Whetstone Benchmark – Historical Performance Test
I continue my work on reviewing old cross-platform performance tests, the first article is about an integer benchmark Dhrystone.
The test was developed by Harold Curnow (CCTA) in 1972 in Algol-60, followed by a Fortran implementation in 1973, and Roy Longbottom’s C implementation in 1996. The test is very simple and includes about 150 instructions with eight active loops, three of which are executed using procedure calls.
The dominant loop, which typically accounts for 30-50% of the time, performs floating point calculations through procedure calls. Performance estimates are expressed in millions of Whetstone operations per second (MWIPS). For better performance evaluation, the processor must have an FPU unit.
Whetstone Test Structure
The test consists of 8 sections:
N1 floating point: Test of arithmetic over floating point arrays;
N2 floating point: Test of arithmetic over floating point arrays, which are passed as parameters to the function;
N3 if then else: Conditional jump test;
N4 fixed point: Integer arithmetic;
N5 sin, cos etc.: Test of trigonometric functions;
N6 floating point: Test of calling procedures with passing parameters through pointers;
N7 assignments: Assignments;
N8 exp, sqrt etc.: Test of other mathematical functions (root, exponent, logarithm).
Advantages of the test:
Small amount of code;
It is possible to compare the results of processors of different architectures;
Easy to compile and port (but requires FPU).
Flaws:
Does not allow you to use all the capabilities of the processor;
The results are highly dependent on compiler optimization;
Placed in the processor cache (reduces the load on the memory subsystem).
Output of the Whetstone benchmark result
Benchmark output example (MWIPS is the main performance indicator)
##############################################
Whetstone Single Precision C Benchmark amd64 _ optimized, Fri Jan 31 22:19:52 2020
Loop content Result MFLOPS MOPS Seconds
N1 floating point -1.12475013732910156 796.837 0.081
N2 floating point -1.12274742126464844 627.771 0.721
N3 if then else 1.00000000000000000 5177.749 0.067
N4 fixed point 12.00000000000000000 3237.101 0.328
N5 sin,cos etc. 0.49911010265350342 81.010 3.461
N6 floating point 0.99999982118606567 647.448 2.808
N7 assignments 3.00000000000000000 1752.687 0.355
N8 exp,sqrt etc. 0.75110864639282227 57.663 2.174
MWIPS 3371.316 9.996
There is a multi-threaded variation of the test (Whetstone MP)
I ported the Whetstone benchmark to other programming languages: Python, JavaScript, PHP, Lua, C#, Java. The code can be viewed here: https://github.com/entityfx/entityfx-bench.
Table with test results for some processors
MWIPS MFLOP MFLOP MFLOP COS EXP FIXPT IF EQUAL
CPU MHz 1 2 3 MOPS MOPS MOPS MOPS MOPS
AM386/387 40 5.68 0.928 0.884 0.673 0.461 0.275 2.36 2.16 0.638
80486DX2 66 15.3 4.92 3.59 2.38 0.501 0.320 6.18 5.91 5.32
AMD 5X86 100 25.0 7.70 6.04 4.05 0.806 0.522 9.52 9.37 7.83
Winchip C6 200 40.6 16.9 12.3 7.81 1.02 0.71 18.4 28.4 17.0
Pentium 75 48.2 19.6 12.6 7.39 1.86 1.12 11.9 14.6 16.7
Cyrix P150 120 53.5 14.1 11.6 7.55 2.07 1.36 18.3 30.0 12.9
IBM C6x86 150 66.1 17.3 14.3 9.32 2.56 1.68 22.6 36.9 15.9
Pentium 100 66.2 27.1 17.3 10.2 2.56 1.53 16.3 19.9 23.1
Apple G3 266 66.5 17.3 13.7 11.4 2.73 1.13 39.6 25.0 10.9
Pentium 120 79.5 32.4 20.8 12.2 3.07 1.83 19.5 24.1 27.7
Cyrix PR233M 188 88.0 24.3 19.5 12.8 3.28 2.16 28.5 48.4 23.4
Pentium 133 88.3 36.1 23.1 13.6 3.41 2.03 21.7 26.5 30.8
Cyrix MII300 233 109 30.6 24.0 15.8 4.09 2.70 35.7 59.4 29.4
Pentium 166 109 44.6 28.5 16.8 4.21 2.50 26.9 33.0 38.0
Pentium MMX 166 112 45.0 28.9 17.0 4.28 2.53 28.4 38.1 38.6
AMD K6 200 124 46.3 29.8 16.9 5.54 2.53 68.6 25.2 35.4
Pentium 200 132 54.2 34.7 20.3 5.12 3.03 32.5 39.8 46.2
Pentium MMX 200 134 54.2 34.3 20.3 5.13 3.03 34.0 46.2 45.8
