This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| IBM Summit | |
|---|---|
| Name | IBM Summit |
| Manufacturer | IBM |
| Developer | Oak Ridge National Laboratory |
| Release | 2018 |
| Power | 13 MW |
| Speed | 200 PFLOPS (peak) |
| Architecture | Heterogeneous CPU/GPU |
IBM Summit
IBM Summit was a flagship supercomputer developed by IBM in collaboration with Oak Ridge National Laboratory and other partners. It combined high-performance hardware and software to achieve leadership in computational science and artificial intelligence, serving research programs across physics, climate science, genomics, and national security. Summit integrated technologies from multiple vendors and institutions to set records in speed and energy efficiency while influencing subsequent exascale initiatives.
Summit was designed as a petascale-to-exascale precursor built at Oak Ridge National Laboratory under the United States Department of Energy's CORAL procurement program, alongside systems at Lawrence Livermore National Laboratory and Argonne National Laboratory. The system featured a heterogeneous architecture combining processors from IBM and accelerators from NVIDIA with interconnects supplied by HPE Cray technologies, and storage solutions influenced by designs from Seagate and Intel. Its deployment involved partnerships with US government agencies, academic consortia such as University of Tennessee collaborators, and industry players including Red Hat for system software integration.
Summit's hardware roadmap built on processor roadmaps from IBM POWER9 families and accelerator advances from NVIDIA Tesla V100. The system used a node architecture pairing two POWER9 CPUs with six NVIDIA GPU accelerators, linked by high-bandwidth interconnects derived from NVLink and InfiniBand technologies developed by Mellanox Technologies. The software stack combined kernels and compilers from IBM XL and GCC toolchains, runtime systems influenced by OpenMP and CUDA, and orchestration layers drawing on SLURM and Rocks Cluster Distribution traditions. Storage and I/O strategies referenced parallel file systems such as Lustre and object stores informed by Ceph, while system management adopted practices from BMC standards and telemetry frameworks used at Argonne National Laboratory and Lawrence Berkeley National Laboratory.
Summit achieved leading positions in the TOP500 list and posted strong results on the HPCG and LINPACK benchmarks during the late 2010s. Peak theoretical performance exploited mixed-precision math popularized in deep learning research from groups at Google and Facebook and leveraged tensor core concepts pioneered in NVIDIA accelerator roadmaps. Benchmarking campaigns were coordinated with centers such as NERSC and evaluated against workloads similar to those run on Titan and later compared to Fugaku and Sierra. Performance tuning employed libraries and frameworks from OpenBLAS, cuBLAS, MAGMA, and optimizations demonstrated at conferences like SC Conference and International Conference for High Performance Computing, Networking, Storage and Analysis.
Summit supported simulations and data analytics across domains including high-energy physics projects linked to CERN, astrophysics collaborations with NASA, climate modeling groups at NOAA and IPCC author teams, and genomics initiatives connected to National Institutes of Health. Machine learning workloads adopted frameworks such as TensorFlow, PyTorch, and Keras to accelerate research in drug discovery with partners like Pfizer and Eli Lilly and in materials science collaborations with MIT and Caltech. Computational chemistry and molecular dynamics used software from GROMACS, LAMMPS, and NAMD, while fusion energy simulations referenced efforts at ITER and Princeton Plasma Physics Laboratory.
Commissioned in 2018, Summit entered operational service supporting research allocations administered through programs at DOE Office of Science and scheduling frameworks similar to allocations at XSEDE and INCITE. Its operational lifecycle included security and compliance practices aligned with standards from NIST, workflow integrations using tools from GitLab and Jenkins, and data provenance facilitated by metadata standards used at National Center for Atmospheric Research. Collaborative projects spanned universities such as University of California, Berkeley, University of Chicago, and Stanford University, and multinational partnerships with institutions in United Kingdom, Germany, and Japan.
Summit influenced the design of successor systems and national exascale strategies pursued by DOE and informed procurement decisions for machines like Frontier and other exascale platforms. Its role in accelerating artificial intelligence research echoed developments at DeepMind and OpenAI in demonstrating the value of heterogeneous architectures for mixed scientific and machine learning workloads. The system contributed to publications in journals such as Nature, Science, and Physical Review Letters and impacted standards discussions at forums like IEEE and ACM. Lessons from Summit shaped training programs at institutions like Oak Ridge National Laboratory Leadership Computing Facility and curricula at Georgia Institute of Technology and Carnegie Mellon University.