Doug S (foo.delete@this.bar.bar) on September 24, 2013 5:54 pm wrote:
[snip]
> I'll believe it when these start showing up in Android tablet statistics as something other
> than "other". Otherwise it is all just more hot air from someone, who still won't concede
> that I was right about Itanium all along. I always said the following things about it:
>
> 1) it would become HP only

How do you define "only"? I seem to recall that not too long before Intel's "fatal announcement", a Chinese company announced Itanium adoption. I would not call even a 95% market share exclusive (unless the 5% were merely rebadged machines).

HP's stated strategy of being distinct through better compilers may have hurt Itanium adoption since EPIC is defined to require excellent compilers. (Delaying CPU release to coordinate with HP having its chipset ready would also tend to clarify that anyone else is a second-class citizen.)

> 2) Microsoft would never seriously get behind Windows on Itanium

While Microsoft seems not to have been committed to Itanium, if Itanium had been the only credible 64-bit merchant server ISA (with new process adoption timing like Xeon, multiple vendors, no alternatives but a POWER3 follow-on), Microsoft would have been much more supportive of Itanium.

With AMD64 (and POWER4), many vendors quickly dropping Itanium, and Intel struggling to produce strong implementations in a newish process, it should not be surprising that Microsoft would abandon Itanium. With Merced being delayed and pitiful (developer test systems only), serious skepticism about Intel's commitment and/or competence might be reasonable. (I seem to recall reading that even in the McKinley time frame there were serious software issued--e.g., compiler bugs.)

> 3) sales would start an inevitable slow decline after the PA-RISC migration completed

Even with AMD64 (and the performance strength of POWER4) and the disaster of Merced, I do not remember it being obvious that SGI would drop Itanium (die maybe but not continue for years after dropping Itanium) or that the Japanese interest in Itanium would fade.

While the slow death of the UNIX market might have been obvious, I do not think it was obvious that Linux on Itanium would not grow at least as fast as UNIX declined, at least for a while.

> The economic downturn made the slow decline a steep decline, but other than that I was 100% correct.

The rise of x86 clusters for HPC probably did not help either, especially in supporting non-HP system vendors. (At least I received the impression that HP was less oriented toward HPC than SGI and perhaps some other Itanium vendors.)

I admit that I have not studied the history of Itanium, but the general ISA does not seem that bad and the idea of a merchant processor seems very alluring (and plausible). (The merchant processor model--particularly with a system vendor emphasis on chipsets with memory controllers--may well have excessively encouraged a delayed implementation of integrated memory controllers and on-chip directory support; so the effect is not entirely positive.)

Even though Itanium was overhyped (some hype might be reasonable for a new ISA) and subject to excessive delays, even when McKinley came out it might not have been obvious that Itanium was almost exclusively going to be a PA-RISC replacement. Also an expectation of significant microarchitectural innovation and competitive process technology adoption might have seemed reasonable. A decade of "Tick" (modest microarchitectural change) should have seemed highly improbable.