Pentium Pro 200 161 50.3 45.2 31.5 4.46 2.77 102 20.6 119
AMD K6 266 167 65.9 41.4 22.9 7.48 3.36 92.0 33.0 47.0
Pentium Pro 233 189 58.9 51.3 36.7 5.36 3.30 119 24.0 116
AMD K6 300 191 75.5 47.1 26.2 8.51 3.83 105 37.3 53.4
Pentium II 233 191 58.8 51.9 36.9 5.37 3.34 119 24.3 141
Pentium II 266 218 67.3 59.6 42.1 6.12 3.81 137 27.4 160
Celeron A 300 245 76.2 66.7 47.4 6.91 4.28 153 31.0 181
Pentium II 300 245 75.7 66.9 47.4 6.90 4.28 152 31.1 180
AMD K63 450 286 113 70.8 39.2 12.8 5.74 157 56.2 79.7
Pentium II 350 286 88.1 78.6 55.4 8.06 4.99 178 36.1 212
AMD K62 500 309 122 76.4 42.3 13.8 6.20 170 61.0 86.0
Pentium III 450 367 114 101 71.2 10.3 6.42 231 46.6 271
Pentium II 450 368 115 100 71.3 10.4 6.41 230 46.5 269
Athlon 500 381 98.9 91.6 60.1 10.9 8.34 265 115 128
Celeron 466 382 118 103 73.8 10.8 6.66 240 48.6 279
Pentium III 550 448 139 122 86.8 12.6 7.83 279 56.9 331
Duron 600 463 121 111 73.1 13.3 10.2 321 140 156
Duron 700 541 140 131 85.3 15.5 11.8 377 164 181
Athlon 700 544 141 131 85.8 15.6 11.9 380 165 185
Atom M 1600 554 248 159 89.9 15.1 10.2 418 338 198
Pentium III 700 572 176 156 111 16.2 9.96 357 72.3 415
Celeron 733 598 185 162 116 16.9 10.4 373 75.6 437
Pentium 4 1700 603 204 166 104 21.3 12.1 369 71.1 155
Pentium 4 1800 639 216 176 110 22.6 12.8 391 75.2 164
Pentium 4 1900 671 227 184 116 23.6 13.4 409 79.4 172
Pentium 4 2052 726 245 200 125 25.6 14.5 443 85.5 187
Athlon Tbird 1000 769 200 185 121 22.0 16.8 532 233 260
Duron 1000 772 200 186 122 22.1 16.9 536 235 260
P4 Xeon 2200 773 260 213 134 26.9 16.2 469 85 198
Pentium III 1000 816 253 222 158 23.1 14.2 510 103 599
Atom Z8300 1840 954 432 292 162 26.5 15.9 986 256 417
PIII Tualatin 1200 972 304 268 188 27.4 16.9 604 123 715
Pentium 4E 3000 1028 313 262 154 38.7 18.5 824 217 298
Celeron M 1295 1034 321 301 202 29.0 17.6 661 132 762
Pentium 4 3066 1119 365 300 188 40.2 22.3 660 152 278
Athlon 4 1533 1193 308 284 186 34.6 26.3 833 358 398
Pentium 4 3382 1233 402 331 206 44.4 24.7 726 167 306
Pentium 4 3678 1342 436 357 224 48.3 26.8 791 184 333
Athlon 4 1789 1389 358 331 217 40.3 30.6 971 416 464
Ath4 Barton 1800 1397 348 333 227 40.3 30.4 957 357 447
Athlon XP 1865 1450 373 345 226 42.1 32.0 1010 437 484
Turion 64 M 1900 1506 375 346 245 42.3 32.1 1291 437 473
Pentium M 1862 1538 471 439 292 42.0 25.3 945 293 1113
Core 2 Duo M 1830 1557 431 437 293 42.8 26.2 1641 286 590
Opteron 1991 1580 393 364 255 44.7 33.8 1349 457 496
Celeron C2 M 2000 1688 487 472 315 46.9 28.4 1792 310 632
Athlon 64 2150 1720 427 395 280 48.5 36.6 1465 495 537
Athlon 64 2211 1766 439 406 286 49.9 37.6 1503 509 552
Athlon XP 2338 1805 457 424 295 51.5 39.7 1224 463 581
Core i5 2467M @@@@ 1813 537 501 317 50.4 32.1 2242 457 561
Core 2 Duo 1 CP 2400 2057 586 580 387 56.7 34.4 2192 381 771
Phenom II 3000 2145 594 492 297 67.1 50.7 2053 694 703
Core i7 930 **** 2496 691 671 441 72.0 43.9 2663 489 838
Core i7 860 #### 2790 793 752 499 81.2 49.5 2133 559 964
Core i7 3930K &&&& 3004 883 829 525 83.6 53.1 3715 749 936
Core i7 4820K $$$1 3063 887 831 533 85.2 53.7 4110 769 972
Core i7 4820K $$$2 3124 920 857 545 86.2 55.3 4149 782 979
Core i7 3930K OC 3722 1096 1017 649 104 65.7 4538 937 1166
#### Rated as 2800 MHz but running at up to 3466 MHz using Turbo Boost
**** Rated as 2800 MHz but running at up to 3066 MHz using Turbo Boost
@@@@ Rated as 1600 MHz but running at up to 2300 MHz using Turbo Boost
&&&& Rated as 3200 MHz but running at up to 3800 MHz using Turbo Boost
$$$1 Rated as 3700 MHz but running at up to 3900 MHz using Turbo Boost
$$$2 Performance not Balanced Power Setting for 3900 MHz
OC OverClocked ~4730 MHz
M Mobile CPU Links
PS
For Windows, the benchmark can be downloaded here: http://www.roylongbottom.org.uk/benchnt.zip. For Linux, compile from sources from my repository: https://github.com/EntityFX/anybench.
