Is it time for the high-performance computing industry to take another look at the Itanium?
Many observers have given up on Intel's high-end platform as a lost cause considering the broad support that IBM's Power architecture, not to mention Intel's own Xeon family, have received from leading number crunchers around the world. Indeed, it was starting to look like Intel itself had all but shelved the Itanium, with nary an upgrade or product announcement for the past two years.
But all that changed last week as Intel introduced the new 8-core Itanium 9500 featuring the 32 nm Poulson architecture that not only doubles the number of cores over the previous Tukwila models, but ups interconnect speed and supports up to 2 TB of RAM as well. The device pushes clock speeds to 2.53 GHz and contains an error-resistant buffering system that helps bolster performance nearly two-and-a-half times over the Tukwila. However, bulk pricing starts at $1,350 to more than $4,600 per chip.
Long-time Itanium supporter HP quickly stepped up to the plate again, releasing a number of new Integrity servers designed to capitalize on the 9500. The list includes everything from new blades for the c-Class enclosure to a Superdome 2 that sports a redesigned board optimized for the Poulson architecture. At the same time, the company has issued an upgraded version of the HP-UX OS said to provide more efficient resource utilization in many-core environments.
More interesting than the raw capabilities of the new Itanium, however, is the roadmap. It's no secret that Itanium has had a tough time working its way into high-end processing environments — a rather specialized market segment and one that requires a hit product to justify the high development costs. But as speed and power continue to increase on the x86 Xeon line, it hasn't been entirely clear how Intel plans to keep its own enterprise architecture from further reducing the more expensive Itanium's appeal.
One way, of course, is to find a way to utilize the development costs of both platforms using new generations of multiprocessor technology — that is, to foster environments that can take advantage of both Xeons and Itaniums. As tech analyst David Kantner told our sister publication eWeek recently, Intel and HP can more easily justify the ROI on the Itanium if it can be shared with Xeon's. The hope is that customers will see value in gaining the performance boost of the Itanium when it's needed, even as the overall cost of the system is kept low due to the prevalence of Xeons.
And to be sure, Intel has made clear and dramatic steps over the past decade to build a high level of commonality between the two platforms — a sharp departure from Itanium's early days. Much of the core logic is the same on both chips, as well as interconnects, memory and even the chipsets themselves. And Intel recently announced a Modular Development Model that calls for socket-compatibility and a number of shared silicon-level design features to be implemented in the next generation Itanium, code-named "Kittson," expected in 2014. The devices won't be fully compatible, but they should operate comfortably in mixed environments, perhaps handling many of the same software applications
High-end users aren't exactly known for being fickle, however, so it could prove tricky to supplant some of the more established HPC environments out there. But as the lines between high, low and mid-level processing continue to blur, it might be that the Itanium won't need to cut it as a top-end design.
Raw power is impressive, but in today's virtual/cloud environment, flexibility is king. And as long as Itanium can prove itself adept in a wide variety of situations, chances are good that it will find a following somewhere.