Intel takes Itanium towards Xeon's standard
November 12, 2012
HP has introduced a new generation of Integrity servers powered by the Itanium 9500 chips, computers which will start to ship in December. For the HP-UX adopter of migration platforms, the Integrity systems have been high-value, high-performing, and high-attraction computers. The servers are blazing fast and a good value for a high transaction box (something crucial to 3000 migrators). They've also been attractive as in sticky. Because taking steps down the HP Unix path has meant treading the tar-pit of Itanium. No other processor will run HP's Unix.
However, Intel is starting to take its own steps to open up the Itanium architecture. With the Intel 9500 announcement, the chipmaker added that there would be shared technology between Itanium and HP's acknowledged industry standard for processors, the Xeon family.
Future generations of Intel Itanium processors will adopt an innovative "Modular Development Model" that enables deeper commonality between Intel Itanium and the Intel Xeon processor E7 family, from shared silicon design elements to full-socket compatibility. This will provide a more sustainable path for Itanium development and greater design flexibility for Intel's partners.
HP itself calls the Xeon server business Industry Standard. When Intel starts to talk about taking steps to sustain Itanium development, it's a sign that the future being sold to HP customers was wearing thin. Oracle tried to prove as much in its attempted pullout from Itanium development, but a judge ruled against that ideal. However, the evidence submitted for the lawsuit trial showed HP's Project Redwood documents were aimed at shoring up Intel's Itanium interests. The project was proposed before Oracle bought Sun, and Itanium sales have gone nowhere but down since then. Those sales have the advantage, however, of still being far more profitable than all of HP's PC business.
This "but it's profitable" perch provided no safety for HP's 3000 plans during 2001. The 3000's sales and installed base were not growing to Carly Fiorina's satisfaction. And so the customers were given an "end of life announcement." In every company's product line, all products die one day -- at least a death of manufacturing. Then there's some loose cannons that cook up an emulator, and heaven knows when the 3000 will see an end of life.
Even if Itanium growth continues to decline, Intel's fresh plans will let the chipmaker keep developing new iterations of Itaniums. However, they're likely to be more incremental than innovative. Innovation requires marketplace growth. In HP's world, as well as Intel's, growth is Xeon's speciality.
Twelve years ago, HP was not yet announcing Itanium IA-64 plans for the 3000. At that time there wasn't even a clear case inside the vendor's 3000 labs, led by Dave Wilde, to get an 8900 into an MPE/iX box.
Do you think you’ll get to an PA-8900 processor in an HP 3000? That’s the last generation anybody’s willing to talk about in a slide. Will you need all that PA-RISC headroom as you watch IA-64 take shape?
We try to understand our customer needs and work hard to understand the HP roadmap, and work to put those together in a way that makes the most sense for our customers and the business. If an element of that is delivering an HP 3000 on a PA-8900, then that would be something we would obviously do. It’s a little early to talk about availability of the 3000 with that [processor]. Watching is exactly what we’re doing: what happens in the overall market, and what happens in the HP product roadmap.
By 2005, at last, a Unix user could finally adopt an Itanium box that could outperform even the 8900 PA-RISC server.
Easier for partners and vendors
However, when you tease apart that Intel announcement of the Itanium 9500 chip line, you'll see a reference to partners. That's software partners as in app builders, and hardware partners including Itanium in servers. Instead of needing to maintain a separate software design team for Itanium and Xeon (we're looking at you, Oracle), developers might reduce the amount of one-off work they do for an HP Unix application. Whenever Intel gets to "full-socket compatibility," then the Itanium chips have a chance -- not a very big one -- to find their way into the higher ends of non-HP product lines. Because when you get out of the HP and NEC product lines, it's the rare Itanium chip to be found on a system's motherboard.
As part of Intel's announcement of the 9500, it mentioned 15 years of alliance with NEC on the chip's designs. Software partners in the release were Oracle, SAP, SAS and more. But these companies also develop for a much larger customer base that uses Xeon systems. Those are computers which HP is also selling to its migrating 3000 customers: HP's ProLiant systems. By working on a melding of Itanium and Xeon, those software vendors may not be forced to choose between resources for Itanium (HP-UX) and resources for Xeon (Linux, Windows).
Make no mistake about who is acquiring whom in this tech merger. Xeon is the larger entity and so will dole out its tech essentials to the Itanium designers. Don't expect that socket compatibility to be an adoption of the Itanium designs, or the shared silicon designs to promote Itanium's nuances on top of a Xeon empire. The whole Intel enterprise will keep HP from being forced to commit to a port of HP's Unix to Xeon, perhaps. That's $147 million estimated by HP that it will save, and maybe drive into Project Odyssey. But the end result of Odyssey is Linux environments, secured better and hosted on Xeon-based hardware -- with all of your favorite characters from the world of HP-UX. Now that HP's promised the best of HP-UX in Odyssey, Intel's gone and promised the best of Itanium in the Xeon family.
These facts about the future are enough make a customer believe, if they're a HP Unix user who's migrated custom code to an Integrity box, that there's a migration waiting out there. We'd guess anytime after 2016, with a 10-year "end of life" countdown for HP-UX. MPE/iX got a nine-year countdown of that sort. Because every product will see an end of its life on a vendor's price list, after all. Perhaps its best elements will live on in emulation, or integration with newer architecture.