HPC Challenge is a benchmark suite comprising 7 micro benchmarks designs to assess a number of HPC system characteristics. The suite includes HPL, DGEMM and FFT which test floating point performance. STREAM and RandomAccess are included to access memory performance. PTRANS and 'Communication bandwidth and latency' are used to assess network performance. source
HPCC website http://icl.cs.utk.edu/hpcc/
HPL Calculator http://hpl-calculator.sourceforge.net/
source Intel Compiler, MKL and MPI.
source /opt/intel/impi/5.0.1.035/bin64/mpivars.sh intel64 source /opt/intel/bin/compilervars.sh intel64 source /opt/intel/mkl/bin/mklvars.sh intel64
Download the source from the website:
cd ~ mkdir HPCC cd HPCC wget http://icl.cs.utk.edu/projectsfiles/hpcc/download/hpcc-1.5.0a.tar.gz tar -xf hpcc-1.5.0a.tar.gz cd hpcc-1.5.0a
Use the Makefile which is adapted from the HPL which ships with the Intel MKL package:
cp Makefile.intel64 ~/HPCC/hpcc-1.5.0a/hpl
Make with:
make arch=intel64
Before running the benchmark, edit the hpccinf.txt file. Set appropriate N, NB, P and Q values (using the calculator linked above.)
Run the benchmark using:
mpirun -np <N> -hostfile <HF> ./hpcc
Download this script to format the output data: format.tar
./format.pl -w -f hpccoutf.txt
Below is the result from the Dell C4130, with its 24 Haswell cores.
HPL_Tflops G-HPL 0.7431 Tera Flops per Second PTRANS_GBs G-PTRANS 0.0010 Tera Bytes per Second MPIRandomAccess_GUPs G-RandomAccess 0.0034 Giga Updates per Second MPIFFT_Gflops G-FFT 0.0011 Tera Flops per Second StarSTREAM_Triad*CommWorldProcs EP-STREAM Sys 0.0897 Tera Bytes per Second StarSTREAM_Triad EP-STREAM Triad 3.7380 Giga Bytes per Second StarDGEMM_Gflops EP-DGEMM 33.3345 Giga Flops per Second RandomlyOrderedRingBandwidth_GBytes RandomRing Bandwidth 1.4625 Giga Bytes per second RandomlyOrderedRingLatency_usec RandomRing Latency 0.8082 micro-seconds