In-Depth

Itanium: Intel’s Once-and-Future 64-bit King

Even though Intel has embraced 64-bit on the low-end with its new Xeon chips, it hasn’t given up on Itanium

• By Stephen Swoyer
• 08/17/2004

What—another 32-bit processor from Intel, the 32-bit volume market king?

Nocona isn’t just any Xeon microprocessor. Like its predecessors, it’s designed to run 32-bit x86 code at blazingly fast speeds. Unlike its predecessors, Nocona also boasts new extensions that enable it to process 64-bit code and address a 64-bit memory space.

The upshot is that Nocona is going where no Xeon processor has gone before—into the 64-bit stratosphere. The catch, of course, is that Intel already has a 64-bit processor—Itanium. What, then, does the advent of 64-bit Xeon portend for Itanium, Intel’s would-be 64-bit volume market king?

The original Itanium processor debuted several years ago—late, over-budget and (as a result of its tardiness) underpowered relative to its RISC competition. Since then, Intel and its partners have consistently reprised and updated a constant theme: Itanium uptake is limited, to be sure, but sales will ramp up when an improved version (e.g., McKinley, Madison, or Montecito) ships later this year or sometime next year.

At the same time, Intel has sought to move the goalposts, so to speak, with respect to the expected pervasiveness of Itanium. As you may recall, the original Merced chip was expected to be the precursor to a 64-bit renaissance that would drive the industry away from 32-bit and toward 64-bit volume shipments. That hasn’t happened, and folks from Intel no longer think that it will—at least, not with Itanium leading the way.

The good news for Intel is that Itanium shipments are ramping up. In the first quarter of this year, the chip-making giant shipped 6,281 Itanium boxes, accounting for $282 million in revenue, per Gartner Inc.’s market research data. That makes for a 500 percent increase in unit shipments and a staggering 740 percent spike in sales. That kind of performance has some of Intel’s partners lauding a potential explosion in Itanium-related uptake among customers.

Take HP, for example, which stresses that other factors have accounted for Itanium’s thus-far-limited uptake.

“Overall, a worldwide recession hurt spending and certainly hurt new technology, but we’re certainly happy with where Itanium is going. We do see a lot of demand in the Windows and Linux space, especially,” says Lorraine Bartlett, director of marketing for Windows on HP’s Integrity server line. “We are at the beginning of where Itanium sits in the scheme of things, and we’re kind of getting to the end of where x86 is. There have been tremendous gains from Merced to the Madison-based Itanium 2 processors, and we’ve got an additional launch of Itanium 2 processors coming out this year. We’re just starting to see that take off now, and the uptake [numbers] from Gartner and IDC show that.”

Scott Schweitzer, product manager for NEC’s Itanium server line, takes a more pragmatic approach. “[Itanium] hasn’t quite picked up as much as we hoped, but from a technological point of view it’s far superior to the Opteron stuff that’s out there today,” he comments. “Where we’ve had some traction is in doing SQL [Server] scale-ups, where [customers have] SQL databases and just need more horsepower and more memory, and their path for upgrading is obvious.”

The bad news, of course, is that even with 500 percent growth in server unit shipments and near-750 percent growth in revenues, Itanium is still a drop in the bucket of the worldwide server market. In fact, Itanium unit shipments were dwarfed by demand for the industry’s other 64-bit microprocessor, Opteron from Advanced Micro Devices (AMD) Inc., which moved 31,184 boxes in Q1. (Unlike Itanium, Opteron is designed to run 32-bit code and 64-bit code with about equal facility.) Itanium is undoubtedly a more lucrative play than Opteron, of course: in the first-quarter, AMD’s 64-bit giant-slayer accounted for only $93 million in sales—just one-third that of Itanium. But the larger point is that both 64-bit chips comprised just two percent of worldwide server unit shipments in Q1.

As a result, Intel officials have sought to reposition Itanium: The would-be 64-bit volume-market champ has now been recast as a high-end-only play. “If you look at a lot of the RISC vendors today—IBM, HP, even Sun—all of those major high-end RISC OEMs have continued to maintain RISC at the high-end and Xeon for the volume market,” says Jason Waxman, director of multiprocessor marketing for Intel’s Enterprise Product Group. “There’s a reason for this. There are two very distinct sets of requirements. It would be nice—it would have been nice—if Xeon could cover both the volume segments and go all the way up, but, fundamentally, there are two different requirements that will continue to coexist.”

The difference between these two distinct requirements is in part elucidated by the price disparity of Itanium-based systems and their Xeon- (or Opteron-) based brethren: Based on Gartner’s Q1 data, the cost of the average Itanium system is about $46,000. The price of the average Opteron-based system, on the other hand, is around $2500. Those are two very distinct numbers.

The upshot, says Nathan Brookwood, a principal with microprocessor consultancy Insight64, is that the two- and four-way 32-bit space is where the volume is. Itanium is right now a mostly a niche player that’s attractive for specific applications. “Itanium still attracts converts in the [high performance computing] space and in the ‘Big Brother to x86-based Windows’ space,” he says. “Anyone looking for larger industry-standard SMP configurations, and/or 64-bit Windows with 64-bit SQL Server still only has one choice—Itanium.”

Intel, for its part, saw where the wind was blowing in the 64-bit volume space. As a result, the chip giant formally announced its 64-bit extensions (the “worst-kept secret in the industry,” according to more than a few analysts) earlier this year. Waxman says that Nocona is designed for a different set of requirements than Itanium. “The Xeon chips with 64-bit extensions are a way of getting full capability for a 64-bit architecture, but doing it for the existing software base. There are a lot of applications out there on the 32-bit architecture, and the vast majority of those are going to remain 32-bit,” he acknowledges. “Itanium was designed really from the ground up for mission-critical, high-end applications. Xeon continues to be for the mainstream.”

Nor does Intel expect the volume sweet spot to shift over to Itanium anytime soon. “By the middle of next year, we expect that more of our enterprise processors that we ship will be 64-bit rather than 32-bit,” he acknowledges. “But what we don’t necessarily see is that all of the 64-bit software or infrastructure will be there, so we don’t see that [volume] shifting to [the 64-bit only] Itanium.”

Instead, Waxman and other Intel officials talk up a hybrid paradigm—called the Common Parity Cost Platform—that will result when the underlying architecture that supports both Xeon and Itanium converges in 2007.

“By 2007, we have planned to deliver a single platform that would allow OEMs to ship either Itanium processors and Xeon [on the same board],” Waxman confirms. “Now you’re able to offer customers at effectively the same price the choice of buying either Itanium or Xeon. We expect that the Itanium processor in that platform will be up to two times the performance of Xeon at that time, so that transition is going to continue to drive Itanium volumes.”

For his part, Insight64’s Brookwood allows that such a convergence could occasion Itanium uptake in greater volumes. “The common platform will … reduce Intel's spending for IPF infrastructure, and thus help its ROI, and … allow the cost of IPF platforms to match that of comparably configured Xeon platforms. Then, if the performance of IPF is still two times that of Xeon and the prices are comparable, users may find it more attractive to migrate to these platforms.”

If this doesn’t work, Brookwood says, Intel may be forced to make some tough decisions. “The issue for Intel is that systems priced under $25K make up 90 percent of the server market, leaving a potential market for Itanium of about 500K systems,” he comments. “Only time will tell if Intel wants to continue investing large amounts of R&D and marketing dollars in such a constrained market, when it can get so much more leverage with its Xeon investments.”