The fastest computer to date was the Frontier, which was built on the HPE Cray EX235a architecture, 64-core 3rd generation AMD EPYC processors clocked at 2 GHz and AMD Instinct MI250X accelerators. It had 9.07 million cores, a consumption of 24.6 MW and a theoretical peak performance of 2.06 ExaFLOPS. The measured value was 1.35 ExaFLOPS. But that is now changing, as a new supercomputer has been launched The Captain.
It is built on the HPE Cray EX255a architecture, while we also have AMD processors here. In this case, it is a 4th generation EPYC with 24 cores and a frequency of only 1.8 GHz, but this is complemented by more powerful AMD Instinct MI300A accelerators. The total number of cores is 11,039,616 and the peak consumption of the computer is 29.581 MW, 20% more than the Frontier. However, it is not fundamentally more efficient compared to it, because the peak performance is supposed to reach 2.746 ExaFLOPS (33.5% more) and the measured performance is 1.742 ExaFLOPS (+28.8%).
The result is that complex 3D simulations that would have taken weeks or months on Sierra, which was the second fastest computer in the world in 2018 (it’s 12th today), now take hours or days. El Capitan, like Frontier, is on US soil. El Capitan got the green light in 2019 when the US Department of Energy decided to invest 600 million USD in its construction. The aim is simulations for various nuclear projects (e.g. nuclear weapons without conducting their physical tests), fusion energy, climate or drug development.
Source: www.svethardware.cz
