NCSA 30 | NCSA transitions to microprocessor-based hardware
51975
post-template-default,single,single-post,postid-51975,single-format-standard,ajax_updown_fade,page_not_loaded,,tactile-ver-1.8.1,smooth_scroll,wpb-js-composer js-comp-ver-4.11.2.1,vc_responsive

NCSA transitions to microprocessor-based hardware

NCSA transitions to microprocessor-based hardware

By the end of 1994 NCSA removed its traditional vector machines—the Cray Y-MP and CONVEX C3880—leaving machines based solely on microprocessor hardware. The center utilized microprocessor-based systems from three vendors: Silicon Graphics Inc. (SGI), Convex Computer Corporation, and Thinking Machines Corporation. A 16-processor SGI POWER CHALLENGE machine was installed in November 1994, a CONVEX Exemplar 8-processor hypernode was installed in April 1994, and the Thinking Machines CM-5 was retained.

The SGI POWER CHALLENGE had 16 superscalar processors, each running at 75 MHz and 300 MFLOPS peak speed and each with 128 MB of RAM. The CONVEX Exemplar consisted of 8 processors, with a peak speed of 200 MFLOPS per processor and 512 MB of shared memory.

For many applications, it became possible to run on a workstation at similar or higher performance than the CRAY Y-MP—for example, one 17-MFLOPS run of ABAQUS on the Y-MP achieved 21 MFLOPS on a single processor of the SGI POWER CHALLENGE. Parallelizing these codes on the SGI CHALLENGE or CONVEX Exemplar yielded a potential of a few to tens of times speedup relative to single vector processor performance.