Benchmarks at CyberInfrastructure Partnership

• What is CIP Benchmarks?
• Application Benchmarks
  • GAMESS
  • MILC
  • NAMD
  • PARATEC
  • WRF
• Synthetic Benchmarks
  • HPCC
  • IOR
• Computing Resources

GAMESS

“GAMESS is the General Atomic and Molecular Electronic Structure System code designed for high-performance simulation of general ab-initio quantum chemistry. GAMESS scales to hundreds of processors on high-end parallel platforms and tens of processors on commodity clusters using TCP/IP protocol and sockets, MPI, or SHMEM.“ [TI-06] The code version for the benchmarks is 22 NOV 2004 (R1). The two problems considered are those in the TI-06 benchmark set.

- The standard problem is a B3LYP(5)/6-311G(d,p) gradient calculation on a 6-atom molecule.

- The large problem is an MP2/6-311G(d,p) gradient calculation on a 5-atom molecule.

The table lists run times on the various computers for both problem sizes. The plots show the same results converted to relative speeds per processor in strong scaling scans. No results are shown for Blue Gene, since it has insufficient memory to run either benchmark problem.

T2 runs the fastest for the standard problem. All of the computers except Cobalt show reasonable scalability up to 128 processors.

DataStar is the fastest at 384 processors for the large problem and continues to scale well to twice that many processors. Except for Mercury, the other computers exhibit poorer scalability.

The differences in scalability between the computers are likely related to the significant amount of I/O that is done in the benchmark problems.