EnigmA Amiga Run 1997 July/EnigmA AMIGA RUN 20 (1997)(G.R. Edizioni)(IT)[!][issue 1997-07 & 08][EAR-CD IV].iso/lightwave/lwmlist/95.lightwave-09/000496_owner-lightwav…mail.webcom.com_Wed Sep 13 14:49:30 1995.msg

Internet Message Format | 1995-10-04 | 4KB

Received: by mail.webcom.com
    (1.37.109.15/16.2) id AA180948970; Wed, 13 Sep 1995 14:49:30 -0700
Return-Path: <[e-mail]>
Received: from lois.dti.com (dti.com) by mail.webcom.com with ESMTP
    (1.37.109.15/16.2) id AA180588951; Wed, 13 Sep 1995 14:49:11 -0700
Received: from assman ([192.84.116.12]) by lois.dti.com (8.6.9/8.6.9) with SMTP id QAA16586 for <[e-mail]>; Wed, 13 Sep 1995 16:05:47 -0500
Message-Id: <[e-mail]>
Comments: Authenticated sender is <[e-mail]>
From: "Blaise Fanning" <[e-mail]>
Organization:  Deskstation Technology
To: [e-mail]
Date:          Wed, 13 Sep 1995 16:20:18 +0000
Subject:       Re: Info on low price Alpha systems
Reply-To: [e-mail]
X-Confirm-Reading-To: [e-mail]
X-Pmrqc:       1
Priority: normal
X-Mailer: Pegasus Mail for Windows (v2.01)
Sender: [e-mail]
Precedence: bulk

I hate to *ever* agree with anything Bruce says, but in THIS ONE
INSTANCE, he's right :-)

***Warning Computer Nerd Hardware Technical Discussion Follows***

I would have to immediately discount the idea that there's a 21164
processor in that design, as the cost of the processor and cache
alone in such a system is more than a good Pentium system costs you
these days.

I instead conclude that they are talking about a 21064(A?) or
21066-based system that has a 64 bit path to cache memory.  This is a
disaster.  Without going into too much detail, the 21064 and 21066
processors have something called write-through primary (internal)
caches, which need to access backup (external) cache every time a
processor write is performed.  As a result, speedy access to backup
cache is critical to the performance of such a system.

I've got actual data (that I can't show you :-() that shows that
rendering applications need from 512K to 2M of backup cache and that
the most critical parameters in the system are clearly CPU clock
frequency and the width and speed of the path to the backup cache.
(The scenes we used were the POVRay examples of PACMan eating a
bunch of balls over a reflective surface and the chessboard with
pieces sitting outside in a field.)  Less important were the I/O
paths and the time taken to access the second, third, and fourth
blocks in the main memory system.

As an example of the kind of hit you're going to take with such a
path to backup cache, look at the benchmarks of the TeraClipse 166
in the July issue of VTU.  The substantive differences between that
machine and the Carrera Cobra 275 are that the Cobra is running at
1.65 times the frequency of the 21066 and that the Cobra has a 128
bit path to backup cache.  The rendering speed factor between the two
is 2.29, which indicates a 38% difference in rendering speed due
simply to the size of the backup cache path.

((CobraSpeed/TeraSpeed)/(CobraFreq/TeraFreq)) = (21.1/9.2)/(275/166)
 =  1.38 or 38% Faster.

Further, in that same benchmark, the Tera166 clocked in at 9.2 times
as fast as an A4000/25MHz, while a Pentium/?MHz showed 8.6x as fast.
9.2/8.6 = 1.0697 or 7% faster.  If you're paying 10% more for a
machine that's 7% faster, it doesn't seem like such a great deal.

I know that I'm probably supposed to say ALPHA ALPHA ALPHA, but if it
were me, I'd probably buy a Pentium :-)

Blaise Fanning
Chief Technology Officer
Deskstation Technology
[phone]
[e-mail]
--
"Blaise Fanning" <[e-mail]> sent this message.
To Post a Message           : [e-mail]
Un/Subscription Requests To : [e-mail]
(DIGEST)                 or : [e-mail]
Administrative Items To     : [e-mail]