someone (someone.delete@this.somewhere.com) on September 27, 2013 12:24 pm wrote:
> bakaneko (nyan.delete@this.hyan.wan) on September 27, 2013 10:10 am wrote:
> > David Hess (davidwhess.delete@this.gmail.com) on September 27, 2013 8:06 am wrote:
> > > bakaneko (nyan.delete@this.hyan.wan) on September 26, 2013 3:06 pm wrote:
> > > >
> > > > What I don't get is how Intel got the idea
> > > > that building Itanium would be successful.
> > > >
> > > > I'm reading through the wikipedia entry
> > > > for i860, which Intel worked on half a
> > > > decade earlier, and it is VLIW and it had
> > > > the same negative downsides from shoving
> > > > prediction into the compiler.
> > > >
> > > > So why did Intel think Itanium was going
> > > > to be any different?
> > >
> > > If AMD had not been successful with AMD64 and Opteron, then Itanium would have been the successor
> > > to x86-32 or so Intel planned. Wasn't that the reason for including support to run legacy x86-32
> > > binaries? Intel's customers had other ideas however which AMD was able to capitalize on.
> >
> > That's besides the point, Intel had data that
> > VLIW sucked. Hard. They may not have been able
> > to pinpoint why, but they had data that showed
> > that VLIW designs don't compete even before
> > Itanium.
>
> Wrong. A simple comparison of the data points of high end processors all
> made in 0.18 um Al interconnect bulk CMOS shows otherwise. (SPEC CPU 2k)
>
> [Power4/1300 - 822/1202 ***SOI/Cu device, perhaps 10% faster than bulk Al 0.18 um impl.***]
> McKinley i2/1000 - 810/1431
> Athlon XP/1733 - 720/613
> Willam. P4/2000 - 681/735
> Alpha ev68b/933 - 579/724 (no official submission, scaled from 0.18 um Cu ev68b 1 GHz result)
>
> Where's the problem?

The metric of interest is not merely "performance".
For a buyer it is "performance/dollar". For a manufacturer it is something even more elusive, "performance per engineer hour", with "performance/mm^2" also important;
with a bunch of associated even more complicated caveats like how easily the design splits into parts that can be handled by mostly independent teams, and how much of the design can be automated.
POWER4, in particular, I suspect does the best by the engineering metric (maybe not as much by mm2). Certainly at the time IBM was very proud of how well the design performed given how much of it was synthesized rather than manual.