Benchmark results of QCDMPI

last updated: 1999/2/27

8**4/8(=2*2*2*1)PE, 5 Kbytes/comm.

Machinemicro sec/linkMB/sec
SX-4 1.74 38.9
Exemplar V2200 2.8 70
AP3000(200MHz) 5.33 26.4
AP3000(167MHz) 7.36 25.3
Cenju-3 29.6 9.51
AP1000+ 38.0 11.2
Paragon 51.7 17.9
Cluster 67.2 0.9
AP1000 151.6 4.0
SR2201 151.9 0.5

16**4/8(=2*2*2*1)PE, 70 Kbytes/comm.

Machinemicro sec/linkMB/sec
SX-4 0.55 515
VPP/500(*) 2.01 169
RS/6000 SP(160MHz) 3.28 63.8
AP3000(200MHz) 5.16 48.1
Exemplar V2200 6.19 26
AP3000(167MHz) 7.29 46.9
SR2201 12.3 9.4
AP1000+ 35.5 12.1
Cenju-3 30.1 13.2
Paragon 51.6 41.9
AP1000 164.7 4.8

32**4/8(=2*2*2*1)PE, 1 Mbytes/comm.

Machinemicro sec/linkMB/sec
SX-4 0.67 232
VPP/500(*) 1.94 186
Exemplar V2200 7.5 135
AP3000(167MHz) 8.21 48.3
SR2201 10.5 20.3

32**4/8(=8*1*1*1)PE, 4 Mbytes/comm.

Machinemicro sec/linkMB/sec
SX-4 0.68 345.2
SR2201 10.1 66.6
(*)not MPI but VPP/spread move, supported at QCDMPI ver.1.6.
Acknowledgments
I thank Dr. Shoji Hashimoto @ KEK for the benchmark of AP3000.
I thank Yukiya Aoyama @ IBM for the benchmark of RS/6000 SP.