Dear Tench:
The reason why it will work is that it is an extension to x86 not a complete replacement. Everyone, including Intel, has failed to replace x86 with a different instruction set. The overhang of x86 applications is far too great for even Intel to overcome. Intel has failed with i860, i960, and ipx432. AMD is taking the tried and true approach. This is why they will succeed and Intel's IA-64 will fail. Even HP is hedging its bets with PA-RISC. IBM, Sun, and the others say they will support it if, and ony when, Intel presents a saleable system.
If customers are presented with two systems roughly comparable in performance, with one that runs as fast with their x86 applications and one the runs those applications very slowly, 99% will pick the faster x86 system. So far, it appears that both the hardware and especially the compiler for IA-64 are not ready. All the rumours consistently point to the current compiler produces lousy performing IA-64 code. Assembly code by practiced hands show the possibility of high performance but very few applications will be written in assembly for customers in general (it is very expensive).
This is vastly different where one can run current apps at faster speeds, take time to validate 32 bit x86 code to work in 64 bit mode, and then slowly convert to full 64 bit software. The steps are smaller and you can take a few years if, the amount to be moved is large. And with AMD becoming a significant portion of the market, may force Intel to give up use x86-64 instead of IA-64.
Pete |
