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 XK7 | |
|---|---|
| Name | Cray XK7 |
| Manufacturer | Cray Inc. |
| Family | XK |
| Released | 2012 |
| Discontinued | 2016 |
| Type | Supercomputer |
| Cpu | AMD Opteron |
| Gpu | NVIDIA Tesla K20X |
| Memory | up to tens of TB |
| Os | UNICOS/lc (Linux-based) |
Cray XK7
The Cray XK7 is a supercomputer family developed by Cray Inc. that combined Cray Inc. engineering with accelerator-based computation to target exascale-era workloads. It integrated technologies from AMD and NVIDIA with Cray's interconnect and system software to serve national laboratories, research institutions, and commercial centers. The platform emphasized hybrid CPU–GPU nodes for simulation, modeling, and data analysis across domains including Lawrence Livermore National Laboratory, Argonne National Laboratory, and other high-performance computing sites.
The XK7 was introduced as a successor to earlier Cray designs such as the Cray XT5 and Cray XE6, inheriting the company's interest in scalability and low-latency networking. It targeted workloads characteristic of projects like the Human Genome Project, ITER, and climate modeling efforts associated with National Oceanic and Atmospheric Administration centers. Customers included entities involved with the National Nuclear Security Administration and scientific programs at Oak Ridge National Laboratory, leveraging the platform for computational fluid dynamics, astrophysics, and materials science. The system aligned with procurement priorities set by agencies such as the Department of Energy.
The XK7 architecture combined multi-core AMD Opteron processors with NVIDIA Tesla accelerator boards, specifically the NVIDIA Tesla K20X. Nodes paired CPUs and GPUs on a shared-memory blade, interconnected by Cray's proprietary Aries interconnect technology derived from earlier Cray interconnects. The physical system used modular cabinets similar to those in systems deployed at Los Alamos National Laboratory and Sandia National Laboratories, with cooling solutions paralleling designs used at National Supercomputing Centre (NSCC) facilities. Storage and I/O layers were commonly integrated with parallel file systems such as Lustre deployed at sites like Oak Ridge Leadership Computing Facility.
XK7 installations were benchmarked across suites including High Performance Linpack and application kernels from climate, combustion, and nuclear simulation communities. Peak performance figures for full cabinets reached multiple petaflops, comparable to contemporaneous systems from IBM and Fujitsu. Measured efficiency often depended on GPU-accelerated codes ported using models developed in laboratories such as Lawrence Berkeley National Laboratory and collaborations with vendors like NVIDIA Corporation. XK7 performance was cited in comparisons with systems participating in the TOP500 and Green500 lists, where energy efficiency and sustained application throughput were emphasized by agencies including the European Centre for Medium-Range Weather Forecasts and the US National Science Foundation.
The XK7 ran a Linux-based OS variant, UNICOS/lc, and supported toolchains including compilers from Cray Inc. and vendor toolkits from PGI and GCC. Programming models exploited hybrid parallelism using MPI for distributed memory and OpenMP for node-level threading, coupled with accelerator offload through CUDA and frameworks such as OpenACC for porting codes from projects at Argonne National Laboratory and Lawrence Livermore National Laboratory. Debugging and performance analysis used tools like TotalView and vendor profilers provided by NVIDIA Nsight. System management integrated software from ADIC-era backup solutions and orchestration practices similar to those at National Energy Research Scientific Computing Center.
Major deployments included systems at national laboratories and university consortia: for example, an XK7-class system provisioned for the Oak Ridge Leadership Computing Facility and installations at Lawrence Livermore National Laboratory for classified and unclassified workload partitions. Other notable customers were research centers collaborating with European Organization for Nuclear Research-adjacent projects and government laboratories such as Los Alamos National Laboratory for weapon-physics simulation support. These installations often interfaced with national research networks like ESnet and used data management strategies common to PRACE projects in Europe.
Cray's development of the XK7 drew on the company's lineage from systems like the Cray T3E and organizational heritage dating to the original Cray Research. The program responded to market movements following acquisitions and leadership changes involving SGI and strategic partnerships with AMD and NVIDIA. Development cycles involved coordination with DOE procurement programs and academic consortia that earlier acquired Cray platforms such as Cray X1 and Cray XC30. The XK7's release coincided with a period of rapid accelerator adoption across high-performance computing driven by work at institutions like Stanford University and Massachusetts Institute of Technology.
The XK7 influenced subsequent Cray architectures, informing designs in the XC series and systems that incorporated next-generation accelerators from NVIDIA and new CPU offerings from Intel Corporation and AMD. Concepts proven on XK7—hybrid node design, Aries interconnect lessons, and UNICOS/lc system software—appeared in successors deployed for exascale roadmaps championed by DOE and international partners. Academic and government users migrated workloads to newer platforms influenced by XK7 architecture choices, as seen in transitions to systems at Oak Ridge National Laboratory and other leading centers.
Category:Supercomputers Category:Cray products