Replies to post #15419 on Advanced Micro Devices Inc (AMD)

[Elmer Phud]

blauboad -

Actually, in recent days it's been Intel who has had to keep pace. Given the superiority of AFX over P4EE (which is supposed to be better than Prescott), there is little reason to think this is going to change soon. Maybe with higher-clocked Precotts, but then only if AMD's 90nm process is not on schedule.

My you can really pack a lot of misstatements into a small space. P4EE easily beats AFX is many benchmarks, SPEC among them. It's supposed to be faster than Prescott? Says who? AMD's 90nm process on schedule? Which schedule? The one from last December had them switched over to nearly 100% 90nm by now. Based on past performance, the chances of Intel delivering are far greater than AMD.

[Golfbum]

blauboad:

the point of my note was the 2004 timeframe, that should have been clear from my .09 assumption...

940/939/754 will be very hard to have a common die with any useful yield due to the pinout positions necessary to meet the memory layouts required. two die (940 for servers), desktop combo (939/754) die could work. it will be interesting to see how they stay on this treadmill of memory changes however.

wrt to the chipset guys it is a double edged sword. they don't need to pay intel a bus license but amd took some of their value add away in the form of the memory controller.

the motherboard guys are the worst hurt on the desktop. the lack of a common pinout for value and midrange in particular as 754 takes over the value position means that they get to eat inventory (or their distributors do) when the mix changes. and there's currently too many mb players chasing too small a market making it worse.

there's a point where segmentation becomes fragmentation. that point easier reached at amd's volumes than intel's.

gb

[sgolds]

Golfbum, blauboad, sockets and chipsets -

First, currently all K8 products (A64, FX & Opteron) are generated from the same die. I had thought differently, but after looking up the specs a few days ago it was obvious.

Second, IMO, when socket 939 is released there will be an updated memory controller that can use DDR or DDR2. This will be the second generation of K8 product.

Third, AMD is moving their production totally over to K8 in 2004. There have been stories of 32-bit socket 754 Athlons next year. When I put 2 & 2 together, I come up with Socket 754 migrating to the cheaper 32-bit processors (yes, the Celeron space) based on K8, and socket 939 becoming the main desktop socket. I think that A64 will migrate away from socket 754 next year.

I hope that AMD makes a smaller die to serve the low-end AthlonXPs on socket 754. Those chips won't need the 1M cache plus other stuff that comes on the current K8 die!

So, Golfbum, that is the strategy I see for AMD. There won't be tension between Socket 754 and Socket 939 on the desktop because, IMHO, Socket 939 will become the only desktop 64-bit socket. Socket 754 is destined for 32-bit low ends. There will be very clear market differentiation.