Advanced Micro Devices Inc (AMD)

[wbmw]

UND, I don't think AMD64 has an advantage in the high end server market. As systems get larger, the price of the processors becomes a smaller percentage of the overall cost. The rest of the cost is built into functionality that can decrease the time between failures and decrease the time to fix failures. Itanium architecture and the supporting chipsets have much higher levels of RAS than Opteron, and performance is better in most cases. The Itanium ramp will start here.

Although AMD has the better 64-bit migration story, Intel is spending the money to catch up in this area, and I think they will be successful. The software ecosystem for the Madison core is already far better than it was for McKinley. Once Intel introduces Itanium in a much less expensive DP design, they will have lowered the barriers for the more mainstream markets. Although I still don't see large volumes in the Deerfield time frame, I believe the follow-ons will really help to ramp the architecture. The Montecito core in 2005 also has a low cost DP variant, and I think by that time, it will be easy for Intel to sell Itanium into the mainstream and high volume server market. The software will be better, the performance will be much better, and AMD64 will begin looking like 3DNow.

In the end, people always want bang for the buck. They don't care whether it's x86 or not. AMD may have a bit of success in getting some high profile software vendors to port to AMD64, but they don't have the resources that Intel has. The best they can offer is full performance 32-bit. Intel will get Itanium to the point where 32-bit is irrelevant, and IPF is the most widely used instruction set. They can't do it overnight, but they'll start with high end servers, move to low end servers, then try to bridge the volume markets. They'll spend what they have to spend to make it possible.