The current setup of 526 compute nodes at FSU HPC system was acquired during 3 different upgrade cycles. The theoretical peak performance of the complete system is 28.7 TeraFlops. To minimize downtime during these upgrades, extensive benchmarking of the whole system was put off and only parts where tested.
- The 512 opteron cores from the 2007 purchase were independently measured using the HPL benchmark and these tests resulted in a performance of 2.3 TF.
- In 2008, the 1024 barcelona cores were benchmarked at 7.6 TF; an efficiency of 81% of the peak performance.
- During the last upgrade cycle, 1024 AMD Shanghai cores were added and the 2048 cores from the 2008 and 2009 acquisitions where benchmarked at 14.5 TF. Clearly the barcelona cores were the limiting factor for this test.
In 2009, a number of 'smaller' systems were added to the HPC, for example, a 144 core 3-Leaf system (not yet benchmarked).
During the downtime on 10/14/2008, the filesystem performance was also measured. With 10 panasas shelves and 96 writers, the system had a data-to-disk rate of over 3.0 GB/sec.