I came here a bit before the P4. The architecture of P4 seemed right
to me: highly-pipelined, clocked fast, and with big bandwidth to
memory. I didn't really understand about the trace cache, but I soon
learned. I was an early adopter, and I was disappointed.
The reasons for my disappointment are too many to go into. For one
thing, I didn't understand what a bodge the Camino chipset was. What
perplexed me most, though, was the P4's abysmal performance on tasks
like compiling gcc.
The responses to my perplexity could be divided into two cl*****:
Intel bashers, and those who actually knew something. As far as I can
remember, the only one who actually knew anything was David Wang (Dean
Kent was probably knowledgeable, but I don't remember learning much
from him). David's initial response to me was "you idiot," but after
some private communication, I got on track to learn some things. I
renewed my member****p in the ACM. I started reading papers. The P4's
problem (as at least some know) is its endless pipeline that doesn't
buy it as much as Intel had hope in clock speed overhead. How much it
costs and what can be done about it produced a lot of interesting
work, until Intel finally just gave up and went back to a P3 design,
which they probably never should have abandoned.
I haven't learned any of this from anyone here but David Wang. I've
listened to the actual P4 architect come clean on the subject and I've
talk to people who know a *lot* more about P4 programming than anyone
here. Everyone else is stroking their own ego or perhaps some body
part.
The problems of the P4 were my entry into learning something about
computer architecture other than the Cray 1, and it started here.
Because of all of the screaming and bellowing about John Corse and
Rambus, I never learned as much from David Wang about Rambus as I
think I could have. More generally, my knowledge of the actual
mechanics of memory and memory controllers and what might be done
about it is still weak. The railing here about AMD and Intel has
produced only venom, personal accusations, billowing clouds of smoke,
and exercises in ego that would make a professional tennis player
blush.
As a result of my investigations into the P4, I became interested in
Itanium, whose problems are related. Neither chip is what you would
call agile. They do well with predictable code (as do stream
processors), but they cope very badly with surprises. As a result of
my public musings on Itanium, I've given invited talks on the subject
to real computer architects, not just to those who play them on
usenet.
Because I once learned something here, I stupidly keep coming back.
As a result of much more reading in the professional literature and
talking to really smart people (none here), I have actually learned a
few things. The time has come, really, to give up. I don't need the
wisdom of people who know how to use a desoldering station. Nor do I
need their contempt. It's too bad, really. Real World Tech has its
own problems, but, if I'm going to find David Wang, for example, he
won't be here.
The people who are bent on showing me something or other about my
personality or some other personal thing about me are just wasting my
time, theirs, and everyone else's. If you don't get the way in which
I am bright, and I am, then stop showing how clueless you are by
trying to tell me how witless I am.
Robert.
|
