Perlmutter (supercomputer)

Perlmutter (supercomputer). It is a pre-exascale supercomputer housed at the National Energy Research Scientific Computing Center (NERSC) within the Lawrence Berkeley National Laboratory in Berkeley, California. Named for the Nobel Prize in Physics-winning astrophysicist Saul Perlmutter, the system was delivered in 2021 to accelerate research for the United States Department of Energy's scientific missions. Perlmutter represents a significant advancement in high-performance computing infrastructure, designed to tackle grand-challenge problems in fields like climate science, materials science, and fundamental physics.

Overview

The system was developed through a partnership between NERSC, the United States Department of Energy's Office of Science, and the industrial vendor Hewlett Packard Enterprise (HPE), which acquired the pioneering supercomputer firm Cray Inc. Its deployment marked a major upgrade to computational resources available to the thousands of researchers supported by NERSC. Perlmutter is specifically engineered to handle both data-intensive and computationally intensive workloads, a design philosophy that supports the evolving needs of modern scientific discovery. The supercomputer's installation was a phased process, with its initial GPU-accelerated partition becoming operational in 2021, followed by a larger CPU-based partition.

Hardware and architecture

Perlmutter features a heterogeneous architecture combining AMD EPYC central processing units with NVIDIA A100 Tensor Core graphics processing units. The system is built on the HPE Cray EX platform, which utilizes the high-performance Slingshot interconnect to facilitate rapid data movement between nodes. This architecture includes a unique "data-centric" design, where the GPU nodes are directly connected to a massive all-flash Lustre (file system) storage system, dramatically accelerating data analysis for experiments like those conducted at the Dark Energy Spectroscopic Instrument (DESI). The final, full system comprises over 6,000 GPU nodes and more than 1,500 CPU nodes.

Performance and capabilities

In benchmark tests, Perlmutter has demonstrated a peak performance of approximately 3.9 exaflops in mixed-precision computing, a metric highly relevant for artificial intelligence and machine learning applications. It achieved a performance of 70.9 petaflops on the High Performance Linpack (HPL) benchmark, securing a top position on the TOP500 list upon its debut. The system's immense I/O bandwidth and integrated storage solution allow it to process the enormous datasets generated by next-generation experimental facilities, including the Linac Coherent Light Source (LCLS) at the SLAC National Accelerator Laboratory. Its computational power is also critical for simulations in quantum chromodynamics and cosmology.

Scientific applications

Research teams use Perlmutter for a vast array of projects, including mapping the universe in three dimensions with data from the Dark Energy Spectroscopic Instrument to study dark energy. In materials science, it enables atomic-scale simulations to discover new materials for batteries and catalysts. Climate scientists employ it for high-resolution models of the Earth's atmosphere and oceans to improve predictions related to climate change. Furthermore, the system is a pivotal tool for advancing fusion energy research, simulating plasma behavior for projects like the ITER international tokamak. Its AI capabilities are leveraged for tasks ranging from analyzing particle collisions at the Large Hadron Collider to optimizing renewable energy grids.

Naming and history

The supercomputer is named in honor of Saul Perlmutter, an astrophysicist at Lawrence Berkeley National Laboratory and a professor at the University of California, Berkeley, who shared the 2011 Nobel Prize in Physics for the discovery of the accelerating expansion of the universe. The naming reflects the machine's primary role in supporting astrophysics and fundamental energy science. The contract for the system, initially known as NERSC-9, was awarded to Cray Inc. (later HPE) in 2018, with a total cost of approximately $146 million funded by the United States Department of Energy. Its phased installation was completed in 2022, making it one of the world's fastest supercomputers for open science.

Category:Supercomputers Category:Lawrence Berkeley National Laboratory Category:2021 establishments in California