Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email firstname.lastname@example.org
Re: ... is dead too
Date: Sun May 25 2003 - 14:55:11 CDT
(Sorry for this not being a direct reply. Redhat is too snobish
to accept direct email from DSL customers of my ISP.)
> From: "Bryan J. Smith" <b.j.smithieee.org>
> Date: 25 May 2003 11:27:04 -0400
> It's hard to fight economies of scale. But Intel could have changed all
> that by sticking with Digital in the early '90s and slowly adopting
> Alpha in the latter '90s.
Yes, that's what you and I and the other idealistic techies
would have wanted. Unfortunately, decisions at Intel are
rarely made on the basis of technological idealism. Since
about 1985 or earlier, decisions have been made on the basis
of internal company politics and what's good for marketing.
While that's good for their stockholders, it does not
necessarily promote the most efficient advancement of
> Intel, on the other hand, bet on EPIC to keep the pipelines full and
> Predication to remove the need for branch prediction. While I and
> everyone thought the EPIC/Predication concepts were extremely
> innovative, sometimes even the greatest innovation does not mean
> real-world performance.
Actually, I have it from pretty good sources that the Itanic
instruction set architecture was actually designed primarily
by HP, not by Intel.
Many of the problems with the Itanic 1 (alias Merced),
mostly the delays in getting to silicon, were caused by
employment conditions and management issues among the design
team. During much of the 1990s, you couldn't hire or retain
quality engineers in Santa Clara, because there were too
many employers seeking too many engineers. I hear they had
a lot of employee turnover, which makes it difficult to
maintain project schedule and keep the design quality good.
Maybe, someday, we'll have the combination of a clean and
efficient instruction set architecture with inexpensive
implementations. Until that day, we'll have to get by with
what we have.
Axp-list mailing list