The CPU market is one that rewards innovation and performance, but chews up and spits out players who don't bother to bring their "A" game. But sometimes, a contestant rides a hot streak and runs the table -- just as AMD has done with the new Athlon 64, toppling Intel's 3.2GHz Pentium 4 to take back the high-end performance crown. The dual-pronged attack of the more mainstream Athlon 64 3200+ and elite-speedster Athlon 64 FX-51 has been one of AMD's best-received processor introductions, with many reviewers going out of their way to praise the entire AMD64 lineup.
Intel, of course, is not a company to take this sort of offensive lying down; time and again, its counterpunch to a new AMD chip has raised the performance bar even higher. This time, however, the picture is less clear. Intel needs to confront the AMD64 platform head-on, but insists it can do so by refining its 32-bit processors. And while the silicon giant has advantages ranging from brand loyalty to Hyper-Threading technology, this may be its toughest fight yet.
For instance, there are rumors circulating that Intel is having trouble with its transition from 0.13-micron to 90-nanometer process technology, delaying the upcoming "Prescott" core. Naturally, Intel denies this, but we've still not heard any firm details on Prescott's release date.
On the other hand, it's entirely possible that Intel's 90nm technology is fine, but the company has pushed Prescott back a bit anyway. This could be a result of AMD's somewhat surprising move in offering the mighty Opteron server CPU as the desktop Athlon 64 FX-51; it's possible that Intel has only now grasped the ramifications of this release and the higher-speed FX-53, FX-55, and other models that AMD will release in future.
It's hard not to think of possible comparisons to the PC graphics-card market. Nvidia was firmly in the driver's seat, then bet the farm on a shift to smaller 0.13-micron process technology. Instead of following suit, ATI upgraded its technical specs quite nicely, while refining its 0.15-micron process to achieve record clock speeds. The result has been a somersault in the power structure, with ATI riding high and Nvidia on the defensive, due to a combination of 0.13-micron delays and perhaps stubborn pride in going ahead with the release of a top-of-the-line product -- the GeForce FX 5800 Ultra -- that clearly couldn't match ATI's best. The results will be felt for years to come.
Vendor loyalty and Intel's heavy presence in PC chipsets make such a reversal of fortune unlikely here. But Intel has seen firsthand what happens when the market leader introduces a new product that fails to exceed its competitor's. Prescott must retake the desktop performance title from the Athlon 64 FX-51, and do so by a noticeable margin, or critics will tag its release as a failure.
Delaying the new core could give Intel time to increase clock speeds, potentially unveiling a surprise model above the expected 3.2GHz and 3.4GHz Prescott CPUs. Intel rarely makes missteps, and if delivering a firm thumping to the FX-51 is the goal, then postponing Prescott may prove to be the right decision in the long run.
While we're waiting for Prescott to exhale, Intel hasn't yet exhausted its possible strategies for the current Pentium 4 line. The company has already announced the Pentium 4 Extreme Edition, which combines the standard "Northwood" core at 3.2GHz with a Xeon-style 2MB of Level 3 cache. The obvious intention was to shunt some publicity away from AMD's Athlon 64 launch, and also to give Intel a stopgap competitor before Prescott reaches the marketplace. There are only two problems: the P4 Extreme Edition has yet to ship, and when it finally does, expect its price to be prohibitive.
Another strategy being speculated about is upgrading the Pentium 4's 512K of Level 2 cache to, say, 1MB. But this doesn't make much sense, as it would put these theoretical Pentium 4s into direct competition with Prescott. If Intel introduced a Pentium 4/3.2 with 1MB L2 cache, it might not have all the SSE3 features of the Prescott core, but it could give the new design a good run in general performance. Intel already has enough to worry about with the P4 Extreme Edition encroaching onto Prescott territory, so it's doubtful we'll see any further upgrades to the Northwood core.
That brings us to the Celeron -- which, as far as we're concerned, has largely outlived its usefulness. The Prescott CPUs will compete against the two Athlon 64 versions, while the Pentium 4 battles the Athlon XP, leaving the Celeron to go head to head with ... itself, maybe. In today's desktop market, AMD offers far greater value at the entry level, while most Intel loyalists are setting their sights on the 2.4GHz or 2.6GHz Pentium 4. There are still low-priced retail PCs that use the Celeron, but this seems to be based more on marketing habit than logic.
The latest Intel roadmaps suggest a future Celeron based on the Prescott core, with less cache (256K) and likely none of the spiffy SSE3 or Hyper-Threading features of the flagship -- so basically, Intel will be offering a slightly upgraded, cooler-running version of the original Pentium 4 rebranded as a Celeron. Intel obviously wants to move all production to the Prescott core, but it still seems a bit of a waste to hamstring the chip that badly, especially when viable Pentium 4 models could easily be used instead.
Right now, AMD is in the driver's seat for performance PC shoppers, and it's Intel that finds itself sprinting to catch up. This game could continue through the rest of 2003, giving AMD more time to release faster processors and work out some of the Athlon 64 FX-51 platform kinks left over from the Opteron server-based design.
We've seen this movie before, when Intel let the Pentium III languish against AMD's Athlon for a time before it dropped the Pentium 4 hammer and began its dizzying climb up the megahertz ladder. Time will tell whether history repeats itself (or goes off the rails with rumors such as Intel's revealing Prescott's secret 64-bit extensions), but with its track record, it's tough to bet against Intel for long.