NCSA's Linux supercomputer will test Itanium

NCSA's Linux supercomputer will test Itanium


How many Intel Itanium processors will it take to execute a trillion floating-point operations per second?

About one-third as many 1-GHz Pentium III chips as it takes to reach 1 TFLOPS, according to developers of a large Linux cluster at the National Center for Supercomputing Applications in Urbana-Champaign, Ill.

The center plans to install the 64-bit chips practically as soon as Intel Corp. introduces them later this year. The hybrid cluster of 32-bit Pentium IIIs and 64-bit Itaniums will be the 'first of its kind,' said Dave Gelardi, director of deep computing for IBM Corp., maker of the cluster platform. NCSA and IBM will install the 2-TFLOPS cluster in two stages, based on CPU architecture.

The first half of the cluster, completed in February, has 512 nodes with two 1-GHz Pentium IIIs each. The second half of the cluster, slated for installation in June, will have 160 nodes, each with two 800-MHz Itanium processors. But it will be capable of executing 1 TFLOPS all by itself, NCSA director Daniel Reed said.

The 64-bit Itaniums, in development since the mid-1990s, are supposed to arrive before summer. NCSA already has 32 of the 160 Itanium nodes running in its test laboratory on a 'pre-general availability' basis, Gelardi said.

IBM is using its commodity rackmount eServer x330 systems for the Pentium III nodes. The housings for the Itanium CPUs are slightly different and don't have a model number because IBM has not yet announced its IA-64 server line, Gelardi said.

The Pentium III nodes will run Red Hat Linux from Red Hat Inc. of Durham, N.C., whereas the Itanium section will run TurboLinux from TurboLinux Inc. of San Francisco. The Red Hat-TurboLinux combination will be compatible with another Linux cluster project called Los Lobos at the University of New Mexico, Reed said.

Bringing it all together

A Myrinet high-speed network fabric from Myricom Inc. of Arcadia, Calif., will unite the cluster. The fabric will provide bandwidth of 200 megabytes/sec in each direction to the distributed-memory cluster, Gelardi said.

Several other high-performance Linux clusters already use 32-bit Myrinet interconnects, but the NCSA system will be the first with 64-bit Myrinet, Reed said.

NCSA, located at the University of Illinois, has been working closely with Intel on development of the Itanium nodes, Reed said.

Under its agreement with the National Science Foundation, NCSA makes most of its supercomputer time available to researchers whose proposals have passed peer review. The new Linux cluster is likely to tackle complex problems in physics and chemistry, Reed said.

The supercomputer center, which gets its funding from NSF and other federal and state agencies, supports collaboration by scientists in the Defense Department's High-Performance Computing Modernization Program, the Energy Department, the National Institutes of Health and NASA.

Reed declined to disclose the price of the new cluster, saying only that it was in the millions of dollars.


  • Records management: Look beyond the NARA mandates

    Pandemic tests electronic records management

    Between the rush enable more virtual collaboration, stalled digitization of archived records and managing records that reside in datasets, records management executives are sorting through new challenges.

  • boy learning at home (Travelpixs/

    Tucson’s community wireless bridges the digital divide

    The city built cell sites at government-owned facilities such as fire departments and libraries that were already connected to Tucson’s existing fiber backbone.

Stay Connected