NCSA's NT Cluster Scalability

Scalability on NCSA's NT Cluster for the ASPCG Kernel

The per processor MFLOP rate on the NCSA NT Cluster is plotted against corresponding data for the NCSA SGI Origin2000. The MFLOP rate is shown on the left-hand side of the graph.

Speedup on the NCSA NT Cluster for the ASPCG kernel is plotted against corresponding data for NCSA's Origin2000. The speedup uses the log scale on the right-hand side of the graph. Per processor performance of the NT is lower by a factor of about 1.45.

Questions? Email Danesh Tafti.

Origin2000 data date: August 1997
NT data date: June 1998

using MPI
single precision (having problems with double precision)
problem size: 1024x1024
block size: 32x32
nswp1 =40
nswp2 = 100

Tuning Highlights | PECM | SCD | NT