Ah yes, the
Dynamic Duo strikes again.
I've contemplated not posting, as a response is pointless and futile in the face of you two (and I'm being nice). But, to let this just go isn't an option.
Lithography is everything Joe, and history will bear this out time and time again just as it's been the deciding factor over the past 3-4 years. Concurrency, architecture irrelevant, is becoming even more important as the factor that is propelling the performance aspect of Dynamic Media applications, of which 3D graphics are a bleeding-edge application. In fact, the underlying concurrency is increasing at a factor that much outstrips the increase seen in physical clock and external communication. As we've discussed (and anyone with a hint of foresight and intelligence will tell you), we're entering a point - and 3D graphics are a prime example - where the tasks are very open-ended/high-programmability and computationally limited as we haven't seen before. To a point where the dynamic moving of information is second to the ability to compute them. This is clear and present shift - which pushes raw concurrency (on several levels) to the deciding factor of an architecture.
An architecture (especially a set-piece one destined for a console) is no exception. In fact, it should serve as
the prominent example of what performance can be extracted from a mid-decade design. This will fall back on lithography, it's impossible to avoid. You can't design a conventional IC without it. It solely provides the definitive upper bound on performance in every possible respect. From concurrency threw logic density increases, to a multiplier effect with regards to power and thermal savings with technolgies like SOI, Low-K, Cu, et al. It is the Alpha, it is the Omega - there is just no way around this. And Microsoft has realized this and subsequently taken to IP licensing, but that's not good enough as time will show.
Now, people like Joe and Dave will go on and on about how 'ATI has done things nVidia said were impossible' and talk up such rhetoric once again that's not important to anyone outside the realm of inter-PC IHV bullshit. I've been there, I know. Don't believe me, look at the physical package size - it's hardly an achievement. This entire statement is actually cyclical when taken out of the PC-biased paradigm and negates itself when applied to the fundamental question of lithography - for if ATI was able to pack (at 150nm) some high percentage of nVidia's 130nm accomplishments (probably hovers around 80%) then it just reflects badly on nVidia's one part. Which is something that we all know already and nobody is fighting that there were several design level problems. The true question to ask - which the article doesn't - is what would/will happen when you take a more advanced
N process and exploit it?
And that's what I'm concerned with. 90nm is capable of over a hundred million logic gates, 65nm will be beyond that. Process technologies like SOI and dielectrics, all CU, etc will only propel the 'curve' forwards. And living on this line
is living in the high performance sector.
I remember Dave making a comment (can't remember if it was to me or who, but I recollect it) in, maybe, 1999 or 2000 about how nVidia's aggressive lithography use (logic and RAM) will catch up to them. There was a lot of chatter about how they were lucky this time, but what happens when all the pieces of the jigsaw don't fall perfectly. And I agreed, living on the edge is a suicide wish - anybody will tell you this.
Yet, for the next 3-4 years I've seen nVidia literally rampage threw the opposition taking with them companies and institutions that seemed infallable. And the entire time they never fell, untill 2003. And looking back it's easy to say that it was living on the edge, that extra push, that finally got them. People, and that article (which is useless as Baumann well knows due to it not addressing the true corperate and design issues - I liken it to saying Ramage or Glaze3D/Bitboy's problems were entirely process based), are quick to point to lithography as the fumble. So, is lithograpahy and the living on the edge mentality to blame -
is it the fallacy? No.
The fundimental problem, as Mfa mentioned, is the disconnect between a fabless corperate body and the process technology. Lithography, unlike what Joe's mumbling about, is key and will remain key for the next decade. What's needed is a more seemless and tight integration of front-end design with back-end technology that's in house for the preformance computing sector so that those jigsaw pieces don't fall. We're currently at what I believe is an inflection point in the industry; one in which paradigm will change, corperate structures will converge and ameliorate to better allow for this methodology of design. While I'm not making a promise, the semiconductor topography of 2005-2010 could very well be quite alien to what we're accustomed.
PS. I'm not talking of just Sony. Necessary disclaimer these days.
PPS. Mfa, I'd like to see AMD and IBM together by the end of the decade.
To just DMGA: Dear Deadmeat, after reading these posts I've come to appreciate your at least attempt at logical reduction of you're thoughts. I might not agree with your conclusions, but most of it's better than this shit.
