PC GAMESS v. 7.1.1 benchmarks and scalability on Infiniband Opteron Windows CCS cluster


Number of cores used

Test 1, Wall clock time and relative speedup

Test 3, Wall clock time and relative speedup

Test 5, Wall clock time and relative speedup

Test 6, Wall clock time and relative speedup

4

1149.1

4.00

2254.0

4.00

2414.6

4.00

6286.3

4.00

8

603.9

7.61

1152.3

7.82

1380.8

6.99

3280.9

7.66

12

415.1

11.07

773.1

11.66

1057.4

9.13

2272.5

11.06

16

319.5

14.39

580.0

15.54

876.1

11.02

1767.2

14.23

20

264.9

17.35

465.0

19.39

784.9

12.31

1477.4

17.02

24

226.9

20.26

382.6

23.57

716.0

13.49

1265.5

19.87

28

202.1

22.74

323.2

27.89

674.3

14.32

1134.6

22.16

32

182.6

25.17

287.6

31.35

618.2

15.62

1020.3

24.64

36

165.3

27.81

254.8

35.38

593.4

16.28

937.3

26.83

40

156.0

29.46

233.8

38.56

584.6

16.52

878.3

28.63

44

146.3

31.42

212.5

42.43

574.6

16.81

825.4

30.46

48

133.3

34.48

195.5

46.12

551.9

17.50

774.9

32.45

52

127.3

36.11

183.7

49.08

536.0

18.02

729.7

34.46

56

129.3

35.55

169.8

53.10

512.4

18.85

707.0

35.57

60

130.3

35.28

162.9

55.35

549.0

17.59

703.5

35.74

64

115.2

39.90

157.7

57.17

507.4

19.05

672.3

37.40

 

Graphical representation of scalability



OS and hardware description


Windows Compute Cluster Server located at the Microsoft Enterprise Engineering Center. Cluster consists of 64 HP BL465c blades , each blade contains 2 Dual-Core Low Power AMD Opteron 2216 HE 2.4 GHz  Processors.  Each blade contains 8 GB EEC DDR 667 MHz of memory.  The interconnect is the 4x DDR IB Switch Module and the IB NIC in each blade is a single-port 4X DDR InfiniBand ® PCI-Express Mezzanine adapter   This all connects to two Voltaire ISR 9024 switches. OS version is Windows Server 2003 R2 64-bit SP2, CCP version 1.0



Tests description


Test 1, single-point direct DFT (B3LYP) energy plus gradient for medium-size system (623 basis functions). View image

Test 3, single-point direct MP2 energy for medium-size system (623 basis functions, the same system as one used for Test 1). View image

Test 5, single-point direct CASSCF(12,12) for medium-size system (retinal molecule, cc-pVDZ, 565 Cartesian basis functions) using ALDET code. View image

Test 6, single-point direct CIS energy plus gradient of first excited state of medium-size system (porphyrin molecule, cc-pVTZ (aug-cc on Nitrogens), 1130 Cartesian basis functions, D2h group). View image


Test comments


All tests were run in standard parallel mode using dynamic load balancing over p2p interface, with four processes per each four-core node. Wall clock times are given on master node in seconds. Test 5 is the most communication intensive and would scale better for larger job. For these tests, the most commonly used MPI calls are MPI_Allreduce and MPI_Bcast.


We are grateful to Dennis Crain and Jeff Swift (Microsoft) for providing us by access to this cluster and support.

Copyright © 2007 by Alex A. Granovsky

Press to visit PC GAMESS v. 7.0.7 benchmarks and scalability on Opteron Myrinet Linux cluster page

Press to visit PC GAMESS v. 7.0.7 benchmarks and scalability on large Opteron Pathscale Infinipath Infiniband Linux cluster page

Press to visit PC GAMESS v. 7.0.4 benchmarks and scalability on SKIF K-1000 (another large Opteron Infiniband Linux cluster) page

Press to visit PC GAMESS v. 7.0.4 benchmarks and scalability on 21-node Pentium 4 Infiniband Linux cluster page

Press to visit PC GAMESS' eight core systems performance comparison page

Press to visit PC GAMESS' Woodcrest vs. Opteron performance comparison page

Press to visit PC GAMESS Pentium 4 family Xeon processor benchmarks page to compare the results of these benchmarks with those obtained on Xeon DP processors.

Press to visit PC GAMESS Pentium 4 family benchmarks page to compare the results of these benchmarks with those obtained on various Netburst (Pentium 4 and Pentium D) processors.

Press to visit the PC GAMESS vs. WinGamess performance comparison page to compare the results of these benchmarks with those obtained on older processors. Input files can be found there too.