Conference hydra::amiga_v1

2683.0. "CPU performance numbers" by WJG::GUINEAU () Mon Jun 26 1989 09:57

Following are benchmark times recieved for 3 different classes of CPU:

	68K (Amiga's)
	VAX (various models)
	IBM (Clones - various)

The following benchmarks (and thier listed variations) were run:

	sieve - Integer test from this months Amazing Computing.
	Osieve - Optimized sieve

	whets - floating point test which I converted from Fortran.
	whets881 - Compiled with specific 68881 support (1.3 libs)
	Owhets881 - Optimized 
	whets020881 - Compiled with specific 68020/68881 support


881, 020881 and Optimized versions run on Amiga only.


AMIGA Notes
-----------
	All compiles done with Lattice C 5.02. 
	Optimization was done using the Lattice Global Optimizer (-O option).
	881 support using the 1.3 libraries (-Lm+lib:lcmieee.lib option)
	020 support using the -m2 option.

	I've since discovered another Lattice option, -f8, which supposedly
	uses thier lcm881.lib library for inline 881 stuff. I tried it and
	halved the time on a Midget Racer based system. However, I later
	discovered that there was something seriously wrong with this
	code since it's accuracy was gone (like went from 10^-9 to 10^+358!).
	I'm still investigating this. It looks like (believe it or not) a
	Midget timing problem.

VAX Notes
---------
	Nothing special here. Just a standard CC compile.

IBM notes
---------
	First let me express my explicit *hate* for these beasts. I had
	a real time getting this thing to compile (Ask Ed!). Other than
	that, nothing special (but then neither are IBM machines :-).
	Microsoft C 5.1.


Lastly, I'd like to thank everyone who ran these benchmarks. We now have
a good set of relative numbers over lots of machines. 

BTW - I'l like to state a bit of a disclaimer on this stuff. I didn't
write these benchmarks and claim no responsibility for thier accuracy or
applicability toward CPU speed measurement. They just give us some
*relative* numbers across several CPU's.


-------------------------------------
Here are the numbers (no speculation provided :-)	


SIEVE.C	(integer test)
------------------------------------------------------
					Osieve	sieve
68K:	Amiga 7Mhz 68000     		27
	Amiga 68020 7MHZ (Midget Racer)	12
	Amiga 2500 (14Mhz 68020/881)	 5	 7
	Amiga 14 Mhz 68000 (CMI)        23	32
	Amiga GVP 68030/68882 25Mhz	 3	 4
	Amiga A2630 68030/881		??

VAX:	VAXstation 3600 (WJG::)		 4
	MicroVAX II (VMS)		13
	MicroVAX II (Ultrix)		14 
	VAX 8800 (LEDS3::)		 2
	VAX 8650 (LEDS2::)		 4
	VAX 8370 (SUBSYS::)		 9
	VAXstation 3520 (FireFox)	 5
	VAXstation 3100 (PVAX) 		 5

MIPS:	DECstation 3100 (PMAX)           2

IBM:	Compaq 286			31
	WEISS 386 8MHZ			31
	WEISS 386 16MHZ			11
	IBM PS/2 Model 80		11




WHETS.C	(floating point test)
------------------------------------------------------

					whets020881  Owhets881  whets881 whets
68K:	Amiga 7Mhz 68000     		                                 258
	Amiga 68020 7MHZ (Midget Racer)	47
 	Amiga 2500 (14Mhz 68020/881)	23           81         49       118
	Amiga 14 Mhz 68000 (CMI)        NA           58		80       288
	Amiga GVP 68030/881 25Mhz	13           12         36       59
	Amiga A2630 68030/881		??                                  
                                          
VAX:	VAXstation 3600 (WJG::)		12
	MicroVAX II (VMS)		25
	MicroVAX II (Ultrix)		27    
	VAX 8800 (LEDS3::)		4
	VAX 8650			14
	VAX 8370			22
	VAXstation 3520 (FireFox)	 8
	VAXstation 3100 (PVAX) 		10
                                          
MIPS:	DECstation 3100 (PMAX)          ??
                                          
IBM:	Compaq 286			378
	WEISS 386 8MHZ			356
	WEISS 386 16MHZ			159
	IBM PS/2 Model 80		??

Lattice global optimizer makes for good fast code.

I took the savage benchmark that came with the Midget Racer and ran the
compiled version which had been compiled by CSA using Manx Aztec C 3.6a (?)
and got 1.9 times a base A2000. I Then recompiled it with Lattice and the global
optimizer and got 2.2 times a A2000. Lattice wins!


881's really do work!

020 are fast, 030's are faster.

I*M 386 machines aren't that fast.

I want a VAX!

The benchmarks used are at

	WJG::AMIGA:SIEVE.ARC and WHETS.ARC

John	

	Does anyone have, or know where I might find, relative
	performance numbers for the 68030, 68040, and the Intel
	i486 ?  (I'm particularly interested in the '040 vs 486
	comparison.)

	Just some rough performance numbers would be welcome; I
	don't need actual benchmark suites.

	Thanks.
	~art