Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!mtune!codas!usfvax2!pdn!alan
From: alan@pdn.UUCP
Newsgroups: comp.sys.mac
Subject: Re: Mac C Compilers, Benchmarks, Stupidity
Message-ID: <1054@pdn.UUCP>
Date: Tue, 18-Aug-87 17:45:20 EDT
Article-I.D.: pdn.1054
Posted: Tue Aug 18 17:45:20 1987
Date-Received: Fri, 21-Aug-87 01:42:02 EDT
References: <3560@cit-vax.Caltech.Edu> <20149@ucbvax.BERKELEY.EDU>
Reply-To: alan@pdn.UUCP (0000-Alan Lovejoy)
Distribution: world
Organization: Paradyne Corporation, Largo, Florida
Lines: 64

In article <20149@ucbvax.BERKELEY.EDU> hilfingr@tully.Berkeley.EDU.UUCP (Paul Hilfinger) quotes Robert Dewar:
>
>Date: Sun, 16 Aug 87 18:42:54 EDT
>From: dewar@acf2.nyu.edu
>
>
>... First I would expect an 80386
>to run faster than a 68020 at the same clock rate. The store overlap alone

At the same clock rate, same memory access speed, same bus speed and
same algorithm in hand-coded assembly, the 68020 averages twice the
speed of the '386 according to IEEE benchmarks--using the 68020's cache,
68020 opcodes and addressing modes and '386 opcodes and addressing
modes, and using 150ns DRAMS.  Using 45ns SRAMS, the '020 averages 30
per cent faster than the '386.  These are objective, reproducible facts.

>will improve things. Also I assume that the comparison was on a MAC II
>without the memory management chip, does this chip slow things down further?

Yes, the 68851 or 68461 MMUs engender one extra clock cycle per memory
fetch.  Score one for the '386.  

Remember also that all Mac II's come with either the 68461 or the 68851
as standard equipment, and that Sun, Apollo, Masscomp and others produce
machines using 68020/MMU CPU chip sets that are 2 to 3 times faster than
the Mac II, which has TWO wait states (120ns DRAMs).

However, consider the following chart of minimum cycle times for memory
accesses (assuming no MMU for the '020, built-in MMUs for the other
two):

                '386       '020      '030
data in cache    ---          2         1
data in ram        4          3         2

Considering the fact that the '030 has a data cache as well as an
instruction cache, its cycle time superiority is even more significant.

>If so the comparison is unfair in any case, since the 80386 has built in
>memory management. Of course the PS2/80 is a fairly poor implementation of
>the 80386 -- there are unconditional wait states. The DP386 is generally
>faster than the PS2/80, and faster designs with static RAM (e.g. the
>PC Limited design), are faster still.
>
>I am quite surprised that anyone would expect the 68020 to be faster than
>the 80386, or even as fast, where does this idea come from?
>

From people who have done professional, unbiased and complete benchmarks
of the two CPUs.

>[comments on dearth of good compilers for either CPU]

Microsoft C v5.0 is a VERY good compiler, certainly exceeding any
Mac compiler by a long shot.  My Modula-2 compiler for my Stride 440
(68000 12Mhz 1 wait state 120ns drams) produces better code than
any Macintosh compiler for any language (Sieve runs 10 times in
1.14 seconds, for example, which is better than the Mac II's time
as reported in Byte).

>[Robert Dewar 
>Net Address: dewar@acf2.nyu.edu ]
>
--Alan "Heard the latest rumor on the 68040, yet?" Lovejoy