, and the basis of the annual Top500 supercomputer rankings that were released earlier this week.

The leading supercomputer in 2022 is the Frontier system, located at the Oak Ridge National Laboratory in Tennessee. It is “the first true exascale machine,” Top500 says, as it is the first known to demonstrate exaflop-level performance. The system was built by Hewlett Packard Enterprise—which helped build 19.2% of the computers on Top500’s list—and uses chips from AMD.

• An exaflop is equal to 10¹⁸ flops. As the New York Times puts it, that means it can handle “a billion billion calculations” per second.
• There might be two true exascale-level supercomputers in China, but the country did not submit results to Top500 this year.

These room-sized, high-speed computers are used for more than just the computational Olympics. For its part, Frontier is expected to help with research in areas like medicine and clean energy. In general, the plurality (26%) of systems on the Top500 list are intended to aid researchers, but there are other applications too, including...

• IT services (13%), cloud services (13%), and weather and climate research (8.7%).
• Nearly half (47%) of the computers are owned by “industry.”

Plus, as the NYT points out, the companies involved—HPE and AMD—could see increased interest in their computing products thanks to their part in claiming the No.1 spot.

History lesson: ENIAC, an apartment-sized system considered to be the first computer, could process ~500 flops in 1946. The smallest computers on the 2022 Top500 list are capable of 1.65 petaflops of performance. Petaflop= 10¹⁵, or 1,000 trillion, flops.