World’s new fastest supercomputer is built to simulate nuclear bombs

The top spot in the league table of the world’s most powerful computers has changed hands, with one supercomputer built for US national security research ousting another.

Top500, the definitive list of the most capable computers, is based on a single metric: how fast a machine can solve vast numbers of equations, measured in floating-point operations per second, or FLOPS. A machine called Frontier built in 2022 was the first publicly acknowledged to have reached the exascale – a billion billion FLOPS.

Frontier was created by Oak Ridge National Laboratory in Tennessee to work on a range of complex scientific problems including climate modelling, nuclear fusion simulation, drug discovery and national security applications.

Now, Lawrence Livermore National Laboratory (LLNL) in California has created El Capitan, which is capable of 1.742 exaFLOPS, more than any other supercomputer.

The machine has been built in collaboration with the National Nuclear Security Administration, an arm of the Department of Energy dedicated to developing nuclear weapons science under tight security. The agency was formed in 2000 after the discovery that nuclear secrets had leaked to China from the Department of Energy.

El Capitan will essentially provide the vast computational power necessary to ensure the effectiveness of the US nuclear deterrent without having to carry out physical nuclear testing. LLNL claims that complex, high-resolution 3D simulations of nuclear explosions that would take months on Sierra, its most powerful system until now, will be done in just hours or days on El Capitan.

The world’s most powerful supercomputer is a title that doesn’t change hands as often as it once did, says Simon McIntosh-Smith at the University of Bristol, UK. He is developing the UK’s Isambard-AI supercomputer, which is expected to rank in the top 10 machines next year.

Frontier sat at number one for two years and El Capitan is expected to keep its position at least as long. “There used to be more turnaround, more turnover, but that’s all slowed down,” says McIntosh-Smith. This is because the trend known as Moore’s law, in which the computing power of a new chip doubled every two years, no longer holds true.

“As Moore’s law slowed down, the way you compensate for that is you just make it bigger and bigger. So in the past a system of the same size may have doubled in performance in two years,” says McIntosh-Smith. “Now it doesn’t. So it needs to be, say, 50 per cent bigger in two years and use 25 per cent more power. So it all gets more complicated, therefore has got more expensive. It’s all got harder.”