Leibniz-Rechenzentrum (LRZ) now with 2 TFlop/s on its HitachiMunich 16 January 2002 Leibniz-Rechenzentrum (LRZ) of the Bavarian Academy of Sciences upgraded its Hitachi SR8000 and increased its memory capacity and performance by 50%. In addition to the SR8000, LRZ also runs an IBM 8-way Power-4 shared memory multiprocessor, a Linux Cluster, and a Fujitsu Siemens VPP700 vector computer with 52 nodes. The latter has a peak performance of 114 GFlop/s and serves as high performance platform for Bavarian universities and research activities.
During the Christmas and New Year holiday period, the SR8000 operation was interrupted for installation of the upgrade to the second phase of the procurement, conforming with the contractual requirements.
The phase 1 configuration consisted of 112 SMP nodes, each node comprising eight processors working in shared memory mode. The peak performance of each node is 12 GFlop/s, summing up to a
total of 1.344 TFlop/s. Hitachi installed additional 56 nodes. This increased the peak performance of the machine to 2,016 Gigaflop/second. Due to the measured LINPACK performance of 1.65
TFlop/s LRZ hopes to return to a ranking in the top 10 of the June 2002 Top500 list.
The memory capacity of the SR8000 was expanded too, from 900 to 1300 GigaBytes. The disk space was extended to about 10 TeraBytes. The hardware part of the upgrade works proceeded very
quickly with the machine booting correctly immediately after integration of the new nodes. Due to improvements in the operating system, only few software problems are still present,
so after installation of a new operating system release, LRZ expects good stability also for the upgraded machine.
On January 10, user operations were reinitiated. Currently, LRZ runs the 30 day acceptance test, until now without failures.
Physically, the LRZ now runs Hitachi's biggest machine worldwide, the 168 compute nodes require four cabinet heads. The SR8000 at University of Tokyo contains only 144 nodes in three
cabinet heads, however the processors run at a higher frequency resulting in a peak performance of 2.07 TFlop/s. This means a node has a peak performance of 14.4 GFlop/s.
Currently, about 60 projects from different application areas run on the Hitachi supercomputer: applied mathematics, astrophysics, bio sciences, fluid dynamics, chemistry, geo
sciences, informatics and physics. The competence network (competence network for technical and scientific supercomputing in Bavaria, KONWIHR) has about 4.6 million Euro (more than 4
million US$) at its disposal for strategic improvement of software and algorithms for use in high performance computing.
The Leibniz-Rechenzentrum is available at http://www.lrz.de.
Uwe Harms [News on Advanced IT] [Calendar] [Analysis] [IT in Medicine] |
 |