Linpack benchmarks

These benchmarks are based on Intel's Optimized LINPACK Benchmark as distributed with Intel's mkl libraries v.10.1.1.019. On a single node and using the defaults, the xlinpack_xeon64 executable gives:

Linpack 1000, size vs. Gflops (one node, four cores)

The log file from the run is ...

With 33 Gflops per node, and for embarrassingly parallel problems, the cluster would deliver 0.3 Tflops. To get a feeling for this number, top500's November 2008 list reported its slowest entry to be at 22.1 Tflops (and this was not for an embarrassingly parallel problem ;-) ).

Since top500 was mentioned, this is possibly a proper place for a little cross-validated propaganda. Here it comes: the representation of the various operating system families in November's 2008 list is:

Count Share (%)
Linux 439 87.8 %
Windoze 5 1.0 %
Unix 23 4.6 %
BSD Based 1 0.2 %
Mixed 31 6.2 %
Mac OS 1 0.2 %
Totals 500 100.0 %

about/benchmarks/linpack.txt · Last modified: 2009/02/04 19:14 (external edit)