Lawrence Livermore National Laboratory (LLNL) has begun deploying the supercomputer El Capitan, which promises to achieve more than 2 FP64 ExaFLOPS of computational performance for sensitive national security research. In parallel, LLNL also plans to introduce a less powerful but very fast supercomputer named Tuolumne. This companion model is for unclassified research only and provides him 10%-15% of El Capitan’s computing power. This is enough to surpass most systems on the current Supercomputer Top 500 list.
LLNL’s El Capitan isn’t the first exascale system, nor the first to break the 2 FP64 ExaFLOPS performance record, but it will be the first exascale supercomputer to use AMD’s Instinct MI300A accelerated processing unit. It’s still a very specific system. It features a Zen 4 x86 core and a CDNA 3 based compute GPU. When El Capitan comes online, Rpeak performance will significantly surpass Frontier, the world’s fastest PC rated at his 1.679 FP64 ExaFLOPS.
Tuolumne is expected to use an architecture similar to El Capitan, so it will also use an Instinct MI300A APU. Even with just 1/10th the performance of El Capitan, Tuolumne joins the top of the global list of supercomputers capable of delivering 200 petaflops. This potential performance could put it among the top 10 supercomputers on the current Top 500 list. Still, if you can get 15% of El Capitan’s performance, you’d have a supercomputer with an Intel Xeon Platinum 8358 and an Nvidia A100 with an Rpeak performance of 304.47 PetaFLOPS, currently ranked as the 4th fastest supercomputer. It could be comparable to Leonardo.
“We are planning to acquire an unclassified system called Tuolumne.‘” Bronis R. De Spinski said in an interview. ExaScaleProject.org. De Supinski is Chief Technology Officer for Livermore Computing at LLNL. “It will be about 10% to 15% the size of El Capitan.“
Classified supercomputers, such as the existing Sierra and LLNL’s planned El Capitan, primarily serve national security needs. According to Bronis R. de Spinski, El Capitan will be used primarily for stockpile management programs to ensure the reliability of nuclear weapons without resorting to actual testing. Conversely, unclassified supercomputers like the upcoming Tuolumne will serve diverse computational requirements across different scientific disciplines, from research and engineering simulations to data analysis and weather forecasting.
“Tuolumne will make even more contributions to the wider scientific arena,” De Spinski added. “There’s a lot of materials modeling. We’ve typically done broad molecular dynamics. applications, climates, and things like that run in Tuolumne, and sometimes offer shorter runs on larger systems if there are special cases.“