AMD to Help Upcoming US Supercomputer Achieve Not One, But Two Exaflops

The US government is tapping AMD chips to help it build a machine capable of beating the combined might of the world's top 200 leading supercomputers.

The Department of Energy today announced updated details regarding “El Capitan,” a supercomputer capable of achieving an exaflop, or a quintillion (10^18) calculations per second.

The $600 million system was originally slated to achieve only 1.5 exaflops once it’s completed at the Lawrence Livermore National Lab in California.

However, the US government now says El Capitan will be capable of hitting two exaflops when the system is delivered in 2023.

The speed increase is thanks to a partnership between the supercomputer’s maker Cray and chip vendor AMD, which will supply next-generation EPYC server CPUs and Radeon Instinct GPUs to power El Capitan.

Specifically, the upcoming supercomputer will use EPYC processors code-named “Genoa,” which will be built with Zen 4 Cores, a chip microarchitecture AMD has yet to mention in its roadmaps.

Each Genoa CPU will be connected to four AMD Radeon Instinct GPUs, which will do most of the processing.

It's unclear how many processor cores will be used to build El Capitan.

But the world’s current fastest supercomputers leverage over a million cores, while only achieving around a tenth of an exaflop.

Many other supercomputers on the top 200 list can only achieve around a thousandth of an exaflop. 

In El Capitan’s case, the supercomputer will be primarily used to create 3D simulations of nuclear explosions, saving the US the need to physically test nuclear warheads in the real world.

The US government is already carrying out these virtual experiments using the Sierra supercomputer, also based in the Lawrence Livermore National Lab.

However, the El Capitan system promises be 16 times more powerful, and will let researchers run more detailed nuclear simulations at a faster rate, enabling them to come up with new ways to secure the US’s nuclear stockpile.

Along with El Capitan, the US is building two other systems capable of achieving at least one exaflop.

Both Aurora and Frontier are set to go online 2021.

But whether the US will be the first to build an exascale system remains to be seen.

China has been prototyping its own exascale supercomputers with the goal of bringing them online as soon as this year.

The US government is tapping AMD chips to help it build a machine capable of beating the combined might of the world's top 200 leading supercomputers.

The Department of Energy today announced updated details regarding “El Capitan,” a supercomputer capable of achieving an exaflop, or a quintillion (10^18) calculations per second.

The $600 million system was originally slated to achieve only 1.5 exaflops once it’s completed at the Lawrence Livermore National Lab in California.

However, the US government now says El Capitan will be capable of hitting two exaflops when the system is delivered in 2023.

The speed increase is thanks to a partnership between the supercomputer’s maker Cray and chip vendor AMD, which will supply next-generation EPYC server CPUs and Radeon Instinct GPUs to power El Capitan.

Specifically, the upcoming supercomputer will use EPYC processors code-named “Genoa,” which will be built with Zen 4 Cores, a chip microarchitecture AMD has yet to mention in its roadmaps.

Each Genoa CPU will be connected to four AMD Radeon Instinct GPUs, which will do most of the processing.

It's unclear how many processor cores will be used to build El Capitan.

But the world’s current fastest supercomputers leverage over a million cores, while only achieving around a tenth of an exaflop.

Many other supercomputers on the top 200 list can only achieve around a thousandth of an exaflop. 

In El Capitan’s case, the supercomputer will be primarily used to create 3D simulations of nuclear explosions, saving the US the need to physically test nuclear warheads in the real world.

The US government is already carrying out these virtual experiments using the Sierra supercomputer, also based in the Lawrence Livermore National Lab.

However, the El Capitan system promises be 16 times more powerful, and will let researchers run more detailed nuclear simulations at a faster rate, enabling them to come up with new ways to secure the US’s nuclear stockpile.

Along with El Capitan, the US is building two other systems capable of achieving at least one exaflop.

Both Aurora and Frontier are set to go online 2021.

But whether the US will be the first to build an exascale system remains to be seen.

China has been prototyping its own exascale supercomputers with the goal of bringing them online as soon as this year.