Path: utzoo!attcan!uunet!bu.edu!rpi!batcomputer!riley
From: riley@batcomputer.tn.cornell.edu (Daniel S. Riley)
Newsgroups: comp.sys.atari.st
Subject: Re: 68030 speed increases (was Re: Atari TT 030 Launched!)
Keywords: Atari TT
Message-ID: <10379@batcomputer.tn.cornell.edu>
Date: 7 Jun 90 06:05:50 GMT
References: <1990Jun5.143231.4977@watserv1.waterloo.edu> <13266@wpi.wpi.edu> <81214@tut.cis.ohio-state.edu> <1990Jun6.044350.20403@cbnewsh.att.com> <10373@batcomputer.tn.cornell.edu> <3647@rodan.acs.syr.edu>
Reply-To: riley@tcgould.tn.cornell.edu (Daniel S. Riley)
Organization: Cornell Theory Center, Cornell University, Ithaca NY
Lines: 43

In article <3647@rodan.acs.syr.edu> jfbruno@rodan.acs.syr.edu (John Bruno) writes:
> >That wasn't mindless ravings.  Seriously, that 1/3 number confuses me--it
> >should be lots higher.  Going from a 16 bit to 32 bit machine and doubling
> >the clock speed (8 MHz to 16 MHz) should give you at least a factor of 4,
>
>How do you figure that???? I would expect a CPU with twice the clock rate to
>execute instructions twice as fast.  However, the difference between a 32
>bit bus and a 16 bit bus will not double your execution speed (unless every
>single instruction in the testing program operates on 32 bit objects). Sure
>there will be more speed improvement, but definitely NOT enough to double the
>speed.  

My "double the bus width, double the speed" was a bit glib.  I was assuming
that the TT has a decent 32-bit memory subsystem, and that the instruction
and data caches are enabled.  If those conditions are true, then a '30
should have at least a factor of two advantage over a 68000 of the same
speed.  If you turn off the cache, that drops.  If you turn off the cache 
and force it to use 16-bit memory, a '30 may well perform worse than a 68000 
at the same clock speed.  

It is true that, in general, a 32-bit bus machine isn't a priori twice as 
fast as a 16.  However, with the built-in 256-byte instruction and data 
caches and a pipelined architecture, a 68030 in practice is at least that 
much faster than a 68000, if it has decently fast 32-bit memory to talk to.

>You can always whip up some special optimal program that will run 4 times
>as fast with the above configuration. That is, if you just want to impress
>those people that like to say "MY machine does more Giga-hoozits than..."

Yeah, but I'm not talking about special cases.  On generic 68000 integer
code, with decent 32-bit memory and both caches enabled, a '30 should be
more than twice as fast as a 68000 of the same clock speed.  If we give
Atari the benefit of the doubt and assume a competent hardware design for
the TT, then it should be at least 4 times as fast as an 8 MHz 68000 running
generic 68000 code.  I'm just curious why they might not get this running
ST applications.

Anyway, as has been pointed out, we don't have real specs or benchmarks
for the TT, so this line of discussion is premature.  I'll shut up now
until some real specs are out.

-Dan Riley (riley@riley@tcgould.tn.cornell.edu, cornell!batcomputer!riley)
-Wilson Lab, Cornell University