InvestorsHub Logo
icon url

wbmw

11/22/02 1:10 PM

#149 RE: dhellman #148

Dhellman, Re: AMD plans to release about 100,000 K8 processors in the first quarter of 2003, rising to 300,000 units in the second quarter and five million units in the third.

That's quite a bit more than they alluded to during their Q3 conference, when they said that more emphasis would be put on Opteron than Clawhammer. It appears that they have once again changed strategies, to get a larger ramp of desktop parts. That tells me one of three things:

A) (If you're the optimist), AMD is moving faster with Hammer than they originally predicted, and thus they can accelerate their ramp to H2 2003 from H1 2004.

B) (If you're the realist), AMD realized that pushing Opteron wasn't going to be a huge win for them without tier-1 support. Likely, they are also having issues getting 64-bit software, and perhaps they even have had trouble securing a tier-2 design win. The solution - sell Opteron into the whitebox market, and fund Newisys, while ramping Hammer as a desktop part.

C) (If you're the pessimist), AMD realized that Barton would not be competitive with Pentium 4 with Hyperthreading and 800MHz front side buses, and they would be sunk if they lost any more of the desktop market. Therefore, cut back on Opteron to save the sinking desktop ship.

Pick your favorite scenario.

wbmw
icon url

wbmw

11/22/02 1:17 PM

#150 RE: dhellman #148

Dhellman, Re: That situation may arise if AMD is unable to fully supply the new K8 processors when Intel formally introduces its Springdale platform, board makers said. Like the recently-updated Springdale products, AMD’s K8 processors support 800MHz FSBs as well. If AMD fails to offer enough processors on the market, Intel is likely to further expand its market share with the new Springdale products.

This is worth posting in a different response. Digitimes actually has this incorrect. The K8 has the equivalent bandwidth of an 800MHz FSB with Hypertransport, but that only helps when interfacing with AGP and legacy I/O, both of which need only a small fraction of what Hypertransport can deliver.

Furthermore, AMD's equivalent memory interface tops out at 2.7GB/s, which is less than half of what Springdale will offer for the Pentium 4 (unless AMD can manage DDR400 support in Hammer before launch, at which point they will have exactly half the memory bandwidth). Of course, the latency will be greatly improved in Hammer over Pentium 4 due to the integrated memory controller, but not all applications are as sensitive in latency as they are in bandwidth.

wbmw