Dolphin Interconnect Solutions- High-Performance Computing (HPC) System at Umea University Built with Dolphin High-Speed Three-Dimensional Interconnect

6/24/2002 - The High-Performance Computing Center North (HPC2N), located at Umeå University, Sweden, has built a new HPC Linux cluster with a peak performance of 800 GigaFLOPS/s. This high performance is achieved using 240 AMD Athlon™ MP2000+ processors in 120 dual nodes and interconnected with the innovative WulfKit3, a high-speed interconnect system provided by Dolphin Interconnect Solutions that features a three-dimensional torus distributed switch topology.

According to Kåre Løchsen, Dolphin Interconnect CEO Managing Director, “the Umeå cluster is one of the first installations adopting the Dolphin three-dimensional cluster architecture. The fact that it is showcased at this national center for high-performance computing especially spotlights the power and innovation of the WulfKit3 interconnect system.”

The 120 nodes in the supercluster, also known as a Beowulf cluster, are interconnected as a 4x5x6 “torus,” or logical cube, using one WulfKit3 card in each node. WulfKit3 is designed around the scalable coherent interface (SCI) standard, which accounts for the very high bandwidth, low-latency interconnect performance. WulfKit3 is a combination of a PCI SCI Adapter Card developed by Dolphin Interconnect, and a powerful message-passing interface (MPI) implementation developed by Scali, also located in Oslo. Very large superclusters can be built and managed using WulfKit3.

In recent testing, the supercluster has shown superior performance on the HP-Linpack benchmark. The performance of 480.7 Gflops/s was obtained when solving an NxN dense linear system Ax = b for N = 116,100. This performance level clearly places the system in the top 100 of the world’s fastest supercomputers as tracked by, a joint project of the University of Mannheim and the University of Tennessee. The Umeå supercluster is the most powerful system in Sweden.

"This new supercluster is an impressive addition to the University's computing capabilities and substantially strengthens our leadership in HPC," said Bo Kågström, Director of HPC2N and Professor, Department of Computing Science, Umeå University. "The increased computing capabilities will be immensely beneficial to the academic and research programs that HPC2N supports," Kågström added.

The primary objective of HPC2N is to raise the national level of competence in high performance computing and to transfer HPC knowledge and technology to new users in academia and industry. The Center supports a variety of HPC users and projects, including scientific visualization, parallel numerical linear algebra, bioinformatics and computational bioscience. In addition to providing an HPC facility for research computing, the Center is actively involved in computer and HPC academic programs at the University.

Umeå University has been offering education in high-performance and parallel computing since 1985 and was the first in Sweden to provide HPC education at the graduate level.

The WulfKit3 cluster joins two existing HPC clusters at HPC2N, a 68-node IBM RS/6000 SP system and a smaller SGI Onyx2 system. As one of three national HPC centers in Sweden, HPC2N and these systems are funded by the Swedish Council for Planning and Coordination of Research (FRN), the Swedish Council for High Performance Computing (HPDR), the Knut and Alice Wallenberg Foundation, and the Kempe Foundations.

Partners in HPC2N are Umeå University, Luleå University of Technology, The Swedish Institute of Space Physics (IRF) in Kiruna, Swedish University of Agricultural Sciences (SLU) in Umeå, and Mid-Sweden University in Sundsvall, Härnösand, Örnsköldsvik and Östersund.

Dolphin Interconnect, PCI-SCI, WulfKit and WulfKit3 are trademarks of Dolphin Interconnect Solutions, Inc., in the United States and other countries. All other registered trademarks, trademarks and service marks are the property of their respective owners.

Previous Page | News by Category | News Search

If you found this page useful, bookmark and share it on: