Itanium Proponents Say Ecosystem Is Growing
Published: February 23, 2006
by Timothy Prickett Morgan
It is perhaps ironic on an absurd scale that the 64-bit Itanium processor, which was pitched in the 1990s as a means of bringing RISC and Pentium architectures together and creating a united 64-bit platform for Unix and Windows and other platforms, for many years only succeeded in uniting all but a few staunch supporters. If platform providers were united, it was in their vocal disdain for the Itanium chip, which was late, underperforming, and incompatible with existing 32-bit Pentium and Xeon processors.
In the past six years, the Itanium has been the whipping boy of the big RISC/Unix server suppliers, such as Sun Microsystems and IBM, which originally had big plans for Itanium and then dumped it, as well as more neutral players like Dell. The advent Itanium chips with acceptable performance with the "McKinley" generation in 2001 was nonetheless not able to prevent AMD from making plenty of hay in the server space with its 64-bit Opteron processors and their Xeon-compatibility and X86-64 64-bit memory extensions. As Intel was forced to counter with its own (and AMD-compatible) 64-bit memory extensions, the company has spoken less and less about the Itanium processor, leaving the job of spokesperson to Hewlett-Packard, which joined the Itanium development effort in 1996 and which is probably the cause of much of the incompatibility that went into the Itanium chips. That's a guess, but the EPIC architecture is largely the brainchild of HP, and while it is technically elegant, it is not compatible with Pentium and Xeon processors. Had Intel not partnered with HP to co-create Itanium, I believe that the Itanium would have looked a lot more like today's 64-bit Xeons and Opterons and a lot less like HP's then-future plans for its PA-RISC processors.
Last year, the many Itanium supporters banded together to form the Itanium Solutions Alliance, which took on the job of promoting the Itanium architecture and trying to ensure the enterprise computing space that the Itanium platform not only has a big future, but that they are investing collectively to make that future happen. The founding members of the Itanium Solutions Alliance include Bull, Fujitsu, Fujitsu-Siemens Computers, Hitachi, HP, Intel, NEC, Silicon Graphics, and Unisys. The charter members include BEA Systems, Microsoft, Novell, Oracle, Red Hat, SAP, SAS Institute, and Sybase.
To make the point that vendors were investing in the Itanium architecture and intended to make money on that investment, the Itanium Solutions Alliance held a big shindig a few weeks ago to talk investments and revenue opportunities. The key founding members of the alliance said they had committed $10 billion in collective investment in Itanium-related technologies, development, marketing, and sales to chase an Itanium server market between now and 2010. Over that same five-year time, according to Steven Howard, HP's representative for the Itanium Solutions Alliance, the alliance expects the mission-critical server market estimated by IDC to be worth $140 billion. By the way, these sales numbers are estimates for the servers alone--not extra storage, systems and application software, and services, which have a large drag-through for Itanium-class machines. The important thing, according to the alliance, is this: That $140 billion sales estimate for the mission-critical server opportunity is larger than the opportunity for so-called volume servers, which usually means X86 and X64 servers. The issue now is how much can Itanium take out of that $140 billion pie.
Approximately 70,000 Itanium servers have been sold through the end of 2005, with hundreds of thousands of Itanium processors sold into these boxes. While Itanium has not taken over the server market from top to bottom as Intel and the IT analyst community were predicting 10 years ago, and while the Itanium chip has been relegated to a high-end niche market (for very complex reasons) that makes it only a marginally more volume product than HP's PA-RISC chips might have been, the fact remains that the Itanium platform has 10 operating systems running on it--more than any other chip architecture--and has over 6,000 applications running on it and over 75 platform partners. And, according to Howard, in the Japanese market, the Itanium platform--driven by NEC, Fujitsu, Hitachi, and HP--has overtaken both the Sparc and Power platforms in terms of aggregate revenue, and worldwide, Itanium sales are about 58 percent of the Sparc platform and 33 percent of the Power platform--and growing.
