Earlier this week, Lawrence Livermore National Laboratory announced it had begun building an El Capitan supercomputer designed to deliver over 2 FP64 ExaFLOPS of computational performance for a variety of classified national security-related research. bottom. In addition to El Capitan, LLNL has a small companion system called Tuolumne for unclassified research that provides him 10%-15% of El Capitan’s performance.
“We plan to introduce an unclassified system called Tuolumne,” Bronis R. de Supinski 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.”
El Capitan is the first exascale supercomputer based on AMD’s accelerated processing unit, which consists of Zen 4 general-purpose cores and CDNA 3-based computational GPUs, promising to deliver more than 2 ExaFLOPS of performance. increase. This is significantly higher than Frontier’s performance with Rpeak. 1.679 FP64 ExaFLOPS performance.
Only 10% of 2 ExaFLOPS (200 PetaFLOPS) would put Tuolumne in the top 10 supercomputers of the current version. top 500 list. At 15% of 2 ExaFLOPS, this machine challenges Leonardo, an Intel Xeon Platinum 8358 and Nvidia A100 based system with Rpeak performance of 304.47 PetaFLOPS.
Sensitive supercomputers such as LLNL’s current Sierra and upcoming El Capitan are used in matters of national security. For example, El Capitan is expected to be used primarily for stockpile management, a US program for reliability testing and maintenance of nuclear weapons without actual testing. In contrast, unclassified supercomputers such as LLNL’s Tuolumne are used for a variety of computational workloads across multiple disciplines, including scientific research, engineering simulations, 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 a wide range of molecular dynamics. Applications, climate, etc. One of them runs on Tuolumne, and in special cases we also offer shorter runs on larger systems.”
