Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!cis.ohio-state.edu!ucbvax!bloom-beacon!eru!hagbard!sunic!dkuug!diku!basker
From: basker@diku.dk (Tom Thuneby)
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: 486SX - Intel now telling lies
Message-ID: <1991Jun11.214144.27150@odin.diku.dk>
Date: 11 Jun 91 21:41:44 GMT
References: <1991May31.210116.16012@odin.diku.dk> <1991Jun4.174624.8972@inferno.peri.com>
Sender: basker@freja.diku.dk
Organization: Department of Computer Science, U of Copenhagen
Lines: 67


I tried to mail, but the mail bounced. I'll post it instead in the
hope that it will not annoy too many people  :)

shane@inferno.peri.com (Shane Bouslough) writes:

>From article <1991May31.210116.16012@odin.diku.dk>,
>	by basker@diku.dk (Tom Thuneby) speculates:
>> 
>> 1) The yield of a batch of 25 MHz 486's is around 14%
>
>???

What I ment was, that out of 100 chips on the wafers, only 14 chips
actually function (at any reasonable frequency). I hope this is
clearer. Another newsreader posted an article saying that 10% might
be a more reasonable number.

>> 2) More than half of the silicon is used by the coprocessor
>
>Not from the photos I've seen. The cache alone takes up a pretty big chunk.

I do not recall any details from photos of the chip. However, if
they use six transistors per bit in the cache, then the raw cache
memory would take 192K transistors, or around 1/5th of the silicon.
I don't know how much cache controller, address decorders and the
like would take, but I don't think it would be quite as much. 

A little obnoxious remark (sorry, but I couldn't resist :)
How do you know it was the cache that took that chunk of silicon?
Couldn't it just as well have been the multiplication array of the
coprocessor? I for one couldn't tell the two apart. End of remark.

I was also surprised of a coprocessor of this size. I wouldn't be
surprised if somebody told its size was, say 1/3rd or 1/4th of the
silicon. But then I also heard that the 486 had floating-point
performance somewhat superior to the 387. That would cost
transistors.

>> 3) The coprocessor has a relatively high probability of errors
>
>Vs. what?

The original poster didn't mention that. I imagine that it's some
sort of bugs-per-transistor measure. That's what I took it for,
anyway.


I'd like to know if you disagree with my estimate of the yield of
the 486SX, or if you only disagree with the numbers upon which the
estimate is based. I'd also like to know what numbers you believe is
correct.

Thank you for the criticism. Post or E-mail (or forget) a reply.

			Tom Thuneby (basker@diku.dk)

P.S.
>-- 
>Periphonics Corp.  | Shane Bouslough is: shane@inferno.peri.com    Ride Bike!
                                          ^^^^^^^^^^^^^^^^^^^^^^
>4000 Veterans Hwy. |
>Bohemia, NY 11716  | "Too slow, Chicken Marango! Too slow for *this* cat!"
>516-467-0500       |                                      -Cat, Red Dwarf

I tried that address, but my mail bounced. Can you give me (us) another?
			Tom Thuneby