HPC MSU

High Performance Computing at the HPC²

The HPC² provides substantial high performance computing resources for use by its member centers.

Raptor is a 2048 processor cluster composed of 512 Sun Microsystems SunFire X2200 M2 servers, each with two dual-core AMD Opteron 2218 processors (2.6GHz) and 8 GB of memory (for a total of 4TB). All of the compute nodes are diskless. The system uses gigabit ethernet to connect the 32 nodes in each rack together, and 10 GbE to connect the 16 racks to one another. Raptor has a peak performance of over 10.6 teraflops (trillion calculations per second). On the November 2006 Top 500 Supercomputer Sites list, Raptor debuted as the 115th most powerful computer in the world, and the 18th most powerful computer at any academic site in the United States.

Maverick is a 384 processor Linux-based supercluster composed of 192 IBM xSeries x335 servers, each with two 3.06GHz Xeon processors, and 2.5 GB of memory. Each of the compute nodes are diskless, so they must load their operating system at boot-time into a RAM disk located in system memory. Each node is connected to the cluster's internal network via InfiniBand, which is a very high speed (10 gigabits per second), very low latency network architecture. At the time of its installation in November 2003, Maverick was one of the largest InfiniBand-based clusters in the world, and the largest diskless InfiniBand cluster in the world. Maverick has a theoretical peak performance of 2.35 teraflops (trillion calculations per second), and a measured performance on the LINPACK benchmark of nearly 1.4 teraflops. On the June 2004 Top 500 Supercomputer Sites list, Maverick was listed as the 160th most powerful computer in the world and the 16th most powerful computer at any academic site in the United States.

The HPC² also has an SGI ORIGIN 3900 server with 512 700MHz R16000 processors and 512GB of memory as well as two SGI ORIGIN 3800 servers, one with 128 400MHz R12000 processors and 128GB of memory, and another with 64 400MHz R12000 processors and 64GB of memory. The large shared memory images on these machines make them extremely useful for applications that do not perform well in distributed memory environments like those found in clusters.

In addition to these systems, the HPC² has many smaller SUN, SGI, and IBM/Linux based servers that provide computational tasks for various groups and individual projects.