Help: Benchmark Heaven II

What's this?

Benchmark Heaven II is a visualization tool for data that represent computer performance and characteristics. The performance data presented here is taken by me, the author, over several years using machines he obtained, or was granted access to.

One quick note: on this page, a system that has larger number in benchmark score means that it is a speedier system.

Basic Ideas

I have measured Dhrystone benchmark on all machines on the list. Additionally, Whetstone, HPCC STREAM, NAS Parallel Benchmarks (NPB) 3.3, SPEC CPU2000 and SPEC CPU2006 benchmarks have been measured on machines that are capable of running them.

With the exception of NAS Parallel Benchmarks OpenMP implementation and the STREAM benchmark, all the other benchmarks are single-threaded. The idea here is that the single-thread performance reflects the (micro)architectural features of the processor core, and is the basis for the discussion of performance characteristics of parallel and concurrent computer systems.

The power consumption of some systems are also measured along with the processing speed. The actual power consumption figures of historic computers are hard to find, but with the running computer, it's easy to measure one.

Field Descriptions

#cores, #threads
Physical and logical processor count.
Year
The first year the same type of the system went on market in Gregorian calendar.
MHz
Processor frequency. Reciprocal of cycle time.
Mtr
The transistor count of the processor in millions. Incomplete.
Pidle[W]
The power consumption of the system in Watts, when the system stays idle in the OS shell.
Pquarter[W], Phalf[W], P3q[W], Pfull[W]
The power consumption of the system, when the ¼, ½, ¾ and all of the processor cores in the system are running Dhrystone.
L1I[KB], L1D[KB], L2[KB], L3[KB]
The memory cache size in kilobytes.
Memory[MB]
The main storage size in megabytes. Usually it's an array of DRAM.
MemType
The type and speed of the main storage, e.g. DDR3-1600.
OS
The name and version of the operating system on the system such as IRIX 6.5.22m.
Compiler
The compiler that was used to compile the benchmark suites.
NPB-SER CFLAGS, NPB-SER FFLAGS, NPB-OMP CFLAGS, NPB-OMP FFLAGS
The compiler flags with which NAS Parallel Benchmarks 3.3 were compiled. The NPB-SER CFLAGS is also used to compile Dhrystone and Whetstone. The NPB-OMP CFLAGS is also used to compile STREAM.
CPU2000 OPTIMIZE, CPU2006 OPTIMIZE
The basic optimization flags with which the SPEC benchmarks were compiled.
DhrystoneMIPS, WhetstoneMIPS
The Dhrystone VAX MIPS score and Whetstone MIPS score. Dhrystone is a benchmark that plays around with strings. Whetstone is a floating-point benchmark but in my opinion it's not as good a benchmark as Dhrystone.
STREAM Copy [MB/s], STREAM Scale [MB/s], STREAM Add [MB/s], STREAM Triad [MB/s]
The HPCC memory bandwidth benchmark. I used OpenMP implementation because generally all processor cores have be used to saturate the memory subsystem.
STREAM mean
The harmonic mean of STREAM Copy [MB/s], STREAM Scale [MB/s], STREAM Add [MB/s] and STREAM Triad [MB/s].
SER bt.A, SER cg.A, SER dc.A, SER ep.A, SER ft.A, SER is.A, SER lu.A, SER mg.A, SER sp.A, SER ua.A
The NAS Parallel Benchmarks 3.3 serial implementation, in Mop/s. The NPB is a set of mostly floating-point oriented applications.
OMP bt.A, OMP cg.A, OMP dc.A, OMP ep.A, OMP ft.A, OMP is.A, OMP lu.A, OMP mg.A, OMP sp.A, OMP ua.A
The NAS Parallel Benchmarks 3.3 OpenMP implementation, in Mop/s. The number of logical core count is used as OMP_NUM_THREADS.
164.gzip, 175.vpr, 176.gcc, 181.mcf, 186.crafty, 197.parser, 252.eon, 253.perlbmk, 254.gap, 255.vortex, 256.bzip2, 300.twolf
The SPEC CINT2000 subbenchmarks in the SPEC CPU2000 benchmark suite. Requires around 256 MBytes of memory in 32-bit pointer system. These figures are speedups over the base system that is a variant of Sun Ultra2. Only a single instance of each benchmark is run at a time.
CINT2000
The geometric mean of the scores of SPEC CINT2000 subbenchmarks.
168.wupwise, 171.swim, 172.mgrid, 173.applu, 177.mesa, 178.galgel, 179.art, 183.equake, 187.facerec, 188.ammp, 189.lucas, 191.fma3d, 200.sixtrack, 301.apsi
CFP2000: The floating point side of the CPU2000 benchmark suite. These figures are speedups over the base system that is a variant of Sun Ultra2. Only a single instance of each benchmark is run at a time.
CFP2000
The geometric mean of the scores of SPEC CFP2000 subbenchmarks.
400.perlbench, 401.bzip2, 403.gcc, 429.mcf, 445.gobmk, 456.hmmer, 458.sjeng, 462.libquantum, 464.h264ref, 471.omnetpp, 473.astar, 483.xalancbmk
The integer subbenchmarks of SPEC CPU2006 benchmark suite. These require around one gigabyte of memory in 32-bit pointer system. In my configuration only a single instance of each benchmark is run at a time.
CINT2006
The geometric mean of the integer subbenchmarks of SPEC CPU2006.
410.bwaves, 416.gamess, 433.milc, 434.zeusmp, 435.gromacs, 436.cactusADM, 437.leslie3d, 444.namd, 447.dealII, 450.soplex, 453.povray, 454.calculix, 459.GemsFDTD, 465.tonto, 470.lbm, 481.wrf, 482.sphinx3
The floating point subbenchmarks of SPEC CPU2006 benchmark suite. These require around one gigabyte of memory in 32-bit pointer system. In my configuration only a single instance of each benchmark is run at a time.
CFP2006
The geometric mean of the floating point subbenchmarks of the SPEC CPU2006.