There’s a new name on the Top500 list of supercomputers. El Capitan, housed at the Lawrence Livermore National Laboratory (LLNL) in California, has finally been brought online and immediately became the fastest supercomputer in the world with a High-Performance Linpack (HPL) score of 1.742 exaflops.
El Capitan, which features technology from HPE and AMD, including Cray Slingshot 11 network for data transfer and AMD Instinct MI300A APUs, is somewhat different from many of the other supercomputers on the Top500 list. While many are used for applications such as climate research, biomedical engineering or material science, El Capitan will be used for nuclear weapon science.
Nuclear testing without the fallout
Although El Capitan is housed at LLNL, it’s a resource shared by all three parts of US National Nuclear Security Administration (NNSA), which constitutes Sandia National Laboratories and Los Alamos, as well as LLNL.
Almost 80 years after the Trinity nuclear test at Los Alamos, which saw the first detonation of a nuclear bomb, El Capitan coming online means – in theory at least – further physical nuclear tests won’t be necessary.
According to Jill Hruby, Department of Energy (DOE) under secretary for nuclear security and NNSA administrator: “El Capitan’s introduction continues the capability advancement needed to sustain our stockpile without returning to explosive nuclear testing. This computational capability … is the heart of science-based stockpile stewardship.”
El Capitan is, LLNL says, the culmination of decades of work to enable the modeling of weapons performance and safety in a virtual environment. This includes ten years spent developing code optimized to work on exascale infrastructure, even before such a thing existed.
While nuclear testing and stockpile management may be the primary purpose of El Capitan, it’s not all virtual mushroom clouds. According to LLNL: “It also will be used to model advanced high-energy-density physics experiments, such as inertial confinement fusion … and detailed understanding of material behavior under extreme conditions.”
It will achieve this using tools such as the creation of digital twins and AI assistants trained on classified data.
Cool running
El Capitan has not only taken first place on the Top500 the moment it was turned on but also made the Green500 list of energy-efficient supercomputers, albeit in 18th place.
Top500 described this achievement as “quite impressive” in view of its top-scoring HPL benchmark of 1.742 EFlop/sec.
This is enabled largely by the energy efficiency of the HPE infrastructure used. The company has made a name for itself in fanless, direct liquid cooling – one of the most efficient ways of removing heat from a data center – with its Cray Supercomputing EX systems.
El Capitan runs on one of these direct liquid-cooled systems, HPE Cray EX255a. It also features 11,039,616 cores based on AMD 4th generation EPYC processors with 24 cores, AMD Instinct MI300A accelerators – which are also highly energy efficient – and uses the Cray Slingshot 11 network for data transfer.
This is the third exascale computer that HPE has rolled out in partnership with the DOE, the other two being Frontier and Aurora, and the second it has created in partnership with AMD.