HP is, of course, the poster child for the Itanium migration, in that it ported HP-UX, OpenVMS, and NonStop to the platform and was a key partner in getting Windows and Linux working on the chip. Howard says that in fiscal 2005 (ended in October 2005), HP had $1.6 billion in Itanium-based Integrity server revenue, up from about $1 billion in 2004, according to Rich Marcello, general manager of HP's Business Critical Systems unit. In the first quarter of fiscal 2006, HP saw Integrity sales increase by 91 percent, which means Integrity server sales are accelerating. While the Itanium's many detractors in the server space can crab that HP is seeing sales of Alpha, PA-RISC, and Tandem platforms drop even as Integrity takes off, that is what you'd expect to happen during a product transition. The problem with the Integrity line is that the machines are relatively expensive compared to 64-bit Xeon and Opteron servers, which, had the Itanium been architected differently, might not exist. And IBM has been very aggressive in its Power-based AIX and Linux server pricing, which it can do because it charges a lot for Power hardware and software in its iSeries proprietary server line. Take away that proprietary prop, and maybe IBM could not be so generous with its Unix systems.
Howard says that HP estimates that about one-fifth of its installed base of AlphaServer and PA-RISC servers have been converted to the Integrity platform, and obviously, with dual-core "Montecito" processors coming in mid-2006 or so, delivering about two times the performance and about two-and-a-half times the performance per watt of the current single-core "Madison" Itanium 2 processors, HP is looking forward to seeing Integrity sales accelerate. "Early performance testing is bearing out Intel's initial performance estimates--and on real workloads, not just highly optimized benchmarks."
Windows is also helping. According to Marcello, sales of Windows-based Integrity servers grew at approximately 60 to 70 percent in 2005 and Unix sales were growing at from 50 to 60 percent. "And we think that we will now begin to start selling more Superdome-class boxes running OpenVMS now, too," says Marcello. "But don't get me wrong. This is not a huge growth business for us, but the people who are on OpenVMS pretty much want to be there forever." What has been popular on Itanium processors is Linux, particularly in the high-performance computing space. Early on, Linux was the dominant operating system on HP's Integrity line.
The dual-core Montecito Itaniums are, of course, running late, as Itanium chips tend to. The Montecito chip has two full Itanium cores and two 12 MB L3 caches on-chip, with a total of 1.72 billion transistors. Each Montecito core has 16 KB of L1 data cache, 256 KB of L2 data cache and 1 MB of L2 instruction cache, plus 12 MB of L3 cache. Intel is not unifying the L2 or L3 caches, as it could have done. Each core will have two virtual threads, enabled by HyperThreading, yielding four virtual threads per chip socket. The 667 MHz front side bus that was expected for the high-end Montecito part has been tossed out the window with "Foxton" over-clocking feature, and the Montecito will only come with 400 MHz and 533 MHz front side bus speeds. The Montecito chips will not have support for X86 emulation, which was included in prior Itanium processors and which ran at only 50 percent efficiency. The Montecitos will have "Vanderpool" VT virtualization features (but will no longer beat the Xeons to market with this feature), as well as "Pellston" error correction technology, and DBS power management features. When running at 1.6 GHz, the initial clock speed Intel can deliver (which is lower than expected), a Montecito chip is expected to consume about 100 watts.
In a recent survey of 500 IT managers by IDC, about a third of the respondents said that they would buy an Itanium server in the next 12 to 18 months. Among HP customers, the propensity to buy an Itanium machine in that timeframe shot up to around 40 percent, which makes sense considering that HP accounts for 69 percent of Itanium-based server sales today. In that same study, IDC projected that Itanium server sales would grow at a compound annual growth rate of 35 percent to reach $6.6 billion in 2009, up quite a bit from the $1.4 billion in sales in 2004. Over that same 2004 to 2009 timeframe, IDC expects the overall server market to grow at a compound annual growth rate of only 3.4 percent.
While Itanium might not have done as well as many had been hoping a decade ago, it is clear that the expected growth rates for Itanium--and the profit margins these machines will engender--are a lot more attractive than the Xeon and Opteron server markets. The next five years look to be a lot better for Itanium than the past five years. That much is for sure.