From:	Robert Nesius <nesius@******.COM>
Subject:	Re: Real-Life Computing ...(OT, obviously)
Date:	Tue, 26 May 1998 21:31:50 -0700

>Thus spake Ereskanti:
>>
>> In a message dated 5/26/98 4:58:36 PM US Eastern Standard Time,
>> mike@*******.com writes:
>>
>> > However, Intel shipping speeds like this to mass market consumers for
>> > somewhat reasonable prices, is important. The more power on distributed
>> > systems, whatever the processor, will only bring us nearer to the
>> > Matrix-like netowrks of the future. I'm already saving up for a datajack
>> > and an encephalon. :-)
>> >
>> funny, I always thought one of the things in our way was connection speeds,
>> not processor speeds at this point....
>
<Discourse on routing/switching technologies snipped>

There is a larger issue surrounding superfast microprocessors - the bus
speed on the board. The reason Exponential, who was touting a 500-700mghz
PPC chip, went bottom up was because designing a chip with backside cache
offered the same performace increase on slower chips for a much lower cost.
And what's more, the system-bus speeds on both PowerMac and Intel boards
are increasing. However, there are some very difficult problems involved
with bus speeds that high... I was talking with the Gossamer design team
this summer (The group that designed the board on the PowerMac G3's. I
worked with that team and went to lunch with them when we took the System
Software for that project Golden Master). One of them told me that
when the bus speeds get up to 100mghz or so, current technologies only
allow for the bus to be 1-2 centimeters long. That's not much room to
hook up a memory management unit, etc... (Oh, and to clarify, I wasn't
/on/ the Gossamer team. I was in a cross-functional support role.)

Anyway, the point is that the real gating item on obtaining performance
from mega-fast CPUs is how fast you can feed the processor instructions.
Even with the backside cache, these processors still end up
spending cycles either flushing their pipelines or waiting for
instructions to arrive.

There's a whole new angle developing in computing technology though -
Quantum Computers. They've actually successfully solved a problem
using a device exploiting quantum theory stuff. It's mostly over
my head, but the example used involved 4 "operations" to be performed
to make a decision. ie: a comparison. The quantum computer
solved all four possiblities at once... This is technology that
probably won't see the light of day though until....2050? :)

-Rob