There's been much talk in recent years about the problems that Intel's been having with the high-end Itanium processor, much of it fuelled by the delay in releasing the latest version, codenamed Montecito. Many argue that Intel's own Xeon processors, by climbing up the price performance-curve, have cannibalised sales and helped flatten the Itanium sales curve. Some say that Intel should give up on the so-called Itanic -- but it never will because its arch-rival AMD is a core reason for the Itanium's woes.
Itanium's acceptance by the market has been slow and patchy since its launch in 1994, and the Santa Clara company's had to rely heavily on the processor's co-developer HP to deliver 80 per cent of the sales of Itanium-based servers. And that's largely because HP bet the farm on Itanium, winding down its own CPU development programme in a huge gamble that resulted in its dominating a small market. All the other vendors -- which includes the ailing SGI, Unisys, Fujitsu Siemens, Hitachi, NEC and Bull -- contribute the rest of the sales, but HP's dominance suggests just how thinly spread general support for the chip is.
Since then, updating of the CPU has proceeded at a fairly glacial pace -- but then, this isn't the world of commodity servers but one where many applications will be expensively custom-built around that particular processor's capabilities. Think Sun Sparc, think IBM Power. You don't casually drop in a new server and hope it'll work.
All the same, even the relatively large number of general purpose applications for Itanium that Intel reckons are now available -- some 8,000 and climbing according to product manager Richard George -- are a mere spit in the bucket compared to the uncountable numbers of applications that run on Intel's mainstream x86 product lines.
Itanium's 'dark years'
Intel senior VP Pat Gelsinger has said with respect to Itanium that the company's gone through "some dark years" but points out that its market share is gaining. Which it seems to be. No-one is arguing that it's coming up from a small base and can therefore be expected to show the highest market share growth figures. Likewise, few have argued that the chip isn't a good performer -- just that the market is limited, and that this sector doesn't move quickly, making it tough to make the kind of progress that Intel's accustomed to with its x86 products.
Why doesn't Intel just junk the chip, since there's little evidence that it'll recoup its investment? There are many reasons.
It's pledged a huge chunk of dollars -- some $10 billion under the Itanium Alliance banner -- to its hardware and software development partners for a start, including those companies mentioned above, plus Microsoft, Oracle, Red Hat, Novell, BEA Systems, SAP and SAS. We're talking big enterprise-level applications here.
So there's a lot of money at stake with Intel providing a very public statement of intent and support.
Why make that statement? There's a lot of face to be lost. Once touted by Intel as a replacement for the x86 product line, expectations for Itanium have been throttled well back. All the same, it's a flagship product, and it would cost Intel more than money to chop the programme now -- especially when it appears to be making ground, albeit very slowly.
The rot set in when Intel failed to notice how much users started to care about the now much-touted performance per watt metric. That was one core reason why Montecito, with two cores, hyperthreading and virtualisation, was delayed by nine months. It now uses 20 per cent less power but only after the designers were forced to go back to the drawing board. The delay in releasing Montecito was down to processor design rather than manufacturing issues, said a spokesman blandly.
Itanium has also faced competition from Intel's own Xeon products for the reasons stated above: they're fast, widely software compatible, and cheap by comparison to Itanium. But the recent scramble by Intel to release new versions of its Xeon chips results from the hammer blow that AMD's Opteron dealt Intel.
Released in 2003, Opteron's wide acceptance and rapid gobbling up of 20 per cent of server market share was a surprise to Intel. Opteron was fast, used much less power and was more advanced in areas such as 64-bit computing. Suddenly, Intel was on the back foot.
It took two years for the evidence of a clawback to appear -- about right in terms of the time it takes to design and produce chips for manufacturing -- in the form of the Xeon codenamed Paxville, released last year. It was hot and power-hungry: a pair of idling Xeons could consume a massive 385W between them, when a pair of Opterons would be sipping at the electricity bill by comparison, consuming just 160W.
Even Intel admitted it wasn't the most elegant of designs, and said that most of its customers would be waiting for the new designs which were eventually launched in recent weeks, codenamed Woodcrest. Sure enough, many major server vendors, including HP and Dell, didn't bother building products around Paxville.
No switch to Itanium
What has all this to do with Itanium? The drive towards lower power and higher performance has been exemplified by Opteron. Server vendors and, by implication, their customers, have either waited for Woodcrest, or moved to Opteron. What many have not done is switched to Itanium.
The big chip has suffered at the hands of Xeons to a degree but only because Intel was forced by Opteron's success to play the performance per watt game. And it was late to that game, delaying Montecito. For a combination of all those reasons, Itanium has suffered because of the predations -- either directly or indirectly -- of AMD.
And that's the core reason why Intel will never abandon the Itanium. To do so would be an admission of defeat at the hands of AMD, a loss of face that would cost far more than money.
I can't imagine that in a million years -- can you?