A module module that ranks supercomputers on their ability to solve formidable problems rather than on tender speed alone continues to benefit traction in a high-performance computing community.
More than 60 supercomputers were ranked by a rising tool, termed a High Performance Conjugate Gradients (HPCG) benchmark, in ratings expelled during a annual supercomputing assembly SC15 in late November. Eighteen months earlier, usually 15 supercomputers were on a list.
“HPCG is designed to element a normal High Performance Linpack (HPL) benchmark used as a central metric for ranking a top 500 systems,” pronounced Sandia National Laboratories researcher Mike Heroux, who grown a HPCG module in partnership with Jack Dongarra and Piotr Luszczek from a University of Tennessee.
The stream list contains a same entries as many of a tip 50 systems from Linpack’s TOP500 though significantly shuffles HPL rankings, indicating that HPCG puts opposite complement characteristics by their paces.
This is since a opposite measures supposing by HPCG and HPL act as bookends on a opening spectrum of a given system, pronounced Heroux. “While HPL tests supercomputer speed in elucidate comparatively candid problems, HPCG’s some-more formidable criteria exam characteristics such as high-performance interconnects, memory systems and fine-grain mild threading that are critical to a opposite and broader set of applications.”
Heroux pronounced usually time will tell either supercomputer manufacturers and users ride toward HPCG as a useful test. “All vital businessman computing companies have invested heavily in optimizing a benchmark. All participating complement owners have dedicated appurtenance time to make runs. These investments are a strongest acknowledgment that we have grown something useful.
“Many benchmarks have been due as complements or even replacements for Linpack,” he said. “We have had some-more success than prior efforts. But there is still a lot of work to keep a bid going.”