Quocirca's Straight Talking: Itanium - what's in a name?

Iceberg dead ahead?

Reminiscent of Titanium, the incredibly strong yet lightweight element, it was perhaps not too surprising that soon after its less-than-stellar launch Itanium was rechristened "Itanic" by wags, after the ill-fated White Star liner Titanic.

The symbolism present in these rival names is perhaps worth exploring. Depending on who you talk to, Itanium is either fatally holed below the waterline and taking in water fast, or else happily afloat and like the Good Ship Lollipop travelling to a candy land where bonbons play on the sunny beach of Peppermint Bay.

An abundance of hype accompanied the launch of both Titanic and Itanium.

The former was designed to compete with Cunard's fast Atlantic liners and in its day was the largest, most luxurious ship ever built. Considered a pinnacle of naval architecture and technological achievement, it was hailed by The Shipbuilder magazine as "practically unsinkable".

Similarly, during its extended development, Itanium was widely expected to become the dominant processor architecture for servers, workstations, and perhaps even desktops, replacing the ubiquitous x86 architecture to provide an industry-standard architecture across an unprecedented range of computing platforms.

Unsurprisingly, both failed to live up to the pre-launch build-up - Titanic because its watertight bulkheads did not reach to the full height of the decks; Itanium due to a lack of optimised software, yield problems, high manufacturing cost and the fact it ran existing x86 code woefully slowly.

Both have also generated their fair-share of legends, myths, and controversy. What was the band playing when the ship went down? Was it the first time that SOS had been sent? Did the Aberdeen Press and Journal (a paper notorious for its parochial coverage) really report the sinking with the headline 'Aberdeen Man Drowned'? Having to sort the myth out from the reality makes it all the harder to get to a true view.

Did HP's handing over of complete design responsibility for the platform to Intel at the beginning of 2005 show its lack of commitment, or was it just a culmination of HP's move out of processor design to allow it to focus on platform development? Given HP's continued commitment to Itanium in its HP-UX, NonStop and OpenVMS business critical server platforms, the latter seems the more likely explanation. Some believe any company that hitches its wagon to the Itanium star is doomed to failure, however the recently formed Itanium Solutions Alliance (comprising Bull, Fujitsu, Fujitsu-Siemens, Hitachi, HP, NEC, Silicon Graphics and Unisys, together with Intel and a number of software vendors) have committed themselves to spending $10bn on individual research and development projects between the beginning of 2006 and the end of 2010.

Those are big numbers by anyone's standards (double what IBM plans to spend on mainframes over the same period) and with so much at stake for both proponents and detractors the level of controversy surrounding Itanium is hardly surprising.

The Titanic disaster led to the convening of the first International Convention for the Safety of Life at Sea, and the formation of the International Ice Patrol. It also led to changes in passenger ship design, including the use of double hulls, and the provision of enough lifeboats for everyone on board.

You what…?

Bust through tech jargon with silicon.com's Cheat Sheets - including this one on Itanium 2.

In response to the performance issues of the original Merced version, the Itanium 2 (McKinley) released in 2002 was substantially more powerful, with roughly double the performance. The most recent generation of Itanium 2, a dual-core version ('Montecito') has double the performance of the previous, single-core version, and about 3.4 times the performance of the original Itanium 2 processor. Addressing the lack of applications, the Itanium Solutions Alliance recently announced that more than 10,000 applications have been ported to various operating systems on the Itanium processor, a 50 per cent increase in the application portfolio in the past year, and should reach 12,000 to 13,000 applications by the end of 2006 if the trend continues.

Coupled with these changes, there has also been a repositioning of Itanium. In response to AMD's introduction of the Opteron chip that ran both 32-bit and 64-bit software, Intel was forced to follow suit with a hybrid 32/64-bit version of Xeon, and to recognise that these would occupy the lower and middle ground of the server landscape. No longer the architecture to span the entire server spectrum, Itanium has retreated to the high-end of the market where it competes with the biggest Risc iron providing a very high level of parallel processing, to enable high performance without high clock frequencies (thus reducing power and heat requirements). Granted, it may have replaced partners' Mips Risc architecture as well as HP's own Alpha and is currently eating its way through the PA-Risc market but how is it doing against the competition?

Collectively, Itanium platforms generate just under half of the revenue stream that IBM's Power and Sun Microsystems Sparc platforms do in any given quarter but it is growing at a much higher rate than the Power and Sparc platforms (close to 40 per cent annually).

Itanium has its detractors - Sun president Jonathan Schwartz has condemned it to a "lingering death" and Linus Torvalds is no fan either (but then he was employed by Transmeta, which licenses AMDs architecture). Even HP and Intel would admit times have been challenging for their problem child. But through it all, Itanium allies maintain their optimism. Have Intel and its partners done enough to turn it around, or is it a case of too little, too late?

Why did Intel persevere with Itanium instead of killing it off years ago? Well, it would have upset partners who had invested in it so heavily, leaving their customers stranded. Then again, why should Intel kill it? It builds Itanium in the same fabs as the x86 parts but on silicon technology that is one generation behind the mass-market products (the current Montecito version is manufactured using 90nm technology, whilst Intel's mass-market products are on 65nm). This means that instead of having to invest in building new fab facilities (at a cost of $2.5bn-plus these days), Intel can use facilities that have already been used and amortised. It also gains economies of scales over its Risc rivals who have to build dedicated facilities for a more limited production run. Last but not least it also provides Intel with a "sand-pit" to play in - enabling it to try out new ideas and techniques on a more limited scale before introducing them into its mass-market products.

Intel's official Itanium roadmap includes a next-generation dual-core processor, manufactured using a 65nm silicon process technology, and future quad-core designs. If it can stick to its delivery targets and make its strategy work, there is a compelling argument to say that there is place for Itanium to survive - if not thrive - in the server ecosystem of the future. Perhaps, to misquote Mark Twain, rumors of Itanium's death may have been greatly exaggerated.