| CPU2006 license: | 11 | Test date: | Sep-2015 |
|---|---|---|---|
| Test sponsor: | IBM Corporation | Hardware Availability: | Oct-2015 |
| Tested by: | IBM Corporation | Software Availability: | Dec-2015 |
| Hardware | |
|---|---|
| CPU Name: | POWER8 |
| CPU Characteristics: | Intelligent Energy Optimization enabled, up to 3.49 GHz |
| CPU MHz: | 2926 |
| FPU: | Integrated |
| CPU(s) enabled: | 20 cores, 2 chips, 10 cores/chip, 8 threads/core |
| CPU(s) orderable: | 2 Modules |
| Primary Cache: | 32 KB I + 64 KB D on chip per core |
| Secondary Cache: | 512 KB I+D on chip per core |
| L3 Cache: | 8 MB I+D on chip per core |
| Other Cache: | 16 MB I+D off chip per 4 DIMMs |
| Memory: | 256 GB (32 x 8 GB DIMMs) DDR3 1333 MHz |
| Disk Subsystem: | 2 x 500 GB 15K RPM SAS SFF-2 Raid5 |
| Other Hardware: | None |
| Software | |
|---|---|
| Operating System: | Ubuntu 14.04 LTS Updated to 14.04.3 (ppc64le) kernel <3.16.0-46-generic> |
| Compiler: | C/C++: Version 13.1.3 of IBM XL C/C++ for Linux |
| Auto Parallel: | No |
| File System: | ext4 |
| System State: | Run level 3 (multi-user) |
| Base Pointers: | 64-bit |
| Peak Pointers: | 64-bit |
| Other Software: | Post-Link Optimization for Linux on POWER, version 5.6.2-6f IBM Advance Toolchain 8.0-3 |
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. | ||||||||||||||
| 400.perlbench | 80 | 1350 | 579 | 1248 | 626 | 1249 | 626 | 80 | 1167 | 670 | 1073 | 728 | 1075 | 727 |
| 401.bzip2 | 80 | 1163 | 664 | 1090 | 708 | 1089 | 709 | 120 | 1680 | 689 | 1607 | 720 | 1608 | 720 |
| 403.gcc | 80 | 712 | 905 | 712 | 904 | 712 | 904 | 140 | 1076 | 1050 | 1084 | 1040 | 1081 | 1040 |
| 429.mcf | 80 | 846 | 863 | 857 | 851 | 851 | 857 | 80 | 846 | 863 | 857 | 851 | 851 | 857 |
| 445.gobmk | 80 | 973 | 863 | 973 | 863 | 974 | 862 | 160 | 1740 | 965 | 1740 | 965 | 1742 | 964 |
| 456.hmmer | 80 | 552 | 1350 | 552 | 1350 | 563 | 1330 | 80 | 479 | 1560 | 477 | 1570 | 478 | 1560 |
| 458.sjeng | 80 | 1355 | 714 | 1358 | 713 | 1357 | 713 | 80 | 1262 | 767 | 1251 | 774 | 1261 | 768 |
| 462.libquantum | 80 | 1153 | 1440 | 1153 | 1440 | 1156 | 1430 | 120 | 242 | 10300 | 241 | 10300 | 241 | 10300 |
| 464.h264ref | 80 | 1452 | 1220 | 1452 | 1220 | 1452 | 1220 | 80 | 1348 | 1310 | 1368 | 1290 | 1354 | 1310 |
| 471.omnetpp | 80 | 1044 | 479 | 1044 | 479 | 1044 | 479 | 160 | 1599 | 626 | 1598 | 626 | 1598 | 626 |
| 473.astar | 80 | 824 | 682 | 828 | 678 | 828 | 678 | 80 | 824 | 682 | 828 | 678 | 828 | 678 |
| 483.xalancbmk | 80 | 617 | 895 | 608 | 908 | 607 | 910 | 80 | 579 | 954 | 568 | 972 | 569 | 970 |
Peak Tuning Notes
400.perlbench fdpr options: -O4 -m power8 -A 2 -sls -dir -vrox 401.bzip2 fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 403.gcc fdpr options: -O4 -m power8 -A 2 -sls -dir -vrox 429.mcf fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 456.hmmer fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 458.sjeng fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 462.libquantum fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 464.h264ref fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 471.omnetpp fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox 473.astar fdpr options: -O4 -m power8 -A 2 -rcl 2 -sls -dir -vrox
Submit Notes
The config file option 'submit' was used to assign benchmark copy to specific kernel thread using the "numactl" command (see flags file for details).
