SPEC® MPIM2007 Result

Copyright 2006-2010 Standard Performance Evaluation Corporation

IBM Corporation

IBM Power 575

MPI2007 license: 0005 Test date: Jun-2008
Test sponsor: IBM Corporation Hardware Availability: May-2008
Tested by: IBM Corporation Software Availability: May-2008
Benchmark results graph

Results Table

Benchmark Base Peak
Ranks Seconds Ratio Seconds Ratio Seconds Ratio Ranks Seconds Ratio Seconds Ratio Seconds Ratio
Results appear in the order in which they were run. Bold underlined text indicates a median measurement.
104.milc 32 571 2.74 571 2.74 570 2.75 32 571 2.74 571 2.74 570 2.75
107.leslie3d 32 757 6.90 755 6.92 755 6.91 32 757 6.90 755 6.92 755 6.91
113.GemsFDTD 32 707 8.92 707 8.92 707 8.92 32 707 8.92 707 8.92 707 8.92
115.fds4 32 411 4.75 411 4.75 411 4.74 32 411 4.75 411 4.75 411 4.74
121.pop2 32 849 4.86 848 4.87 850 4.86 32 849 4.86 848 4.87 850 4.86
122.tachyon 32 1538 1.82 1537 1.82 1539 1.82 32 1538 1.82 1537 1.82 1539 1.82
126.lammps 32 755 3.86 755 3.86 755 3.86 32 755 3.86 755 3.86 755 3.86
127.wrf2 32 1317 5.92 1316 5.93 1317 5.92 32 1317 5.92 1316 5.93 1317 5.92
128.GAPgeofem 32 325 6.35 326 6.34 327 6.32 32 325 6.35 326 6.34 327 6.32
129.tera_tf 32 1225 2.26 1224 2.26 1224 2.26 32 1225 2.26 1224 2.26 1224 2.26
130.socorro 32 264 14.4  265 14.4  266 14.4  32 264 14.4  265 14.4  266 14.4 
132.zeusmp2 32 941 3.30 943 3.29 943 3.29 32 941 3.30 943 3.29 943 3.29
137.lu 32 552 6.65 553 6.65 555 6.63 32 552 6.65 553 6.65 555 6.63
Hardware Summary
Type of System: SMP
Compute Node: IBM Power 575
File Server Node: IBM Power 575
Head Node: IBM Power 575
Total Compute Nodes: 1
Total Chips: 16
Total Cores: 32
Total Threads: 32
Total Memory: 128 GB
Base Ranks Run: 32
Minimum Peak Ranks: 32
Maximum Peak Ranks: 32
Software Summary
C Compiler: IBM XL C/C++ Enterprise Edition V9.0
Updated with the Oct2007 PTF
C++ Compiler: IBM XL C/C++ Enterprise Edition V9.0
Updated with the Oct2007 PTF
Fortran Compiler: IBM XL Fortran Enterprise Edition V11.1
Updated with the Oct2007 PTF
Base Pointers: 64-bit
Peak Pointers: 64-bit
MPI Library: IBM Parallel Environment for AIX
V4.3.2.2
Other MPI Info: --
Pre-processors: --
Other Software: None

Node Description: IBM Power 575

Hardware
Number of nodes: 1
Uses of the node: compute, head, fileserver
Vendor: IBM Corporation
Model: IBM Power 575
CPU Name: POWER6
CPU(s) orderable: 32 cores
Chips enabled: 16
Cores enabled: 32
Cores per chip: 2
Threads per core: 1
CPU Characteristics:
CPU MHz: 4700
Primary Cache: 64 KB I + 64 KB D on chip per core
Secondary Cache: 4 MB I+D on chip per core
L3 Cache: 32 MB I+D off chip per chip
Other Cache: None
Memory: 128 GB (64x2 GB) DDR2 533 MHz
Disk Subsystem: 1x146 GB SFF SAS, 10K RPM
Other Hardware: None
Adapter: 0
Number of Adapters: 0
Slot Type: 0
Data Rate: 0
Ports Used: 0
Interconnect Type: 0
Software
Adapter: 0
Adapter Driver: 0
Adapter Firmware: --
Operating System: IBM AIX V5.3
with the 5300-08-02 Technology Level
Local File System: AIX/JFS2
Shared File System: NFS over ethernet
System State: Multi-user
Other Software: APAR IZ26983
software update for InfiniBand adapter drivers
IBM LoadLeveler for AIX
V3.4.3.2

