Grimes, good questions. Here's my opinion -
The consensus on this board appears to be that A64/FX/Opteron 'owns' the high end in terms of X86 32 and 64 bit performance.
1. Is this a realistic view of the X86 market?
Maybe. Intel does not have a product that can challenge the high end today, and the tumult from Intel would imply that they are panicking - which they do whenever they are seriously challenged (remember the MTH?).
Still they can tweak the design. They have a tweak in the works for release in a couple of months to support the NX bit. It is likely that there will be design improvements to give better high end bin splits. I would expect them to be able to cherry pick a few high end dies to muddy the performance facts on the ground. All indications are that they will run hot in any event.
2. If so, for how long is this advantage likely to endure?
Intel will be behind the 8-ball until they get a new design qualified, undoubtably based on Dothan. See next item.
3. What is Intel's best horse for the next leg of this race? Dothan seems to be mentioned often, and is (IIRC) their candidate for dual core '...sometime next spring...', but given its various defficiencies (word length, bus structure, missing NX bit blah blah) lot of work will be needed to bring it up to speed.
4. It seems to me they essentially have to do a complete redesign to produce a truly competative 32/64 hybrid. How long does that take? Assuming the clock started when they moved the design team from Isreal to Oregon (Spring 04?) when would a resulting new product be deliverable in quantity, and where will AMD likely be at that point?
AMD repeated over the years that 64-bit (with all the new modes & registers) was not all that difficult. Taking them at their word I expect Intel can put equivalent functionality in Dothan in about a year, and take about another 6 months to productize it. (Probably around Q1/06.) It really isn't a new design, there are more similarities between K7 and K8 cores than there are between Dothan and Prescott. Rather it is widening and embellishing an existing design. Still, it has to be done right and all those 64-bit instructions do have to be well tested.
However, there is a lot that Intel can do in the meantime. They can spin a desktop version of Dothan with little changes (although they should add the NX bit). This would make a nice low-cost, mid-performance desktop part that they can crank out in the millions. This is more important than recapturing the high end for Intel - they need a mass quantity solution that can be put in less expansive SFF PCs with lower heat and power requirements. (Prescott burdens the SFF PC with expensive power and cooling solutions.) I think they will have such a product in early 2005.
There are some unknowns, chief among them is how much Dothan can scale on 90nm bulk processing. It may be that heat and power starts to go up pretty quickly when the clock rate hits the mid-2GHz range. It is not clear how much of Prescott's problems are bad design - or bad design meeting a limiting process.
Bottom line: Intel may have a competitive P4 product in a couple of months (and it may even be on 130nm), matching AMD's top 32-bit performance. It is likely to be low numbers of units and scarce in the market. They can have a very good mid-performance product in early 2005 based on the Dothan design, 32-bit. The real solution - 64-bit Dothan derivatives for high end desktops and Xeon - will probably come after New Years, 2006. It may require 65nm production; Intel may very well apply the lessons of 90nm to the next generation rather than puting engineering effort into fixing it.
