Dear Charles:
It is simple prudence that will keep them from ramping if for no other reason. A bad production error could make useless a number of wafers, so they will test. That takes at least 13 weeks for normal production assuming everything goes flawlessly. They remember the P3-1.13G. It actually started production and was found to be defective. Another fiasco like that with a high profile product like IA-x86-64, will make them take big hits in reliability, prestige and stock price. Thus they will ramp each line after a test of 13 weeks for each one. Doubling the number of lines each quarter. To get to 8 million a quarter will take a few quarters. At least 3, if not more, plus two to sample. Figure a year and a half once the instruction set and chip is made public (to us). Since rumors do not say more than they are working on it, the clock is still showing a delay in getting started and AMD has 2003 to get Hammer to ship both as AXP64 and Opteron.
And Intel may still have troubles, if new infrastructure is required (read chipsets, sockets, etc). If not, the glueless connectivity of Opteron will still favor AMD over P4 (Yamhill, Prestonia or whatever). It puts Intel into a bind. Add something similar for IA-x86-64 and further delay it giving even more time to AMD or leave the FSB and architecture alone and leave the bullet points to AMD for OEMs including cost, speed, scalability and ease of use.
Intel could build millions of Itaniums, but if the market will not buy them, production does not matter. So fabs only matter if, the product is saleable above the quanity AMD can produce its competitive product. This is the trap that many fell into with IA-64. They will wait until either has proven themselves and that gives the edge for AMD to crack this server market wide open.
Pete |
