Testing provided by:
Andy Funk and Jeremy Kepner
MIT Lincoln Laboratory, Lexington, MA 02420
The HPC Challenge benchmark suite is a public and freely available collection of parallel benchmarks intended to stress various aspects of a high performance computing systems. These benchmarks include HPL, DGEMM, STREAM, PTRANS, RandomAccess, FFT, and Communication bandwidth and latency. More information about these benchmarks can be found on the HPC Challenge project web site: /hpcc/
The LLGrid TX2500 is a cluster of 416 compute nodes connected by InfiniBand. Each compute node consists of a dual 3.2 GHz Intel Xeon dual-processor with 8 Gigabytes of memory.
The reference implementation of HPC Challenge Version 1.0.0, which is available for download from the web site, is written in C and depends on third-party MPI and BLAS libraries. On the TX2500, OpenMPI 1.1.1 and ATLAS 3.0 BLAS libraries were used.
The attached results represent 77 runs of the HPC Challenge reference implementation on the TX2500. The runs were varied by number of processors and problem size. The numbers of processors used were 1, 2, 4, 8 , 16, 32, 64, 128, 256, 384,and 416, using one processor per node in each case. The problem sizes used were 128, 256, 512, 1000, 2000, 3000, and 4000 MB per process. Note: datasets below 2000 MB per process are less than the official HPC Challenge rules allows. These data are included for research purposes only.
This work is sponsored by Defense Advanced Research Projects Administration, under Air Force Contract FA8721-05-C-0002. Opinions, interpretations, conclusions and recommendations are those of the author and are not necessarily endorsed by the United States Government.
The spreadsheet of performance results obtained by running the HPC Challenge benchmark suite on the LLGrid TX2500 cluster is available here: hpccmitll.xls