General Notes

113.GemsFDTD (base): Applied maxprocandstop src.alt
129.tera_tf (base): Applied fixbuffer src.alt
127.wrf2 (base): Applied fixcalling src.alt
all ulimits set to unlimited
"petaskbind.sh" script used to bind each task to a unique processor
POE Environment variables set before executing benchmarks:
 CWD		     =/specmpi/mpi2007-1.0
 MP_ADAPTER_USE      =shared
 MP_EUILIB           =us
 MP_EUIDEVICE        =sn_all
 MP_SHARED_MEMORY	 =yes
 MP_SINGLE_THREAD	 =yes
 MP_WAIT_MODE        =poll
 MP_EAGER_LIMIT      =65536
 MP_BUFFER_MEM       =67108864
 MP_POLLING_INTERVAL =80000000
 MP_USE_BULK_XFER    =yes
 MP_BULK_MIN_MSG_SIZE=65536
 MP_STDINMODE        =none
 MP_LABELIO          =no
 MP_HOSTFILE         =$CWD/r35.32-1node
Other Environment variables
 MEMORY_AFFINITY     =MCM
 LDR_CNTRL 	     =DATAPSIZE=64K@TEXTPSIZE=64K@STACKPSIZE=64K
 XLFRTEOTPS          =intrinthds=1
submit command uses petaskbind.sh script to bind logical processors to ranks
 poe $CWD/petaskbind.sh $command -procs $ranks
The Gigabit ethernet switch is shared among many nodes, not just the cluster used in this benchmark.

Base Compiler Invocation

C benchmarks:

 /usr/bin/mpcc_r 

C++ benchmarks:

126.lammps:  /usr/bin/mpCC_r 

Fortran benchmarks:

 /usr/bin/mpxlf95_r 

Benchmarks using both Fortran and C:

 /usr/bin/mpcc_r   /usr/bin/mpxlf95_r 

Base Portability Flags

107.leslie3d:  -qfixed 
115.fds4:  -DSPEC_MPI_LC_NO_TRAILING_UNDERSCORE   -qfixed 
121.pop2:  -DSPEC_MPI_AIX 
127.wrf2:  -DNOUNDERSCORE   -DSPEC_MPI_AIX 
130.socorro:  -DSPEC_NO_UNDERSCORE   -qcpluscmt 
132.zeusmp2:  -qfixed   -DSPEC_SINGLE_UNDERSCORE 
137.lu:  -qfixed 

Base Optimization Flags

C benchmarks:

 -O4   -qarch=pwr6   -qtune=pwr6   -q64 

C++ benchmarks:

126.lammps:  -O4   -qarch=pwr6   -qtune=pwr6   -qstrict   -q64 

Fortran benchmarks:

 -O4   -qarch=pwr6   -qtune=pwr6   -qalias=nostd   -q64 

Benchmarks using both Fortran and C:

 -O4   -qarch=pwr6   -qtune=pwr6   -qalias=nostd   -q64 

Base Other Flags

C benchmarks:

 -w   -qsuppress=1500-036   -qipa=noobject   -qipa=threads 

C++ benchmarks:

126.lammps:  -w   -qsuppress=1500-036   -qipa=noobject   -qipa=threads 

Fortran benchmarks:

 -w   -qsuppress=1500-036   -qsuppress=cmpmsg   -qipa=noobject   -qipa=threads 

Benchmarks using both Fortran and C:

 -w   -qsuppress=1500-036   -qsuppress=cmpmsg   -qipa=noobject   -qipa=threads 

Peak Optimization Flags

C benchmarks:

104.milc:  basepeak = yes 
122.tachyon:  basepeak = yes 

C++ benchmarks:

126.lammps:  basepeak = yes 

Fortran benchmarks:

107.leslie3d:  basepeak = yes 
113.GemsFDTD:  basepeak = yes 
129.tera_tf:  basepeak = yes 
137.lu:  basepeak = yes 

Benchmarks using both Fortran and C:

115.fds4:  basepeak = yes 
121.pop2:  basepeak = yes 
127.wrf2:  basepeak = yes 
128.GAPgeofem:  basepeak = yes 
130.socorro:  basepeak = yes 
132.zeusmp2:  basepeak = yes 

The flags files that were used to format this result can be browsed at
http://www.spec.org/mpi2007/flags/MPI2007_flags.20080828.html,
http://www.spec.org/mpi2007/flags/MPI2007_flags.0.20080828.html,
http://www.spec.org/mpi2007/flags/MPI2007_flags.1.html.

You can also download the XML flags sources by saving the following links:
http://www.spec.org/mpi2007/flags/MPI2007_flags.20080828.xml,
http://www.spec.org/mpi2007/flags/MPI2007_flags.0.20080828.xml,
http://www.spec.org/mpi2007/flags/MPI2007_flags.1.xml.