AMD Athlon 64 is delayed - can Barton keep up?

WHILE AMD'S announcement of Opteron availability on April 22nd is a promising step towards Hammer availability, there's no doubt that the company's decision to push Athlon 64 back until September is going to frustrate a large number of consumers who were planning to build systems around the new processor in the first half of this year.

(See AMD Athlon 64s delayed until September)

AMD's announcement may serve as tacit confirmation that the company is having yield problems, possibly caused by the difficulty of implementing SOI.

Although no company wants to crank out CPUs at a low yield, it's more important for AMD to keep yields high than it is for Intel. Intel not only has a much greater volume of revenue on hand to cover the cost of throwing away large chunks of wafer, it also has the fab space to crank out greater numbers of CPUs to compensate somewhat.

AMD, by contrast, has only one fabrication facility currently producing CPUs and much less elasticity in their manufacturing capability.

For this reason, AMD's preferred strategy has always been to hold back from releasing CPUs until their yields are very high. Delaying desktop/workstation Hammer could have another benefit if the CPU market begins to recover this year, as AMD would be able to charge more for the processors and strengthen their [its, companies is singular Jack, Ed.] badly-beaten margins.

Hopefully we'll also see a few improvements to the core as well. When the Hammer architecture was first introduced an onboard memory controller and DDR333 seemed like a blazing-fast combination, capable of pounding anything Intel had to offer. The growth of dual-DDR and the rapid scaling of DDR-I together, however, make Athlon 64's single-channel DDR333 make more anaemic than it once did. While its unlikely Athlon 64 will appear with dual-channel DDR, hopefully AMD will take advantage of the delayed release date and incorporate support for at least DDR400.

AMD's official reason for why it can afford to push back Athlon 64's launch is that Barton is scaling so strongly that the company doesn't need to push its Hammer solution into the desktop. This is both good and bad news. If Hammer was yielding at 90% its highly likely we'd see Athlon 64 launching simultaneously with Opteron, but, the fact that AMD feels it can push the AthlonXP architecture still higher to continue to compete with P4 implies good things about the AthlonXP's ability to scale on its 512K cache design.

The AthlonXP, in fact, has performed beyond what most consumers or even reviewers thought it would be capable of back when the P4 was introduced. Back in 2000, everyone expected the AthlonXP to run out of gas fairly quickly while the P4 architecture rocketed ahead in terms of total performance. The Pentium 4's performance has indeed jumped dramatically since then, but AMD's 2600+ in August and its move to a 333 MHz FSB has substantially closed the gap.

The true test, however, is still to come. In a matter of months the P4 is scheduled to leap to an 800 MHz FSB. Even assuming AMD scales the Barton to a 400 MHz FSB as well, it will still leave the AthlonXP running on a bus only half as fast as that of its rival. Barton had best have wings on its feet if it's going to scale high enough to counter that attack. µ

See Also
The battle between X86-64, the Intel Itanium and Planet Other
Full Hammer coverage