IBM ships Cell as part of canned cluster
ClearSpeed brought on board too
SC06 IBM has added some serious floating point wallop to its pre-packaged cluster line.
Customers can now order IBM's System Cluster 1350 product with the Cell-based BladeCenter QS20 blade server (3.0GHz) and with a floating point accelerator from ClearSpeed. These latest options join a wide range of Xeon-based, Power-based and Opteron-based servers that can slot into the 1350 Linux cluster.
IBM, like all the major server vendors, has been going at the pre-packaged Linux cluster idea for awhile. Later this month at the Supercomputing conference in Tampa, a number of vendors will announce increased support for the cluster edition of Windows with their systems as well.
IBM clearly timed the Cell and ClearSpeed announcements for the Supercomputing crowd. Both products cater to customers in the high performance computing community that often desire the best possible floating point performance for their mathematical and scientific software.
The Cell chip, for example, is a champ with single precision floating point operations, while ClearSpeed sells a floating point accelerator that plugs into a server's PCI-X slot. Some of the top supercomputers in the world make use of the ClearSpeed technology today.
"When a call is made by an application to a math library supported by the accelerator card, ClearSpeed software intercepts the call, calculates whether it should be sent to the accelerator hardware and, if so, transfers the required data to the board to compute the function," IBM explained in an announcement. "The answer is calculated on the board and the results are stored into host memory before returning to the application."
Customers can outfit the 1350 with some of IBM's latest Opteron servers, including the 1U x3455, the 2U x3655 and the 4U x3755. You can also pick from the latest and greatest Xeon- and Power-based boxes. All told, the 1350 can scale up to a 1,024-node cluster.
There's more information available here. ®
Sponsored: Benefits from the lessons learned in HPC