Path: utzoo!utgpu!news-server.csri.toronto.edu!turing.toronto.edu!ruhtra
Newsgroups: comp.unix.questions
From: ruhtra@turing.toronto.edu (Arthur Tateishi)
Subject: Re: Cache vs. Mhz
Message-ID: <1990Aug1.155331.5528@jarvis.csri.toronto.edu>
Organization: CSRI, University of Toronto
References: <1990Jul25.030258.11568@cunixf.cc.columbia.edu> <1366@sixhub.UUCP>
Date: 1 Aug 90 19:53:31 GMT
Lines: 27

In article <1366@sixhub.UUCP> davidsen@sixhub.UUCP (Wm E. Davidsen Jr) writes:
>
>In article <1990Jul25.030258.11568@cunixf.cc.columbia.edu> chaiklin@cunixf.cc.columbia.edu (Seth Chaiklin) writes:
>| 
>| I have a dilemma.  I must take either a 25 Mhz 386 machine
>| with no cache or a 20 Mhz 386 machine with a 64K cache.
>
>For almost any other application I would go with the 20MHz and cache.
>The cache will give you about 15% improvement with a 1w/s memory. I
>would be very sure whats happening if the 25MHz machine claims to be
>0w/s, unless it is running fast interleaved memory.

This touches on a choice I made a few months ago. I bought a 20MHz
386 (AMI, C&T) because it was reported to be 0w/s. All the 25-33MHz machines
around had caches and up here the cache basicly added $600 to the price. 
I opted for a 20MHz 386+387 combination since I figured the 387 would be
more useful to me than a cache. Later I learn about memory pre-charge time
and started to question the meaning of 0w/s. Recently, I have also heard
about pipelined memory accesses( does this eliminate pre-charge delays?).
I'd like to get a proper explanation of what a cache does to improve the
performance of a machine with 0w/s memory.
I currently have 4MB RAM (100ns) set to 0w/s.

-- 
``Sex and drugs? They're nothing compared with a good proof!''
                                - A Cambridge student  (r.h.f)
Arthur Tateishi                 g9ruhtra@zero.cdf.utoronto.edu