Sierra (supercomputer)

Sierra (supercomputer). Sierra is a petascale supercomputer housed at the Lawrence Livermore National Laboratory (LLNL) in California. It was developed through a partnership between LLNL, IBM, and NVIDIA as part of the United States Department of Energy's (DOE) Advanced Simulation and Computing Program. Primarily tasked with supporting the National Nuclear Security Administration's (NNSA) stockpile stewardship mission, Sierra performs advanced computational physics and engineering simulations to ensure the safety and reliability of the nation's nuclear arsenal without underground testing.

Overview

Sierra represents a cornerstone of the United States' high-performance computing (HPC) infrastructure for national security science. The system is managed by the Tri-Laboratory Operating System Stack (TOSS) team, a collaboration between Lawrence Livermore National Laboratory, Los Alamos National Laboratory, and Sandia National Laboratories. Its primary operational focus is on running large-scale, three-dimensional multiphysics codes that model complex phenomena in areas like hydrodynamics, radiation transport, and material science. The acquisition and deployment of Sierra were critical milestones in the NNSA's Advanced Simulation and Computing Program, following its predecessor Sequoia and preceding the exascale computing system El Capitan.

Hardware and architecture

Sierra utilizes a heterogeneous computer architecture combining central processing unit (CPU) and graphics processing unit (GPU) technologies. Each of its 4,320 compute nodes pairs two 22-core IBM POWER9 processors with four NVIDIA Tesla V100 GPU accelerators, interconnected via a dual-rail Mellanox InfiniBand network topology. This design, similar to the Oak Ridge National Laboratory's Summit system, emphasizes energy efficiency and computational throughput for data parallel workloads. The system incorporates over 1.5 petabytes of DDR4 memory and utilizes the IBM Spectrum Scale parallel file system for high-speed data storage and I/O.

Software and applications

The software environment on Sierra is built upon a Red Hat Enterprise Linux (RHEL) operating system, managed by the Tri-Laboratory Operating System Stack. Key applications are legacy Fortran and C++ codes from the Advanced Simulation and Computing Program, such as those used for weapons physics simulations, which have been refactored to leverage the GPU architecture using programming models like OpenMP, OpenACC, and CUDA. The system also supports software frameworks for machine learning and data analytics, enabling research in areas like uncertainty quantification and scientific visualization. Development efforts were supported by the DOE's Exascale Computing Project (ECP) to prepare codes for future architectures.

Performance and rankings

Upon its acceptance in late 2018, Sierra achieved a performance of 125 petaFLOPS on the LINPACK benchmark, securing the number two position on the November 2018 TOP500 list of the world's most powerful supercomputers, behind its sister system Summit. It also demonstrated high efficiency on the HPCG benchmark, reflecting its capability for real-world applications. In terms of Green500 rankings, which measure energy efficiency in FLOPS per watt, Sierra's hybrid CPU-GPU design placed it among the most efficient systems of its generation. Its sustained performance on large-scale NNSA workload simulations far exceeds that of prior systems like Sequoia.

History and development

The development of Sierra was initiated under the United States Department of Energy's CORAL (Collaboration of Oak Ridge, Argonne, and Livermore) initiative, a joint procurement effort aimed at delivering pre-exascale capability. The contract was awarded to IBM in November 2014, with NVIDIA and Mellanox as key technology partners. Installation at Lawrence Livermore National Laboratory's Terascale Simulation Facility occurred throughout 2018, with formal acceptance achieved in December of that year. Sierra entered full production in 2019, directly supporting the National Nuclear Security Administration's mission and serving as a technological precursor for the El Capitan system expected later in the 2020s.

Category:Supercomputers Category:Lawrence Livermore National Laboratory Category:IBM Category:Products introduced in 2018