For reference, the mpi timings can be approximately represented as Total Time ~ Const1/Cores *((Jobs per Bus)**0.6) + Const2 with Const1 ~ 1900, how much of the CPU time gets parallelized Const2 a non-parallelizable overhead of about 60
The dual quad-core has two memory buses. On Wed, Apr 23, 2008 at 7:17 PM, Laurence Marks <L-marks at northwestern.edu> wrote: > Dual Quad-Core E5410 @ 2.33GHz 1333 FSB > 8GB 667MHz DDR2 FB-DIMM > Infiniband (more than fast enough, WALL=CPU within delta) > Adjacent Cache Line Prefetch Enabled (might not be best, not sure) > ifort 10.1.015 mkl 10.0.2.018 > Blocksize = 64 > > Serial benchmark > Jobs/Node Threads Time > 1 8 53.58 > 2 4 85.37 > 4 2 138.03 > 8 1 274.03 > > MPI benchmark > Jobs/Node Nodes Time > 1 8 296.4 > 2 4 298.4 > 4 2 406.2 > 4 4 222.3 > 4 8 150.8 > 8 1 613.5 > 8 8 136.8 > > Run serial, 1 Job 8 Threads 764.58 (25% slower than mpi) > > first number: jobs/node; 2nd number: nodes > Run_1_1: TIME HAMILT (CPU) = 212.5, HNS = 229.9, HORB = 0.0, DIAG = > 1589.8 > Run_1_1: TOTAL CPU TIME: 2034.4 (INIT = 2.1 + K-POINTS = 2032.3) > Run_1_8: TIME HAMILT (CPU) = 31.8, HNS = 30.6, HORB = 0.0, DIAG = 231.4 > Run_1_8: TOTAL CPU TIME: 296.4 (INIT = 2.1 + K-POINTS = 294.3) > Run_2_4: TIME HAMILT (CPU) = 32.4, HNS = 31.0, HORB = 0.0, DIAG = 232.8 > Run_2_4: TOTAL CPU TIME: 298.4 (INIT = 2.1 + K-POINTS = 296.4) > Run_4_2: TIME HAMILT (CPU) = 34.8, HNS = 38.2, HORB = 0.0, DIAG = 331.0 > Run_4_2: TOTAL CPU TIME: 406.2 (INIT = 2.1 + K-POINTS = 404.1) > Run_4_4: TIME HAMILT (CPU) = 18.0, HNS = 21.0, HORB = 0.0, DIAG = 181.0 > Run_4_4: TOTAL CPU TIME: 222.3 (INIT = 2.1 + K-POINTS = 220.2) > Run_4_8: TIME HAMILT (CPU) = 10.7, HNS = 10.5, HORB = 0.0, DIAG = 127.3 > Run_4_8: TOTAL CPU TIME: 150.8 (INIT = 2.1 + K-POINTS = 148.7) > Run_8_1: TIME HAMILT (CPU) = 39.4, HNS = 68.6, HORB = 0.0, DIAG = 503.3 > Run_8_1: TOTAL CPU TIME: 613.5 (INIT = 2.1 + K-POINTS = 611.5) > Run_8_8: TIME HAMILT (CPU) = 7.0, HNS = 9.6, HORB = 0.0, DIAG = 117.9 > Run_8_8: TOTAL CPU TIME: 136.8 (INIT = 2.1 + K-POINTS = 134.8) > > -- > Laurence Marks > Department of Materials Science and Engineering > MSE Rm 2036 Cook Hall > 2220 N Campus Drive > Northwestern University > Evanston, IL 60208, USA > Tel: (847) 491-3996 Fax: (847) 491-7820 > email: L-marks at northwestern dot edu > Web: www.numis.northwestern.edu > Commission on Electron Diffraction of IUCR > www.numis.northwestern.edu/IUCR_CED > -- Laurence Marks Department of Materials Science and Engineering MSE Rm 2036 Cook Hall 2220 N Campus Drive Northwestern University Evanston, IL 60208, USA Tel: (847) 491-3996 Fax: (847) 491-7820 email: L-marks at northwestern dot edu Web: www.numis.northwestern.edu Commission on Electron Diffraction of IUCR www.numis.northwestern.edu/IUCR_CED