the rulerMarks and flowerMarks page

rulerMarks 97 and flowerMarks 97

THESE ARE a couple of benchmark routines I use to test new machines to gauge how fast they are (hey, techies are just like that, OK?)

They make no claim to test processors thoroughly, and no attempt has been made to equalise other factors across these widely different platforms (though all machines except the Acorns ran some sort of Unix). But they do have one advantage over many benchmarks: they are both programs I've actually wanted to use and which I found horribly slow on the machines I had at the time (the ones with benchmarks defined as 1).

rulerMarks 97 contains only integer code, whereas flowerMarks 97 contains mostly floating-point. The flowerMarks benchmark has quite a large dataset and so likes big caches and fast memory systems (see P2/400 vs Celeron 500) whereas rulerMarks and its data will fit in L1 cache on any modern CPU.

OK, I admit it, not all these cool toys are on my desk: the Alpha and Origin 2000 results were contributed by an astrophysicist friend of mine... all the nice Sun kit, the nicer HP and the Origin 200 (which is dual-processor, though the benchmark only exercises one) are on my brother's desk though.

Results

Machine	CPU	MHz	rulerMarks (Integer)		flowerMarks (FP)
Machine	CPU	MHz	Time/s	rM 97	Time/s	fM 97
Acorn A540	ARM3	25	*207.11	def 1
Acorn RiscPC	ARM6	33	146.25	1.4	819.00	def 1
Acorn NC	ARM7500FE	40	112.83	1.8	46.64	17.6
Acorn RiscPC	StrongARM	200	17.26	12.0	86.30	9.5
	i386SX	25	1,176.85	0.2
	i486DX2	66	53.88	3.8	31.11	26.3
Dan	Pentium	90	24.38	8.5	11.80	69.2
	Pentium	133	23.96	8.6	5.95	137.6
	Pentium MMX	233	9.90	21.0	3.43	239.0
AST	Pentium Pro	200	10.70	19.4	3.17	258.0
Dell	Pentium II	266	7.95	26.0	2.32	353.0
	Pentium II	350	6.07	33.8	1.74	470.7
	Pentium II	400	5.43	38.1	1.56	525.0
Dell	Celeron	500	6.47	32.0	1.57	521.7
	Pentium III e	600	4.39	47.1	1.11	737.8
Dell	Pentium III eb	733	3.79	54.6	0.72	1,137.5
Dell	Pentium III e	800	**2.92	70.9	**0.81	1,009.9
	Pentium III eb	1,000	2.43	85.2	0.71	1,153.5
	Pentium IV Xeon	3,066	1.66	125.5	0.21	3,900.0
Apple Mac Pro	Xeon 5160	3,000	0.46	450.0	0.07	11,700.0
	Athlon T'bird	850	**2.45	84.5	**0.74	1,105.3
	Athlon 2200+	1,800	**1.18	175.5	**0.39	2,100.0
	Athlon XP 2500+	1,833	1.55	134.0	0.31	2,600.0
	Athlon XP 3200+	2,200	1.35	153.5	0.24	3,400.0
	Athlon64 X2 4400+	2,200	1.47	140.9	0.15	5,460.0
	Opteron 270	2,000	0.98	211.0	0.17	4,800.0
	Opteron 250	2,400	0.72	288.0	0.15	5,460.0
IBM RS6000	PowerPC 604e	233	7.44	27.8	3.27	250.5
Apple Imac	PowerPC G4	700	2.82	73.4	0.91	900.0
Apple Macintosh	PowerPC G5	1,800	1.18	175.5	0.26	3,150.0
SparcStation 5	microSparc	70	*42.60	4.9
SparcStation 20	superSparc	50	*43.18	4.8
Ultra 1	ultraSparc ?	170	16.51	12.5	3.97	206.3
Ultra 5	ultraSparc IIi	270	8.24	25.1	2.25	364.0
Ultra 10	ultraSparc IIi	360	4.66	44.4	1.40	585.9
Ultra E250	ultraSparc IIi	400	4.49	46.1	1.27	644.9
HP 9000 D220	PA-7300LC	132	19.86	10.4	4.17	196.4
HP Visualize B180L	PA-7300LC	180	14.59	14.2	2.87	285.0
DECstation 3000	Alpha 21064	125	*32.31	6.4
DEC Multia	Alpha 21066	166	44.58	4.6	22.51	36.4
	Alpha 21164	300			5.81	141.0
	Alpha 21164	600	*6.20	33.4
SGI O2	MIPS R10000	250	8.13	25.5	1.72	476.2
SGI Origin 200	MIPS R10000	180	9.95	20.8	1.67	490.4
SGI Origin 2000	MIPS R10000	195	11.91	17.4	1.22	677.0

def 1: defined as 1. All other benchmarks are relative to this system.

* interpolated from rulerMarks 95 times (a very similar but less optimised ruler program). Conversion factor is t₉₇ = t₉₅/3.404

** Windows machines with Visual C++ 6

The benchmark programs

rulerMarks is (part of) a program which calculates the longest Golomb ruler achievable with eleven marks. An ordinary ruler lets you measure one unit in many places: for instance you can measure 1cm between 0cm and 1cm, between 1cm and 2cm, and so on. Golomb's question was, how few marks can you get away with and still be able to measure every integer distance up to, say, 20? Or, put another way, what distance can you get up to with, say, 11 marks? The best brute-force search method for answers to this, involves lots of bitwise logical operators.

flowerMarks is a colour quantisation program, which takes a 24bpp image input and calculates the optimal 16-entry palette. I didn't write this program! -- the bit that does all the work comes from a program called GIMP, which is GPL'd. There's lots of hairy FP code in it for taking a collection of weighted points in 3-space and finding the vector of minimum angular momentum. Or something.

Downloading and running

Just click here [350K], which gives you a gzipped tar file. ruler97 and flower97 are each single C files, which are ANSI-compliant and should compile on just about anything. Most of the archive is taken up with the input file used by flower97: if you're interested, it's a RiscOS 32bpp sprite file of a photo I took of a field of sunflowers somewhere near Vouvray in France.

These programs have not been tested on big-endian machines, but should work. If flower97 reports "image size 464x324" and "60246 colours found" then everything is probably fine.

All Rites Reversed -- Copy What You Like