Replies to post #15709 on Advanced Micro Devices Inc (AMD)

[sgolds]

jhalada, I consider artificial segmentation to be a necessary evil. It should be used only judiciously to prevent wholesale cannibalism of high margin product by low margin product. However, if you do too much of this then competitors will take your market by offering more for less. It is a balancing act, certainly!

There was a rumor posted here a few weeks ago which talked about a socket 754 32-bit AthlonXP. Someone thinks it is going to happen. However, you make a good argument that we need to get as much AMD64 out in the wild as possible, so I am convinced that the socket 754 replacement for AthlonXP should be 64-bits.

However, there still is that cannibalism effect. I don't think the reduced cache is enough to prevent this, so I like your idea of limiting memory to 4GB (32-bits). This part could still be marketed as AthlonXP because it leaves out one important part of Athlon64, yet still supports AMD64.

We seem to mostly agree about the benefits of a small die offering for the value market, so there probably isn't need for further followup on this thread. Whether it limits addressing is not the most important aspect. Having a cheap part to soak up extra 90nm capacity is the important item, IMO.

[KeithDust2000]

Joe, I think L2 cache size is enough of a distinction between value line and mainstream.[...] I am not too crazy about artificial segmentation.


The lack of market segmentation is currently one of AMD´s major problems. INTEL has successfully positioned their Celeron processor against AMD´s mainstream XP line, resulting in artificially depressed ASP levels for AMD up to the point where Celeron is sold (currently 2.7, soon 2.8 Ghz). INTEL has exploited this weakness to the fullest by greatly accelerating their Celeron roadmap from earlier plans, without hurting their higher-end margins - this is the reason why they still get $170 for their 2.4 Ghz P4, while the AXP 2400+ only gets a mere $70. Celeron of course differs from their mainstream/high-end line in brand name (which opposed to Duron has established itself), cache size, FSB speed and lack of hyperthreading (and clock speed, now to a lesser extent). The speed difference is enormous, the segmentation is clear, and so is the pricing scheme. OEMs need this segmentation for their product positioning.

Compare that to AMD´s current line - it´s (basically) AXP all the way, with QS ratings from top to bottom, the cache size difference doesn´t segment anything here. However, it is vital for AMD to segment their market, and the plans I´ve seen for the K8 line look like the best bet out there, the segmentation is clear - OEMs appreciate it, and have asked for it (and so would I). Time will tell how well it works out.

The return (or continuation, depending on your viewpoint) of 533 Mhz P4´s is further proof of OEM´s desire to further segment their markets.

AMD´s line-up for 2004, with (now well established) AthlonXP (value), Athlon64 (mainstream) and Athlon 64 FX (high-end) with the well-known clear differentiations between them is exactly what AMD needs to protect their margins and offer OEMs what they need. And that´s ultimately what counts.

P.S.

As for the 4Gb limit someone was proposing - let me comment on that from an OEM perspective:

It would be impossible (and pointless) to market, customers wouldn´t understand nor care about it, and more than 1 GB (or slightly higher as time progresses) is just an illusion for the mainstream and value markets for the foreseeable future anyway.
It´s a bogus idea, as simple as that.

As for continuing AXP sales hurting AMD64 acceptance, I think the merit of this argument is limited at best. 64bit will remain a server/professional workstation/high-end play for a long time to come, and developers and AMD know that.