| CPU2006 license: | 11 | Test date: | Jan-2010 |
|---|---|---|---|
| Test sponsor: | IBM Corporation | Hardware Availability: | Feb-2010 |
| Tested by: | IBM Corporation | Software Availability: | Dec-2009 |
| Hardware | |
|---|---|
| CPU Name: | POWER7 |
| CPU Characteristics: | Intelligent Energy Optimization enabled, up to 3.64 GHz |
| CPU MHz: | 3300 |
| FPU: | Integrated |
| CPU(s) enabled: | 32 cores, 4 chips, 8 cores/chip, 4 threads/core |
| CPU(s) orderable: | 32 cores |
| Primary Cache: | 32 KB I + 32 KB D on chip per core |
| Secondary Cache: | 256 KB I+D on chip per core |
| L3 Cache: | 4 MB I+D on chip per core |
| Other Cache: | None |
| Memory: | 256 GB (32x8 GB) DDR3 1066 MHz |
| Disk Subsystem: | 8x146.8 GB SAS SFF 15K RPM |
| Other Hardware: | None |
| Software | |
|---|---|
| Operating System: | SUSE Linux Enterprise Server 11 (ppc64), Kernel 2.6.27.19-5-ppc64 |
| Compiler: | IBM XL C/C++ for Linux, V10.1 Updated with the Oct2009 PTF IBM XL Fortran for Linux, V12.1 Updated with the Oct2009 PTF |
| Auto Parallel: | No |
| File System: | ext3 |
| System State: | Run level 3 (multi-user) |
| Base Pointers: | 32-bit |
| Peak Pointers: | 32/64-bit |
| Other Software: | -Post-Link Optimization for Linux on POWER, Version 5.5.0-1 -MicroQuill SmartHeap 9 |
Results Table
| Benchmark | Base | Peak | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Copies | Seconds | Ratio | Seconds | Ratio | Seconds | Ratio | Copies | Seconds | Ratio | Seconds | Ratio | Seconds | Ratio | |
| Results appear in the order in which they were run. Bold underlined text indicates a median measurement. | ||||||||||||||
| 410.bwaves | 128 | 2296 | 758 | 2302 | 756 | 2288 | 760 | 32 | 528 | 823 | 527 | 826 | 529 | 823 |
| 416.gamess | 128 | 2709 | 925 | 2711 | 924 | 2712 | 924 | 128 | 2621 | 956 | 2686 | 933 | 2646 | 947 |
| 433.milc | 128 | 2364 | 497 | 2366 | 497 | 2367 | 496 | 64 | 1101 | 534 | 1101 | 534 | 1101 | 534 |
| 434.zeusmp | 128 | 1496 | 779 | 1493 | 780 | 1493 | 780 | 128 | 1052 | 1110 | 1067 | 1090 | 1062 | 1100 |
| 435.gromacs | 128 | 1248 | 732 | 1249 | 732 | 1245 | 734 | 128 | 1139 | 802 | 1138 | 803 | 1138 | 803 |
| 436.cactusADM | 128 | 1842 | 831 | 1844 | 829 | 1847 | 828 | 32 | 368 | 1040 | 367 | 1040 | 366 | 1040 |
| 437.leslie3d | 128 | 2301 | 523 | 2301 | 523 | 2298 | 524 | 32 | 504 | 597 | 502 | 599 | 502 | 600 |
| 444.namd | 128 | 857 | 1200 | 851 | 1210 | 850 | 1210 | 128 | 838 | 1220 | 840 | 1220 | 838 | 1220 |
| 447.dealII | 128 | 843 | 1740 | 841 | 1740 | 839 | 1740 | 128 | 800 | 1830 | 805 | 1820 | 803 | 1820 |
| 450.soplex | 128 | 2153 | 496 | 2129 | 501 | 2126 | 502 | 32 | 500 | 534 | 446 | 598 | 441 | 604 |
| 453.povray | 128 | 665 | 1020 | 663 | 1030 | 660 | 1030 | 128 | 497 | 1370 | 490 | 1390 | 491 | 1390 |
| 454.calculix | 128 | 1375 | 768 | 1367 | 773 | 1358 | 777 | 128 | 1298 | 814 | 1289 | 819 | 1292 | 817 |
| 459.GemsFDTD | 128 | 3309 | 410 | 3305 | 411 | 3308 | 411 | 64 | 1588 | 428 | 1589 | 427 | 1589 | 427 |
| 465.tonto | 128 | 1968 | 640 | 1969 | 640 | 1965 | 641 | 128 | 1191 | 1060 | 1192 | 1060 | 1182 | 1070 |
| 470.lbm | 128 | 3896 | 451 | 3900 | 451 | 3902 | 451 | 32 | 940 | 468 | 939 | 468 | 938 | 469 |
| 481.wrf | 128 | 1761 | 812 | 1752 | 816 | 1754 | 815 | 128 | 1637 | 873 | 1636 | 874 | 1638 | 873 |
| 482.sphinx3 | 128 | 3184 | 784 | 3167 | 788 | 3182 | 784 | 128 | 3192 | 782 | 3142 | 794 | 3139 | 795 |
Submit Notes
The config file option 'submit' was used.
Operating System Notes
ulimit -s (stack) set to 1048576.
