The High Performance Computing Linpack Benchmark (HPL) is a software package that solves a (random) dense linear system in double precision (64 bits) arithmetic on distributed-memory computers. At present time, HPL is the benchmark used to provide values for the TOP500 List.

To compare your cluster with the ones at the TOP500, you can download the HPL package from the Critical HPC website. In the downloads section of csWMPI II, you'll find references to High Perfomance Linpack for all supported Platforms.

The official HPL website can be found here.

The HPL packages from Critical Software for MS Windows come with the source of HPL, a VisualStudio project, a reference BLAS library and a compiled binary, xhpl.

The executable and the needed configuration files are found in the bin directory of the unziped package. The full documentation of the HPL package can be found in the www directory. It includes detailed information on how to configure the different parameters of HPL.

Platform specific BLAS libraries usually produce better results than the included BLAS library. The Intel Math Kernel Library is a good choice and it's likely to increase the performance by 100%. To use a different BLAS library, you can use the provided VisualStudio Project to recompile the xhpl, removing the references to BLAS.lib and adding references to the new BLAS library.

The HPL package from Critical Software is a copy of the official HPL package, but all Makefile example scripts are set up to use csWMPI II for Linux. This is a source code package and compilation will be needed.

Please refer to the INSTALL file included in the HPL package for detailed information. Before running, make sure you create the csWMPI configuration files (csWMPI.clusterconf and pg file).

HPL needs an Algebra Library. You can download ATLAS from here. This is an open source BLAS implementation with platform specific instructions and it will also need compilation procedures. You are free to use any other BLAS library.