High Performance Computing Center North (HPC2N) has built an HPC Linux cluster with a peak performance of 800 Gflops/s. The 0.8 teraflops cluster was completed during spring 2002.

The cluster consists of 240 AMD Athlon MP2000+ processors in 120 dual nodes. The CPU's runs at 1.667 GHz giving each CPU a peak floating point performance of 3.334 Gflops/s. For high parallell performance the system is equipped with the new generation, 3-dimensional, high bandwidth, low latency SCI interconnect from Dolphin ICS. The network has 667 Mbytes/s bandwidth and an application level latency of 1.46 µs. The network connects the nodes in a 3-dimensional torus organized as a 4x5x6 grid. Each node is also equipped with fast Ethernet.

The cluster is intended for production use, and the majority of CPU-time is allocated through SNAC.

Superior benchmark performance

The system 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. In order to reach half of that performance, the matrix size needs to be 24 570. (This value is normally referred to as N1/2.)

This benchmark result corresponds to place 59 on the most recent Top500 list at the time of installation (from November 2001). The Top500 list ranks the world's 500 fastest computers according to their performance on the HP-Linpack benchmark. On the subsequent list of June 2002, the system is on the 94th position, still making it Sweden's fastest supercomputer.

Further information

The software installation was done using FAI, Fully Automatic Installation for Debian.

If you have any questions about the new cluster please send a mail to: info@hpc2n.umu.se or support@hpc2n.umu.se