Given Intel's recent announcement that it is changing its server processor roadmaps, pushing enhancements further into the future and, in some cases, cancelling them altogether, is the big chipmaker's strategy in tatters?
Everyone, outside of Intel that is, loves to have a go at the Itanium, not least because the company raised expectations to stratospheric levels when it started developing the chip in conjunction with HP, back in the mid-1990s. It would revolutionise the face of computing, we were told. The reality has been less striking, and the processor has struggled to gain market traction against entrenched opposition from the likes -- initially at least -- of HP's now-defunct PA-RISC, Sun SPARC, and IBM's Power chips.
We now hear that Intel is delaying its first dual-core Itanium 2, codenamed Montecito, which will not be available in volume until mid-2006, rather then early 2006 as promised. The biggest disappointment for those awaiting a chip with more power but less heat -- most hardware OEMs and data centre managers -- is that Montecito also won't include two key enhancements.
The first is a power management technology known as Foxton, which could help contain its power draw. The second is a consequence of the removal of Foxton. The front-side bus, which connects the chip to the rest of the server, was planned to run at 667MHz, compared to today's 533MHz. However, without Foxton's power management, that would simply overheat the chip. An Intel spokesman said the changes were as a result of production-level quality issues.
Meanwhile, future versions are being pushed further into the future. Montecito's successor, dubbed Montvale, will now debut in 2007 instead of late 2006, while Tukwila, planned to be a major redesign -- it might be the Itanium 3 -- comes out in 2008 rather than 2007. There are few if any details publicly available about the capabilities of these processors.
Meanwhile, back at the volume end of the server market, things haven't been going too well either. Intel has scrapped a Xeon design, dubbed Whitefield, only to replace it with Tigerton. Though both are next-generation architecture quad-core systems, the difference between them is that Tigerton is slated to include a new high-speed interconnect, CSI, that allows each core to link directly to the chipset rather than share the front-side bus, as Whitefield would have done, and as today's dual core chips do. It's also rumoured that Whitefield was the first design to be outsourced to a team in India.
The upshot of this is that the Common Platform Architecture that Intel trumpeted only a year or so ago looks to be dead or at least delayed, because CSI was a key element of the design. The aim was to get to a point where the same interconnects and chipsets could be used for either Itanium or Xeon. The result of this would be cheaper Itanium systems, as the greater volumes and simpler designs would allow hardware OEMs to lower prices, thereby giving the high-end chip a much-needed boost.
This strategy -- dubbed 64-bit computing for the masses -- looks now as if it has quietly scrapped, which has implications for the future of Itanium. Its economics cannot be good, as its large silicon area makes it expensive, and it was predicated on much larger volumes than have in reality been attained. What's more, both Dell and IBM have dropped support for Itanium. Although, in its favour, Intel's design partner HP is about to launch a suite of Itanium-bearing blades, the longer term prognosis for Itanium is not good. Analysts have for some time questioned Intel's strategy of designing, producing and selling a separate processor architecture, just to compete with SPARC and Power. The death of the common platform may mean that Intel has to go back to the drawing board.
And the slow-down on the Xeon front means, as one analyst put it, that Intel is still behind AMD, lacking as it does an embedded memory controller and standard processor interconnect. So AMD's multi-core systems are more integrated both at the interconnect and cache coherency levels, which ultimately means better performance for multi-core chips.
There's a lot of catching up to do in Santa Clara. But while we know from experience that the chip giant is not averse to a challenge, AMD now has two, three or maybe four years during which to build a solid base in the server market. You can bet that it will make the most of the opportunity.