This article is more than 1 year old
Intel answers Itanium mystery with riddle
One full Swope
The strain of defending the Itanium processor appears to have crushed the spirit of one Intel executive. Will Swope, a VP in Intel's enterprise group, has turned to riddles in order to explain why the company has dropped hardware that could execute x86 instructions from the
delayed upcoming dual-core Montecito processor.
CNET uncovered the x86 hardware loss by searching through a new Montecito reference manual published by Intel. The manual does reveal that the transistors present in older versions of Itanic to handle old x86 code have been scrapped in favor of a pure software replacement - the IA-32 Execution Layer. As many of you know, Intel has long championed Itanium's ability to run legacy 32-bit applications in addition to newly ported 64-bit EPIC applications.
The loss of the x86 hardware was presented by CNET as a tacit admission that Intel's broader Itanium ambitions have failed. Intel and HP never really expected customers to run tons of 32-bit applications on Itanium, but they did expect the x86 hardware to make Itanium look more attractive. A customer might be willing to buy new Itanic gear instead of another Xeon box, if they could run an old 32-bit application or two on the Itanium server. By axing the hardware and going software only, Intel has lessened its focus on making such a dream possible.
(We seem to recall Intel showing benchmarks of Microsoft Word running on the x86 transistors in Itanium. If anyone can confirm this hilarity, we'd appreciate it. Intel tells customers that a 1.5GHz Itanium can crunch x86 instructions with the IA-32 Execution Layer at about the speed of a 1.5GHz Xeon.)
As mentioned, however, Intel's Swope has taken "exception" with CNET's analysis of the situation and done so via a letter to the editor.
The problem is that we can't tell what Swope's argument against the story is.
Swope's letter starts by defining his role at Intel and the nature of the missive.
"As the co-general manager that developed the software technology for the Itanium processor family, and now as the brand manager for all of Intel's enterprise platforms, I must take exception to your analysis on IA-32 EL," Swope writes.
Okay, we get that bit.
Then, he explains the point of the x86 technology.
"We put the feature into the initial Itanium silicon because we expected that IT shops would run a large amount of mixed code--meaning some that were executing to the 64-bit Itanium address/compute space and a fair amount that would not be rewritten but would just execute their existing code."
We're still with you.
But it's in the next paragraph that things deteriorate.
"Well, many years, thousands of production deployments and many thousands of ported applications later, we now know why we have the migration to the Itanium processor family."
We've read this over about 100 times and have come up with two theories as to what it means.
The first goes something like: "Itanium has been so successful over the years that Intel's 64-bit decision was the right one." Or as one friend put the tautology: "People have moved to Itanium because so many people have deployed Itanium."
Our second theory is that Swopes is claiming Intel has done so many migrations that it knows exactly what types of code people move to Itanic and is safe making the decision to shift to a software emulation layer instead of hardware because of this rich experience. Although, this theory requires a serious leap of faith, as little in Swope's statement backs it up.
Dare we suggest that attacking a reporter's assertion that Intel has been forced to pare back its Itanium ambitions by using Itanium's strong sales figures as the central line of argument is a weak position?
Too late. That's exactly what we've done.
For the record, Swopes closes out his letter with this zinger.
"And, just to brag one small bit, that software that does emulate the Xeon instruction set is pretty darn good."
Brag and Itanium should never be linked unless you're a manager at IBM, Dell or Sun Microsystems talking about how you didn't bet your company on Intel's hapless chip.
Intel, for example, has recently been talking up Itanium's market gains and the broad software support for the processor.
According to IDC, the quarter-by-quarter unit volume for Itanium servers since the fourth quarter of 2003 goes as follows:
- Q403 7616
- Q104 8678
- Q204 8085
- Q304 8235
- Q404 8996
- Q105 8127
- Q205 8500
- Q305 8596
It's true that Itanium server revenue has gone up as a result of larger systems being sold, but you can see how flat shipments have been. Meanwhile, IBM has surged from similar Power volumes in 2001 to shipping more than 100,000 servers per year, and Sun is up to more than 300,000 SPARC servers shipped per year.
While confused by Intel's response, we can understand why it was made. It's hard to defend the massive investment on Itanium via any standard means of measurement. In the end, the company has decided that only riddles can save it now. ®