Summit (supercomputer)

Summit (supercomputer)
source
Name	Summit
Operator	Oak Ridge National Laboratory
Manufacturer	IBM
Introduced	2018
Location	Oak Ridge, Tennessee
Cost	approximately $200 million
Purpose	scientific research, national security
Power	13 MW (design)
Memory	10 PB (aggregate)
Processors	IBM POWER9, NVIDIA Volta GPUs
Storage	Lustre file system, petabyte-scale

Summit (supercomputer) is a petascale to exascale gateway system deployed at Oak Ridge National Laboratory for high-performance computing and scientific research. Built by IBM and instrumented with accelerators from NVIDIA for projects supported by the United States Department of Energy, Summit serves researchers from institutions such as University of Tennessee, Lawrence Berkeley National Laboratory, Argonne National Laboratory, and Los Alamos National Laboratory. Its procurement and operation involved collaborations with Cray Inc. engineers, federal programs like the Exascale Computing Project, and user allocations through the Oak Ridge Leadership Computing Facility.

Overview

Summit was introduced to provide capabilities for modeling challenges associated with climate change, materials science, astrophysics, genomics, and nuclear physics. Funded through DOE initiatives including the Advanced Scientific Computing Research office and installed at Oak Ridge Leadership Computing Facility, the system aimed to accelerate research previously constrained on systems such as Titan (supercomputer) and to complement capabilities at NERSC and ALCF. Summit's commissioning ceremony included participants from U.S. Secretary of Energy offices, representatives of IBM Research, and scientists from Oak Ridge National Laboratory divisions.

Architecture

Summit couples IBM POWER9 CPUs with NVIDIA Volta GPUs in a hybrid node design influenced by prior systems like Sierra (supercomputer) and enterprise deployments by IBM in collaboration with NVIDIA. Each node pairs multiple POWER9 processors and multiple Tesla V100 GPUs connected via NVLink and proprietary high-speed interconnects derived from technologies like InfiniBand. The topology integrates a high-radix fabric supporting MPI workloads common to codes from LANL, LLNL, and universities such as MIT and Stanford University. Storage and I/O use a parallel Lustre file system managed by software teams from Whamcloud and administrators with experience from Sandia National Laboratories and Pacific Northwest National Laboratory.

Performance and Benchmarks

At peak, Summit achieved LINPACK results that placed it atop the TOP500 list in 2018, surpassing predecessors including Sunway TaihuLight and Fugaku at various times in community rankings. Measured performance leveraged both double-precision and mixed-precision modes important to applications developed at Los Alamos National Laboratory and Lawrence Livermore National Laboratory. Benchmarks such as HPL, HPCG, and domain-specific tests from NERSC and the Exascale Co-design Center demonstrated strengths in machine learning workloads popularized by groups at Carnegie Mellon University, University of Washington, and industry partners like Google and Microsoft Research.

Applications and Use Cases

Researchers have used Summit for cosmology simulations tied to projects at NASA, large-scale molecular dynamics related to Brookhaven National Laboratory collaborations, and genomics analysis associated with Broad Institute datasets. Summit accelerated computational chemistry efforts at institutions such as Caltech and Harvard University and contributed to fusion modeling with teams from Princeton Plasma Physics Laboratory. Machine learning and AI workflows from groups at Stanford University and Berkeley AI Research exploited GPU-accelerated frameworks like those promoted by Facebook AI Research and OpenAI to analyze datasets from Human Genome Project-class initiatives and climate ensembles coordinated with NOAA.

Development and Deployment

The procurement involved contracts between Oak Ridge National Laboratory and IBM, with system integration drawing on expertise from NVIDIA and software stacks influenced by academic partners including University of Illinois Urbana-Champaign and University of Michigan. Deployment required facility upgrades at Oak Ridge National Laboratory, coordination with DOE sites including Argonne National Laboratory for user allocations, and cybersecurity review aligned with policies from National Institute of Standards and Technology and federal oversight by the Department of Energy. Training programs and user support were organized with assistance from university centers such as Texas Advanced Computing Center and community workshops at conferences like SC (supercomputing conference).

Energy Efficiency and Cooling

Summit's design balanced performance with power profiles referenced in Green500 lists and energy-aware studies by researchers from Lawrence Berkeley National Laboratory. Cooling infrastructure at Oak Ridge, Tennessee relied on data-center air and chilled-water systems similar to those deployed at Argonne National Laboratory and Sandia National Laboratories sites, while facility teams coordinated with vendors experienced in liquid-cooling pilots from Cray Inc. and enterprise deployments by IBM. Power usage effectiveness and sustainability planning referenced standards from ASHRAE and reporting frameworks used by Energy Information Administration and DOE sustainability offices.

Legacy and Successors

Summit's era influenced the design of successor systems such as Frontier (supercomputer), informed exascale roadmaps from the Exascale Computing Project, and guided procurement strategies for national labs including Oak Ridge National Laboratory and Argonne National Laboratory. Software and workflows developed for Summit have been ported to platforms at NERSC, ALCF, and international centers like PRACE and Fujitsu collaborations. The system's integration of CPUs and accelerators shaped architectures adopted by vendors including HPE and researchers at University of Cambridge and ETH Zurich studying heterogeneous computing.

Category:Supercomputers