Wolff, your conclusions about Merced are based on very faulty assumptions:
<but the end result is a pathetically delayed part.>
And yet, when Merced is finally released, it is considered to top the big RISC boys like Alpha and UltraSPARC. Sure, the well-publicized Merced delay was a big embarassment for Intel, but that's old news now.
<Essential you have the P8 ready within months of the release of Merced.>
Get it right, it's called Willamette, not P8. Actually, let's talk about Foster, the server version of Willamette. Foster is not intended to replace Merced at all. In fact, it is speculated that Foster's performance will roughly equal that of Merced. The real successor of Merced, McKinley, will blow away both Merced and Foster in terms of performance, but McKinley isn't going to arrive until late 2001.
<The only reason why the Merced will be released is to save face.>
No, the only reason why Merced is being released is to launch a new IA-64 platform. It takes time for the industry to adopt a new platform. Merced is intended to spearhead the transition. Then McKinley will follow up with even more performance. After that, Madison and Deerfield will continue the ramp-up in IA-64 performance. And this path is totally separate from the Willamette/Foster path.
Let me get one thing cleared up: IA-64 and IA-32 are going to be two separate paths. IA-32 will pursue optimum x86 performance (compatibility is a great thing to keep), while IA-64 will pursue pure performance at the expense of x86 compatibility. (Actually, IA-64 is meant to be hardware-compatible with IA-32, but IA-32 code running on an IA-64 processor isn't expected to have mind-blowing performance.)
Tenchusatsu |