Large pages reserved as follows by root user:
echo 8448 > /proc/sys/vm/nr_hugepages
System configured with libhugetlbfs library for application access to large pages
Environment variables set before executing benchmarks.
export HUGETLB_VERBOSE=0
export HUGETLB_MORECORE=yes
export XLFRTEOPTS=intrinthds=1
General Notes
The "IBM Power 750 Express (3.3 GHz)" and "IBM Power 755 (3.3 GHz)"
are electronically equivalent. The results have been measured on the
"IBM Power 755 (3.3 GHz)"
IBM Post-Link optimization tool with
options "-O4 -omullX -see 0 -m power6" used for
433.milc 435.gromacs 436.cactusADM 482.sphinx3
options "-O4 -omullX -see 1" used for
436.cactusADM
options "-O4 -omullX -see 1 -ihf -1" used for
453.povray
options "-O4" used for
465.tonto
Whenever option "-omullX" was used during the optimization phase,
option "-imullX" was also used during the instrumentation phase.
Benchmarks bound to a processor using numactl on the submit command.
See flags file for details on settings.
Base Compiler Invocation
C benchmarks:
| xlc -qlanglvl=extc99 |
C++ benchmarks:
| xlC |
Fortran benchmarks:
| xlf95 |
Benchmarks using both Fortran and C:
| xlc -qlanglvl=extc99 xlf95 |
Base Portability Flags
| 410.bwaves: | -qfixed |
| 416.gamess: | -qfixed |
| 434.zeusmp: | -qfixed |
| 435.gromacs: | -qfixed -qextname |
| 436.cactusADM: | -qfixed -qextname |
| 437.leslie3d: | -qfixed |
| 454.calculix: | -qfixed -qextname |
| 481.wrf: | -DNOUNDERSCORE |
| 482.sphinx3: | -qchars=signed |
Base Optimization Flags
C benchmarks:
| -O5 -qnoenablevmx -lhugetlbfs |
C++ benchmarks:
| -O5 -qrtti -qnoenablevmx -qstaticlink -Wl,--whole-archive /usr/lib/libhugetlbfs.a -Wl,--no-whole-archive |
Fortran benchmarks:
| -O5 -qsmallstack=dynlenonheap -qalias=nostd -qnoenablevmx -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT |
Benchmarks using both Fortran and C:
| -O5 -qnoenablevmx -qsmallstack=dynlenonheap -qalias=nostd -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT |
Base Other Flags
C benchmarks:
| -qipa=noobject -qipa=threads |
C++ benchmarks:
| -qipa=noobject -qipa=threads |
Fortran benchmarks:
| -qipa=noobject -qipa=threads |
Benchmarks using both Fortran and C:
| -qipa=noobject -qipa=threads |
Peak Compiler Invocation
C benchmarks:
| xlc -qlanglvl=extc99 |
C++ benchmarks:
| xlC |
Fortran benchmarks:
| xlf95 |
Benchmarks using both Fortran and C:
| xlc -qlanglvl=extc99 xlf95 |
Peak Portability Flags
| 410.bwaves: | -qfixed |
| 416.gamess: | -qfixed |
| 434.zeusmp: | -qfixed |
| 435.gromacs: | -qfixed -qextname |
| 436.cactusADM: | -qfixed -qextname |
| 437.leslie3d: | -qfixed |
| 454.calculix: | -qfixed -qextname |
| 481.wrf: | -DNOUNDERSCORE |
| 482.sphinx3: | -qchars=signed |
Peak Optimization Flags
C benchmarks:
| 433.milc: | -Wl,-q -O5 -qnoenablevmx -lhugetlbfs |
| 470.lbm: | -qpdf1(pass 1) -qpdf2(pass 2) -O3 -qarch=auto -qtune=auto -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT -q64 |
| 482.sphinx3: | -Wl,-q -qpdf1(pass 1) -qpdf2(pass 2) -O4 -lhugetlbfs |
C++ benchmarks:
| 444.namd: | -qpdf1(pass 1) -qpdf2(pass 2) -O5 |
| 447.dealII: | -O5 -qrtti -qnoenablevmx -qstaticlink -Wl,-z,muldefs -Wl,--whole-archive /usr/lib/libsmartheap.a -Wl,--no-whole-archive |
| 450.soplex: | -qpdf1(pass 1) -qpdf2(pass 2) -O4 -qstrict -lhugetlbfs |
| 453.povray: | -Wl,-q -qpdf1(pass 1) -qpdf2(pass 2) -O5 -lsmartheap |
Fortran benchmarks:
| 410.bwaves: | -O5 -qsmallstack=dynlenonheap -lhugetlbfs |
| 416.gamess: | -qpdf1(pass 1) -qpdf2(pass 2) -O5 -qalias=nostd -qnoenablevmx |
| 434.zeusmp: | -qpdf1(pass 1) -qpdf2(pass 2) -O3 -qarch=auto -qtune=auto -qxlf90=nosignedzero -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT |
| 437.leslie3d: | -O5 -qsmallstack=dynlenonheap -qnoenablevmx -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT |
| 459.GemsFDTD: | -O5 -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT -q64 |
| 465.tonto: | -Wl,-q -qpdf1(pass 1) -qpdf2(pass 2) -O5 -q64 -lsmartheap64 |
Benchmarks using both Fortran and C:
| 435.gromacs: | -Wl,-q -qpdf1(pass 1) -qpdf2(pass 2) -O5 -lhugetlbfs |
| 436.cactusADM: | -Wl,-q -qpdf1(pass 1) -qpdf2(pass 2) -O2 -qarch=auto -qtune=auto -qnostrict -lhugetlbfs |
| 454.calculix: | -O4 -B/usr/share/libhugetlbfs/ -tl -Wl,--hugetlbfs-link=BDT |
| 481.wrf: | -O5 -qnoenablevmx -q64 -lhugetlbfs |