El Capitan, equipped with AMD processors, took first place in the ranking of the world’s most powerful supercomputers with a performance of 1.7 exaflops, surpassing the previous leader Frontier with 1.3 exaflops. Intel’s Aurora dropped to third place.
Image Source: AMD
The supercomputer will be used in the United States to simulate nuclear explosions and assess the state of the country’s nuclear arsenal. In addition, the system will allow the development of new intercontinental ballistic missiles (ICBMs) and solve problems related to high-performance computing and artificial intelligence. El Capitan is capable of processing data with high precision (FP64), which is necessary for scientific and engineering tasks, unlike systems focused only on AI tasks.
Image Source: AMD
El Capitan was built by HPE based on the Shasta architecture, which is also used in other exascale systems such as Frontier and Aurora. All three supercomputers rank top in the Top500, demonstrating HPE’s leadership in high-performance computing. Frontier, which is now in second place, also showed improved results compared to previous tests, increasing its performance to 1,353 exaflops.
The system is also known to consume more than 35 MW of power at full load and ranks 18th on the Green500 list of most energy-efficient supercomputers, boasting 58.89 GFLOPS per watt. The El Capitan supercomputer has more than 11 million processing cores integrated into the Instinct MI300A processors, which combine both the CPU and GPU in one package. Each MI300A processor includes 146 billion transistors and uses advanced 3D chip packaging technologies to significantly improve power efficiency and performance.
The situation with the Aurora supercomputer, built on Intel technology, attracts special attention. Despite previously reported results, the system was unable to provide new rating data, indicating ongoing hardware and cooling issues. At the same time, Aurora still remains the most powerful AI supercomputer in the world with a performance of 10.6 exaflops in mixed precision tasks.