Replies to post #15695 on Advanced Micro Devices Inc (AMD)

[sgolds]

YB, it would not make sense to market a low cache version on socket 939. Here are my reasons:

1. This is intended as a high performance part, hence the dual memory controllers. True, faster DRAM can compensate for less cache, but that would be one real confusing message for the buying public.

2. 90 nm is following quick on the heels of this product, with production starting by mid year. There is a very short window of opportunity for mass production of small socket 939 dice.

3. Q1 and Q2 are the slow quarters of the year. To keep up ASPs, production should be dampened a bit anyhow. (In previous years, AMD had inventory problems from these quarters.)

Having noted that, I do think there is a useful small cache part that AMD could market - socket 754 32-bit K8 (Celeron space). Take the existing K8, reduce cache, disable 64-bit extensions. This moves the entire product line to the new motherboard design and the same manufacturing processes. It would be a useful product on 130 or 90nm. Probably best on 90nm because it would be ready for H2 when the market naturally expands, and it would soak up extra production at a time it is coming online.

[blauboad]

If the frequency ramp is going along as well as it should be, 256k cache models could replace the 3000+ and 3200+ models, freeing up the 1meg models for Opteron and top-model Athlon. Not sure what they would do with the inevitable sub-3000+ 256k cache chips, though... except maybe sandbag for the upcoming socket 754 Duron/XP line. They would have to be confident in their process. The die size savings might make it viable to dump the old XP cores altogether in that case, which I think is the plan.

Are you saying that lower cache is less of a performance hit on FX than single channel?

And I'm not sure whether the 3.2Ghz P4 is absolute standard that AMD has to measure themselves against. I see no reason for them not to release chips that are slower--they will have a market. Getting AMD64 in as many boxes as possible while raising ASPs ought to be AMD's goal for 2004, not biting their nails about P4 3.2g.

As far as manufacturing numbers, didn't some additional capacity come online, or is scheduled to come on-line at Dresden? I recall reading about an add-on the fab there. Might help to understand the numbers, or not.