🚀 go-pugleaf

Roelof Vuurboom wrote:
>With a little back-of-the-enveloping we might even be able to roughly
>quantify this.
 ^^^^^^^^ raising it from an art to a science!
>
>We estimate that [...] to satisfy customer needs we [need]
>at a perfomance doubling every 2 years or a factor of about 1.4 every
>year. This is like inflation. In other words, a 10 Mips machine available
>now offers the "same" performance (value form money) as a 14 Mips machine 
>available over 1 year or a 12 Mips machine available over 6 months.
>[etc]
>[...] a 10 Mips compatible processor is equivalent to
>a 12 Mips incompatible one due to the extra lead time.

Mr. Mashey wrote:
Our general rule-of-thumb is that:
	10% is uninteresting
	50% is interesting to lots of people, although not others
	100% is interesting enough to make people take serious risks
	1000% is uninteresting to some people (really!), i.e.
	compatibility is everything.  
(never get tired of hearing what that guy has to say)

I write:
although this is probably label-able as "marketing", i must confess that
this kind of stuff is exciting for me to read.

it kind of reminds me of Daniel (connection machine) Hillis's approach
to multiprocessing: "forget the 4 processor model.  forget the .5u CMOS
model.  examine the limit as N->infinity, and as processor size goes towards
0.  conway&mead's book worked like this, somewhat; extracting the overall
patterns from the physics.

i've been tagged as ivory-tower-ish cause i "can't tell what's
good enough".  well it's great to have that finally quantizable, not that i
take your numbers as the gospel, but approximately.  i mean, aint' it true
that lots of arch decisions--well maybe not "arch" but "computer design"--are 
like "do we wait for the -30 or build in the -20 now?".  i generally "hate 
marketing stuff" too, but if/when market acceptance is key to whether or not 
you're copmpany will be around till generation 2 is done, or if i'm gonna 
predict whether joe businessman will put up with a water-cooled lotus123 
accelerator under his mahogany desk, i have to know these things.  it seems 
that engineering is "meeting needs with max !/$.  if i'm tryna decide to 
include part X vs part Y then relative !/$ is ok but if i have a new idea 
then absolute benefit vs. hassle is what counts.

so maybe we can grade fabrication lines or semi houses in terms of their
bang-per-buck time constant.  if they can't keep up with "inflation",
don't go with their gate arrays, cause then your product will get buried,
as t->infinity.  just joking, but concievably not.

i never thought i'd argue FOR keeping marketing stuff in comp.arch (that sure
wasn't what motivated me to press the "F" key, just excitement at the data).

----------------------------------------------------------
Spelling brought to you by Tanqueray.  R. Stein was right. 
----------------------------------------------------------