Operating System Notes
ulimit -s (stack) set to unlimited 8000 16M large pages defined Transparent huge page disabled with echo never > /sys/kernel/mm/transparent_hugepage/enabled sysctl vm.nr_hugepages=N and reboot to set large page pool
General Notes
Environment variables set by runspec before the start of the run: HUGETLB_MORECORE = "yes" HUGETLB_VERBOSE = "0" LD_LIBRARY_PATH = "$LD_LIBRARY_PATH:/opt/ibm/fdprpro/lib" TCMALLOC_MEMFS_MALLOC_PATH = "/dev/hugepages/" XLFRTEOPTS = "intrinthds=1" This result uses the post_setup and/or bench_post_setup to drop caches. SPEC has determined that although the effect may have been negligible for this run, future submissions will not be considered rule compliant if the post_setup actions drop caches (e.g. : "echo 3 > /proc/sys/vm/drop_caches").
Base Compiler Invocation
C benchmarks:
| /opt/ibm/xlC/13.1.2/bin/xlc_at -qlanglvl=extc99 |
C++ benchmarks:
| /opt/ibm/xlC/13.1.2/bin/xlC_at |
Base Portability Flags
| 400.perlbench: | -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX |
| 401.bzip2: | -DSPEC_CPU_LP64 |
| 403.gcc: | -DSPEC_CPU_LP64 |
| 429.mcf: | -DSPEC_CPU_LP64 |
| 445.gobmk: | -DSPEC_CPU_LP64 |
| 456.hmmer: | -DSPEC_CPU_LP64 |
| 458.sjeng: | -DSPEC_CPU_LP64 |
| 462.libquantum: | -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX |
| 464.h264ref: | -DSPEC_CPU_LP64 -qchars=signed |
| 471.omnetpp: | -DSPEC_CPU_LP64 |
| 473.astar: | -DSPEC_CPU_LP64 |
| 483.xalancbmk: | -DSPEC_CPU_LP64 -DSPEC_CPU_LINUX |
Base Optimization Flags
C benchmarks:
| -qinline=40 -qipa=threads -O5 -q64 -qalias=noansi -lhugetlbfs |
C++ benchmarks:
| -qinline=40 -qipa=threads -O5 -q64 -qrtti -D__extern_always_inline=inline -ltcmalloc |
Base Other Flags
C benchmarks:
| -qipa=noobject -qsuppress=1500-036 |
C++ benchmarks:
| -qipa=noobject -qsuppress=1500-036 |
Peak Compiler Invocation
C benchmarks:
| /opt/ibm/xlC/13.1.2/bin/xlc_at -qlanglvl=extc99 |
C++ benchmarks:
| /opt/ibm/xlC/13.1.2/bin/xlC_at |
Peak Portability Flags
Same as Base Portability Flags
Peak Optimization Flags
C benchmarks:
| 400.perlbench: | -qinline=40 -qpdf1(pass 1) -qpdf2(pass 2) -O3 -qarch=auto -qtune=auto -q64 -qfdpr -qalias=noansi -lhugetlbfs -Wl,-q |
| 401.bzip2: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O4 -qsimd=noauto -q64 -qfdpr -lhugetlbfs -Wl,-q |
| 403.gcc: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O4 -q64 -qfdpr -lhugetlbfs -Wl,-q |
| 429.mcf: | basepeak = yes |
| 445.gobmk: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O5 -q64 -lhugetlbfs |
| 456.hmmer: | -qinline=40 -qipa=threads -O5 -q64 -qassert=refalign -qfdpr -lhugetlbfs -Wl,-q |
| 458.sjeng: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O3 -qarch=auto -qtune=auto -q64 -qprefetch=dscr=0x54 -qfdpr -lhugetlbfs -Wl,-q |
| 462.libquantum: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O5 -qsimd=noauto -q64 -qnothreaded -qfdpr -lhugetlbfs -Wl,-q |
| 464.h264ref: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O5 -q64 -qfdpr -lhugetlbfs -Wl,-q |
C++ benchmarks:
| 471.omnetpp: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O5 -qsimd=noauto -q64 -qarch=pwr8 -qtune=pwr8 -qprefetch=dscr=0x54 -qfdpr -qrtti -D__extern_always_inline=inline -lhugetlbfs -Wl,-q -ltcmalloc |
| 473.astar: | basepeak = yes |
| 483.xalancbmk: | -qinline=40 -qipa=threads -qpdf1(pass 1) -qpdf2(pass 2) -O3 -qarch=auto -qtune=auto -qsimd -qprefetch=dscr=0x93 -qipa=partition=large -D__extern_always_inline=inline -lhugetlbfs -ltcmalloc |
Peak Other Flags
C benchmarks (except as noted below):
| -qsuppress=1586-476(pass 2) -qipa=noobject -qsuppress=1500-036 | |
| 400.perlbench: | -qsuppress=1586-476(pass 2) -qsuppress=1500-036 |
| 429.mcf: | -qipa=noobject -qsuppress=1500-036 |
| 456.hmmer: | -qipa=noobject -qsuppress=1500-036 |
| 462.libquantum: | -qsuppress=1586-476(pass 2) -qsuppress=1500-036 |
C++ benchmarks (except as noted below):
| -qsuppress=1586-476(pass 2) -qipa=noobject -qsuppress=1500-036 | |
| 473.astar: | -qipa=noobject -qsuppress=1500-036 |