Craig Barrett, Intel’s CEO, was the first executive to be questioned about Itanium at IDF this week. In a question and answer session following his keynote address on the importance of innovation at Intel and in its partner base, he was asked how Intel would differentiate Itanium from Xeon.
Itanium is a big-iron processor platform, he explained, saying that the requirements for big iron were different from workstations and even entry and midrange Xeon servers. Itanium has a happy home in that spot, and its competition–mainly Power and Sparc–is well-defined.
While Intel has committed to eventually offer a unified server platform that supports both Itanium and Xeon processors in the same boards and in the same form factors–perhaps by 2007–Barrett made it clear that Intel had no plans at this time to somehow consolidate the Xeon and Itanium processors and rid itself of these questions, once and for all. They will stay separate as they go forward, and I don’t see a need to merge them going forward.
Abhi Talwalker, co-general manager of Intel’s Digital Enterprise Group, which is responsible for desktop, workstation, and server processor, chipset, and platform development, echoed the distinctions Intel drew when it launched the 64-bit extensions for the Pentium 4 and Xeon processors a year ago at IDF.
The server and workstation market is a $60bn marketplace where a single architecture does not really fit all needs, said Talwalker. We have taken a dual architecture approach. We are focusing Itanium at the RISC replacement market, and the primary target is IBM Power. Talwalker lasered in on IBM Corp’s Power, not mentioning Sun Microsystems Inc’s Sparc or Hewlett-Packard Co’s PA-RISC or Alpha processors, or other architectures, such as the few remaining proprietary mainframe and midrange platforms.
IBM’s Power platform is being singled out for good reason when Intel talks about Itanium. For one thing, IBM has decided not to support Itanium with its Hurricane X3 chipset, which will support the impending Cranford and Potomac 64-bit Xeon MPs and their kickers due in 2006 and 2007.
IBM’s two prior chipsets, the Summit EXA and Summit-II EXA2 chipsets supported both Xeon MP and Itanium processors from Intel. (Well, to be specific, IBM had two slightly different variants of the Summit chipsets, one for Itanium, the other for Xeon.) Because of its desire to sell Power-based Linux and AIX servers against other Unix platforms, IBM has decided not to support Itanium with the Hurricanes.
Withdrawing Itanium support by IBM will not change the server landscape all that much. IBM’s support for Itanium was not exactly enthusiastic, and IBM was in the doghouse this time last year because it was wavering on its support for Itanium. (Moreover, the company’s decision to kill off Project Monterey, which unified SCO Unix and AIX Unix for the Itanium processors, in May 2001 after development had been finished is what has, by and large, landed IBM in its legal woes with Unix vendor and former Linux distributor The SCO Group.)
The Hurricane’s lack of support for Itanium certainly has cooled relations between IBM and Intel, and so has IBM’s whispered statement that it will cease selling Itanium-based systems by the end of 2005.
It is hard to guess what the repercussions might be to IBM, which has a large Pentium 4 and Xeon server business, but you can bet that the person negotiating IBM’s chip and component prices with Intel is feeling a bit nervous about right now.
And you can also bet that Advanced Micro Devices is making a lot of phone calls to Big Blue about right now. After all, an Opteron can do just about anything that a 64-bit Xeon DP or MP processor can do. If relations between IBM and Intel sour enough, IBM could pull a Sun strategy and endorse Opteron for its future xSeries machines. The odds are that it won’t come to that, but stranger things have happened in the computer business.
In a presentation on Intel’s server roadmaps, Talwalker tried to put the best face forward on Itanium and hammered on the idea that Itanium was different from Xeon in that it was focused on RISC/Unix servers. He said that Itanium-based server revenues had tripled in the past year, and that Itanium enjoyed the broadest operating system support of any RISC-class platform (because it supports Windows, Linux, and HP-UX, and it could support AIX and Solaris, which have been ported to Itanium, if IBM and Sun ever delivered these–which they won’t).
Talwalker’s chart during the session said that eight out of nine RISC vendors supported Itanium, but it is really 7 out of 9 with IBM on the outs. He added that Itanium had 2,800 applications and 75 OEM partners, and stressed that Hewlett-Packard, Unisys Corp, Fujitsu-Siemens, NEC Corp, Silicon Graphics Inc, and Hitachi Corp have all signed up to create and promote Itanium platforms.
During his keynote this week, Pat Gelsigner, the other co-general manager of the Digital Enterprise Group, trotted out Jim Allchin, group vice president of Microsoft Corp’s Platforms Division to have him redo Microsoft’s vows to support the Itanium architecture. Actions speak louder than words, Allchin said. We think that Itanium is the scale-up system. He then showed off a new SAP two-tier Sales & Distribution (SD) benchmark on the 32-way Express5800 server from NEC running Windows Server 2003 Datacenter Edition.
However, Allchin also said that Microsoft was just about ready to ship its 64-bit Windows XP Pro and Windows Server editions for Pentium 4 and Xeon machines. In theory, there is now nothing to prevent any one of the Itanium partners from using Xeon MPs to make servers that are arguably just as scalable as an Itanium-based platform. In practice, the wrath of Intel is a pretty substantial thing to consider, which is why none of the Itanium partners (except Unisys, which has supported Xeon and Itanium side-by-side in its ES7000s for years) have thus far committed to building a scale-up server with more than four processors using the imminent 64-bit Xeon MPs.
After his keynote, Gelsinger was questioned about IBM’s waning support for Itanium. It is not particularly surprising that IBM is not enthusiastic about Itanium when Power is their strategy, he said. And when pressed about the limited use of Itanium by Dell, he said much the same thing, that he was not surprised by this and that Dell, being concentrated in the one-, two-, and four- processor server market and on the clustering of machines to replace monolithic servers, would not design a scale-up, big SMP Itanium server.
Both Gelsinger and Talwalker characterized Itanium as being solely a big iron, RISC replacement, which is certainly not what Intel’s glorious plans for Itanium were when it launched the Merced development project back in the mid-1990s. And their contention that Itanium was making headway as a RISC/Unix replacement platform is debatable.
While a certain number of Unix shops have moved to Windows-Itanium platforms and some more are interested in moving to Linux-Itanium platforms, there are nonetheless plenty of Itanium server customers that are greenfield installations–think of Linux clusters in the HPC market, or new data warehouses running on Windows platforms, or new SAP installations on Windows that are replacing homegrown mainframe code–that are not really ports of applications from any platform, including Unix.
Despite the disappointments and questions, Intel remains committed to Itanium. I think this is true because it has a long-term, contractual agreement with HP that requires it to deliver Itaniums for a set number of years (HP has hinted that this is the case) and because it has pumped billions of dollars in investment in the Itanium line, and it is hoping to get some of that money back through product sales where the Xeon cannot match competitive RISC processors.
In the second half of 2005, Intel will roll out its dual-core Montecito Itanium, its first dual-core server chip and, notably, Montecito is a true dual-core chip, not two cores placed side-by-side on a piece of silicon or two cores integrated using multichip module technology.
In 2006, Intel plans to roll out an enhanced Montecito chip called Montvale and another variant called Millington, which is aimed at dual-socket servers. In 2007, Intel has committed to a server platform dubbed Richford that will use the multicore Tukwila Itanium processor and a cut-down variant called Dimona for two-socket servers. Further out, perhaps in 2008 or 2009, Intel will roll out an Itanium code-named Poulson, about which Intel would say nothing other than it existed.
