[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference hydra::amiga_v1

Title:AMIGA NOTES
Notice:Join us in the *NEW* conference - HYDRA::AMIGA_V2
Moderator:HYDRA::MOORE
Created:Sat Apr 26 1986
Last Modified:Wed Feb 05 1992
Last Successful Update:Fri Jun 06 1997
Number of topics:5378
Total number of notes:38326

5289.0. "FPU speed & performance" by LEMAN::BUSSINK (Think Positive and Enjoy Life) Thu Dec 19 1991 05:54

    Could someone tell me how well the FPU are used on the Amiga. Will
    they be used with LHar'ing and programs like ADpro ?
    What is the difference between 68881/68882. 
    How is the performance related to the speed ?
    	(I know that a 68882 at 33mhz gives ~1.78 Mflops) Is there a list
    	somewhere ?
    
    			Cheers,
    				Erik (also known as TRUCKS::BUSSINK_E
    				      during the last summer.)
    						
T.RTitleUserPersonal
Name
DateLines
5289.1PIANST::JANZENThomas MLO21-4/E10 223-5140Thu Dec 19 1991 08:5316
	I don't have accelerator, but my reading tells me:
	LHARC would not use an accelerator; I have written Huffman encoding
	routines, and they aren't floating point.
	I don't not what ADpro is.
	The bigest gainers in floating point accelerators are
	rendering programs, but they must be ready to use the accelerator.
	Either they must have dual versions (you used to have to mail away
	for a special version of Sculpt 3D that would use an accelerator),
	or use the IEEE libraries, which will figure it out on the fly,
	whether the machine has an acceleartor and use it if possible.
	DKBtrace does this (freely redistributable on fish 513/514).
	I think the gain, depending on the clock rate of the accelerator
	board you buy, is around 3X for floating point.
	I have been looking into this, and unless I buy used, it looks
	like 250-400 $, so I am postponing for now.
Tom
5289.2FPU performance varies greatlyWHAMMY::268For three strange days...Thu Dec 19 1991 10:3035
Floating point accelerators vary widely in performance, depending on
the chip and on the clock speed.  Machines without them will end up
emulating the FP calculations using the CPU, and that is very slow on
a M68000.

                                        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           ??
	Amiga 25Mhz 68040               ~1-2 (estimated)

x86:
  	DECstation 425 (25Mhz 486)	3

MIPS:
	DECstation 5000/200		~1

As you can see, even a 14Mhz 68881 is about 10x faster than the "stock" Amiga.
The 25Mhz 882 is about 20x faster than the "stock" Amiga.  From various
magazine articles, it's been stated that the FPU in the M68040 is approx.
10-12x faster than an 882 at the same clock speed.  

That would put the 040s FPU at around 220x faster than a "stock" A500/2000 
without any type of FPU.  The talk on usenet is consistent with the 
magazine results - the 040 really screams for floating point, and is 
considerably faster than the 486 in this regard.  And the 486 is certainly
no slouch.

If you're doing floating intensive work, consider an 040.  Since time is
money, think of all the money you'll be saving. :^) :^)  

Steve