Gaussian - Performance on One Node

The improved intra-node memory architecture of SGI Altix ICE 8200EX Intel® Xeon® processor X5570 (code-named Nehalem) 2.93GHz, 12x4GB 1066MHz per node compared to SGI Altix ICE 8200EX Intel Xeon processor X5472 3.0GHz results in much better performance when all cores (8) in the node are being utilized by one Gaussian job.

Depending on the test case the speed up can range from ~1.1x to ~1.5x.

BLAS3 operations play an important role in the exceptional case, hismp4.

The average speed up (based on the geometrical mean) of the test set used is ~1.2x.

Gaussian – Parallel Performance on Intel Xeon processor E5570 (code-named Nehalem) vs Intel Xeon processor E5472 (code-named Harpertown) in SGI Altix ICE 8200EX

SGI Altix ICE 8200EX Intel Xeon processor X5570 (code-named Nehalem) 2.93GHz, 12x4GB 1066MHz per node shows excellent parallel scaling in this Gaussian DFT calculation.

At lower core-counts the absolute performance and parallel scaling seem similar to SGI Altix ICE 8200EX Intel Xeon processor X5472 3.0GHz.

At higher core counts SGI Altix ICE 8200EX Intel Xeon processor X5570 (code-named Nehalem) 2.93GHz, 12x4GB 1066MHz per node clearly outperforms the Harpertown based SGI Altix ICE, as the effect of better performance at the node level is amplified.