Re: IBM and Gateway... ain't selling any more of those stinking AthWiper CPUs !!
Valid point, but this is a very fast moving industry, and it really looks like the P4 isn't panning out. It does look like Intel is optimistic about Itanium/McKinley (which is, after all, supposed to be the architecture that Intel goes with after 2003).
Going by the patents that cover it, 2GHZ P4 really does run its ALU at 4GHZ. And remember that when it first started shipping, production was very limited because all die had to go through 2 of their newest steppers (out of all of Intel's hundreds or thousands of steppers). So it may be that "2GHZ" P4 is using phase shift to produce a .13 core on what is otherwise a .18 die.
AMD's 1.53GHZ core, judging by its rapid ramp(check out pricewatch), is a standard .18 part, yet it still outperforms Intel's 4GHZ core. Given that AMD can expect to see a minor performance jump from .13, and a major jump from SOI within a year, getting their core to 2.5 to 3GHZ in that time frame looks fairly straightforward.
Meanwhile, going by the production limits and reports of Intel having shortages even in this very, very, down market, Intel is already using some of the benefit from .13 (though none of the benefit of copper) in their existing process - so they have only the benefit they will gain from the .13 Al to .13 CU transition ahead of them, near term. Which should be substantial, but likely will be less than the gains AMD sees from the dual transition of .18 CU to .13CU/SOI.
Intel may have a very hard time getting to the 7GHZ to 8GHZ core they'll need to compete with Athlon next year.
The Itanium core, OTOH, is currently an experimental chip, and was probably produced with a pretty conservative process, and should see considerable performance increases as it moves to Intel's newer processes.
Unfortunately, the Itanium core isn't PC compatible (unless you want to count emulation performance of a 486 or Pentium 100). My guess is that Intel will integrate in a full blown Tualatin core within a year, (the size increase should be barely noticeable, given the size of the Itanium family core) but that will still leave Itanium with Tualatin performance at a McKinley price, and kill off most of the incentive for developers to code for the Itanium/McKinley architecture.
The recent revelations that Bapco/Intel was running an application that was broken for Athlon (Media Encoder) as a background task in the Sysmark 2001 benchmark kills off any remaining possibility that P4 does better on "newer code."
Newer games also do better on Athlon than P4:
firingsquad.gamers.com
I think that Scumbria's analysis back in 1999 that concluded that the P4 architecture was fatally flawed has been confirmed.
It is becoming increasingly clear that most of Intel's remaining eggs are in the Itanium/McKinley basket. Claims of a large performance increase for McKinley are encouraging for Intel, but McKinley is not PC compatible. And we've seen what happens to any company that tries to make the industry abandon its code base for the convenience of the CPU company's new architecture from the failure of the non PC-compatible Intel i860, DEC Alpha, and HP PA-Risc architectures. |
