A comparison of the speed of the various cores in our clusters.
The series600 and series400 are fairly old chips and were purchased 5+ years ago. Only the CPU cores were used in the seriesGPU tests as the code did not address the Tesla cards. The 6130 and 6138 cores are very similar, the main difference being the number of cores on each CPU, as well as the size of the L2 and L3 caches. The performance differences between them being negligible. The Xeon 5506 cores cannot compete as they are 9 years old.
The code used (a simple mpi C program to move values around a matrix) exhibits classic asymptotic behaviour as the number of cores increase due to overhead and thread communication. However the series100 cores are 2 to 3 times faster than the AMD cores on Hex.
Additionally the RAM\core ratio for the series100 servers is far higher with 384GB RAM to 40 cores or 9.6GB RAM per core as opposed to 128GB of RAM for the series600, or 2GB RAM per core on hex.