SUPERCOMPUTER VENDOR Cray has taken the wraps off its XC30 supercomputer cluster formerly known as Cascade.
Cray's Cascade high performance computing (HPC) cluster has been talked about for years and marks not only the first showing of the firm's Aries interconnect but its departure from AMD processors. Cray claims the Cascade XC30 can deliver 66 TeraFLOPS per cabinet at launch.
Cray's Cascade XC30 ends a decade long association with AMD and its Opteron processors, with the firm opting for Intel Sandy Bridge Xeon processors and supporting the Xeon Phi accelerator. While the change isn't surprising given Cray's close links with Intel, the firm also touted its Aries interconnect that makes its debut with the XC30 and gives the firm its unique selling point over competitors.
The Cascade XC30 packs eight Intel Xeon chips into each compute blade, with each chip having access to four DIMM slots each on a separate channel. While Cray hasn't done much with Intel's processors other than drop them into its motherboard, the firm was keen to show off its Aries interconnect that uses PCI Express 3 to link compute nodes and processors.
According to Cray, the structure of the Aries interconnect means physical distance between nodes no longer degrades performance, resulting in higher average bandwidth at all parts of the cluster. The firm said this should mean that jobs are not sent just to nodes that are in the same cabinet, underutilising system capacity, but to all nodes, maximising performance responsiveness and both capacity utilisation and throughput.
Cray's Cascade XC30 uses the firm's Cray Linux Environment, a Suse based Linux distribution that the firm said all its HPC customers must run. The firm also touted compiler support from GCC, Intel and PGI and applications software support from firms such as Mathworks, Accelrys and Simula.
Cray said Cascade XC30 cluster shipments are beginning today with general availability in the first quarter of 2013. The firm expects Cascade XC30 based clusters to serve HPC customers spending $250,000 to over $100m. µ
Uses 20 percent less power than traditional systems
It's becoming more prevalent in car research and development
Sign up for INQbot – a weekly roundup of the best from the INQ