> But by then (if all goes well for AMD), AMD will be fielding a line of 64 bit processors,
> with a smaller die size than the 32 bit P4, a higher IPC than the 32 bit P4,
> and equal or higher clock speed thanks to SOI.
I think this paragraph is a bit optimistic. From what AMD has said, we'll see their 64-bit processors appearing in 2002 Q1 at best. They will have a smaller die size than the 32-bit P4, yes, and a higher IPC is practically a certainty (I would peg it possibly as "much higher"), but equal or higher clock speed? I dunno about that.
PIII is limited by its circuitry (the speed paths and soforth), much in the way that the K6 is. Both Athlon and P4 are limited by heat production and power dissipation. Intel's and AMD's processes ... proces- ... processi (doh) ... manufacturing techniques are pretty much equivalent, with both having 180nm features and either side having features that the other would kill for (Intel wins in less advertisable ways). From what I can see, Intel's 130nm process will have smaller Leffs and similar benefits. AMD's 130nm process will have doodads like SOI. They'll probably be more or less on par, though I'm tempted to crown Intel here if only by an Angstrom.
The Netburst (nee Willamette) core design seems capable of frequencies of about 25% higher than the K7 core design, although this gap will probably decrease over the coming months (Intel projects, if I recall correctly, 2.00GHz in Q3 while AMD looks forward to 1.70GHz at the same time, reducing Intel's lead to under 18% even though Intel's process tech will have become superior to AMD's at the time).
From my incredibly undereducated point of view, it looks like Intel's 130nm process more competently addresses feature and path sizes while AMD's 130nm process addresses heat production and power dissipation. Given that neither P4 nor Athlon seem limited by critical paths, I do think that AMD might benefit from their 130nm process more than Intel might (compared to current process techs). But I still think that -- barring design changes -- we'll have a situation where Intel is a little ahead in frequency while probably being somewhat behind in performance (this, of course, depends on all that optimization jazz and such, of course).
I have no idea how K8 will affect this fight. AMD has (I seem to recall) admitted that the K8's pipeline will be lengthened, but that does not always directly translate to superior ramping potential. Intel, for example, shortened the pipeline from Merced to McKinley, but the latter is still touted to be much higher in frequency, even on the same process!
I do think that K8 will increase ipc, especially in those areas that Intel will grab on the short term due to code optimization (as K8 will support SSEn opcodes).
-JC |
