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.
| Cray Gemini | |
|---|---|
| Name | Cray Gemini |
| Developer | Cray Inc. |
| Type | Network interconnect |
| Introduced | 2010 |
| Discontinued | 2013 |
| Predecessor | SeaStar2 |
| Successor | Aries |
Cray Gemini is a high‑performance network interconnect chip and router developed by Cray Inc. for petascale supercomputers. It provided low‑latency, high‑bandwidth communication for large clusters used in scientific research, digital modeling, and national laboratory installations. Gemini served as the fabric for systems deployed by institutions such as the Oak Ridge National Laboratory, Lawrence Livermore National Laboratory, and commercial vendors integrating Cray hardware.
Gemini was introduced to support Cray's next‑generation supercomputing platforms alongside the XtremeData compute blades and Linux‑based service nodes. Designed during the era of the Petascale Era and the TOP500 competition, Gemini competed with interconnects from InfiniBand Trade Association members and vendors including Mellanox Technologies and Intel. It targeted workloads from projects like Climate Modeling at National Center for Atmospheric Research and simulation programs at Los Alamos National Laboratory.
The Gemini ASIC integrated routing, network interface, and protocol offload in a single chip to connect compute blades and service nodes. Its design drew on concepts from earlier Cray fabrics and contemporary router designs used by Sun Microsystems and IBM in their SMP and cluster systems. Key elements included packet routing, adaptive routing algorithms similar to those used in IBM Blue Gene systems, and support for topologies commonly employed in supercomputers such as 3D torus and dragonfly variants tested at Argonne National Laboratory.
Gemini provided low latency and high bandwidth through a point‑to‑point, direct network topology combining link aggregation and virtual channels. Features included hardware support for Remote Direct Memory Access modeled after interfaces promoted by the OpenFabrics Alliance, congestion control mechanisms akin to those in InfiniBand, and error detection comparable to telecom standards used by Cisco Systems. The chip exposed endpoints for MPI stacks used by research groups at Stanford University, Massachusetts Institute of Technology, and University of California, Berkeley.
Designed for petascale class systems, Gemini offered performance scaling across thousands of nodes, supporting multi‑rack configurations deployed by government labs. Benchmarking efforts compared Gemini‑based systems against platforms from Hewlett Packard Enterprise, Dell EMC, and accelerator‑based systems integrating NVIDIA GPUs. Workloads including computational fluid dynamics used by NASA and molecular dynamics simulations from Brookhaven National Laboratory highlighted Gemini's throughput and latency characteristics.
Gemini was integrated into Cray XE6 and XE6m systems deployed at sites such as Oak Ridge National Laboratory for the Jaguar supercomputer and subsequent upgrades. Other deployments occurred at research centers including NERSC and laboratories under the Department of Energy. System integrators combined Gemini fabrics with blades featuring processors from AMD and later with accelerator hybrids using NVIDIA Tesla units in collaborations influenced by procurement programs run by European Organisation for Nuclear Research researchers and national computing initiatives in Japan.
Software stacks supporting Gemini included Linux distributions tailored by Cray, MPI implementations from the Open MPI community, and libraries used by computational projects at Princeton University and Caltech. Cray provided firmware and driver support aligned with standards pushed by the OpenHPC effort and interoperability testing with management tools from Red Hat and orchestration suites used at European Centre for Medium‑Range Weather Forecasts. Performance analysis tools from vendors like Intel and academic projects at University of Illinois Urbana‑Champaign aided tuning on Gemini networks.
Gemini influenced subsequent Cray interconnects and contributed technical lessons that informed the design of Cray's successor chip families, including Aries. Its deployment in national‑scale systems impacted research agendas at Lawrence Berkeley National Laboratory and informed procurement strategies at UK Science and Technology Facilities Council. Gemini's architectural choices reverberated in later high‑performance fabrics developed by companies such as Arista Networks and Mellanox Technologies, and in designs used for exascale planning at institutions like Oak Ridge National Laboratory and Argonne National Laboratory.
Category:Cray hardware