Cheyenne is a 5.34-petaflops, high-performance computer built for NCAR by SGI. The system was released for production work on January 12, 2017.
An SGI ICE XA Cluster, the Cheyenne supercomputer features 145,152 Intel Xeon processor cores in 4,032 dual-socket nodes (36 cores/node) and 313 TB of total memory.
Cheyenne's login nodes give users access to the GLADE shared-disk resource and other storage systems.
Data storage components provided by DataDirect Networks (DDN) give the GLADE system a total usable capacity of 38 PB. The DDN system transfers data at the rate of 200 GBps, more than twice as fast as the previous file system’s rate of 90 GBps.
Acknowledgements and citations: A requirement of all allocations and use of NCAR HPC resources managed by CISL, including the CMIP Analysis Platform and Research Data Archive, is to acknowledge NCAR and CISL support for your research and cite these resources in your publications. Our ability to identify supported scientific results helps ensure continued support from NSF and other sources for future HPC systems. See this page for how to acknowledge and cite these resources: Acknowledging NCAR and CISL.
|145,152 processor cores||2.3-GHz Intel Xeon E5-2697V4 (Broadwell) processors
16 flops per clock
|4,032 computation nodes||Dual-socket nodes, 18 cores per socket|
|6 login nodes||Dual-socket nodes, 18 cores per socket
256 GB memory/node
|313 TB total system memory||64 GB/node on 3,168 nodes, DDR4-2400
128 GB/node on 864 nodes, DDR4-2400
Mellanox EDR InfiniBand
|Partial 9D Enhanced Hypercube single-plane interconnect topology
Bandwidth: 25 GBps bidirectional per link
Latency: MPI ping-pong < 1 µs; hardware link 130 ns
|3 times Yellowstone computational capacity||Comparison based on the relative performance of CISL High Performance Computing Benchmarks run on each system.|
|> 3.5 times Yellowstone peak performance||5.34 peak petaflops (vs. 1.504)|
Cheyenne allocations are made in core-hours. The recommended method for estimating your resource needs for an allocation request is to perform benchmark runs. Some guidance is provided here.
The core-hours used for a job are calculated by multiplying the number of processor cores used by the wall-clock duration in hours. Cheyenne core-hour calculations should assume that all jobs will run in the regular queue and that they are charged for use of all 36 cores on each node.