backtop


Print 31 comment(s) - last by Ammohunt.. on Apr 18 at 5:24 PM


Intel's sales forecasts (blue, green) for Itanium were vastly greater than the platform's actual sales (orange).  (Source: Arch Dude)

Itanium's Tukwila architecture landed 3 years late, and is pretty pricey.  (Source: Xbit Labs)

Even as Oracle, Microsoft, and Red Hat abandon support for Itanium, Intel and its partner HP plan to go down with the ship.  (Source: Flickr)
"I'll never let go, (Itanium). I'll never let go."

Little known to most, Intel Corp. (INTC) actively supports a rival architecture to x86.  Built on the Explicitly Parallel Instruction Computing (EPIC) reduced instruction set computing (RISC) architecture, Intel's Itanium CPUs fulfill a variety of niche needs from high-end medical data processing to financial sector applications.  The architecture's inherently parallel design makes a perfect fit for certain kinds of applications and has led to consistent, low-volume sales for Intel.

Lately, however, enthusiasm for the task of supporting Itanium has waned and several software partners have jumped ship, including Oracle Corp. (ORCL).  Most recently, Microsoft Corp. (MSFT) and Red Hat, Inc. (RHT) announced that they would be killing support for Itanium in future versions of their operating systems. 

With no clear OS provider lined up, some have speculated that Intel may preemptively kill Itanium and focus efforts on its Xeon line.

But in the wake of its most recent Xeon launch, Kirk Skaugen, VP and general manager for Intel's Data Center group, commented at a press conference, "Instead of Itanium at the top and Xeon at the bottom of the lineup, we're going to have them side by side."

I. An Inglorious History

Whereas the Xeon line -- the direct server counterpart to Intel's consumer PC chips -- was widely lauded, Itanium was negatively received since its launch.

Intel had originally hoped for Itanium to replace x86 as its parallel-optimized, 64-bit addressable architecture.

But the project just didn't turn out well.  When Intel released its first Itanium design, codenamed Merced in 2001, it only mustered half the performance of Intel's mainline consumer/server architecture of the time, Pentium 4.  And according to some reports, x86 emulation tools were so poor that the chip performed like a 100 MHz first-generation Pentium processor.

Undeterred, Intel mustered onward.  Firmly behind it was the world's largest computer maker, Hewlett-Packard, Comp. (HPQ).  Not only did HP build Itanium systems and try to push them in the corporate IT ecosystem, but it also began to co-develop the chips with Intel.

The first joint design was the "Itanium 2" codenamed McKinley, released in 2002, built at 180 nm.  In 2003 Intel released a 130 nm Itanium 2, codenamed Madison.

Then it began to make big claims in 2005 about a design called Tukwila, which would not only pack four cores but would also replace Itanium's bus.  At the time Intel's first generation PC-side Core architecture (codenamed Conroe/Merom) was still a year away, so this claim sounded pretty exciting.

But 2007 rolled around and the chip never came.

In 2006 Intel had released the chip Itanium 9000 Series, codenamed Montecito and in 2007, rather than the promised Tukwila, it delivered the Itanium 9100 Series, code-named Montvale.

Sales reflected this poor performance.  In 2007 8.4 million x86 servers were sold, but only 55,000 Itanium servers shipped, according to Gartner, Inc.  This was bad, even for RISC-servers, which shipped 417,000 units in total.

While exact numbers are sketchy at best, due to lack of official reporting, it is estimated that Itanium was heavily outsold by Sun Microsystems (now owned by Oracle) SPARC chips and International Business Machines' (IBM) POWER architecture.

In February 2010 Tukwila finally landed, officially branded the Itanium 9300 Series.  Three years late, the processor still sported some tempting features like double-device data correction to prevent memory corruption.  These features made it a promise buy for mission-critical applications.

But for the mass market, Tukwila was late to the party and lacked the performance to truly impress

Famous tech news figure John C. Dvorak wrote in 2009, "[Itanium] continues to be one of the great fiascos of the last 50 years" in his piece "How the Itanium Killed the Computer Industry."

II. Tukwila v. Xeon

With the launch of Intel's latest "EX" Xeon processor line, a 32 nm die shrink of Nehalem codenamed Westmere-EX, Xeon had a new chip that could match the memory bandwidth of Tukwila.

A 4-core Itanium processor clocked at 1.73 GHz costs $3,838 USD per chip in quantities of 1,000, while the newly released 10-core 2.4 GHz Xeon E7-2870 costs $4,227 USD, in quantities of 1,000 (note: due to the different architectures, the clock speeds are not directly comparable). 

Given the comparable price it's hard to see the merits of Itanium, which packs less cores, other than from an error-correction/mission-critical standpoint.  Given the vanishing software support for the platform, the picture becomes even dimmer for the platform.

Intel is promising a cheery outlook for the future of Itanium.  Mr. Skaugen bragged that the next generation Itanium core, codenamed Poulson, would bring 8 cores built at 32 nm and more than double the performance of the Itanium 9300 Series.

And an even more distant chip, codenamed Kittson is also reportedly being worked on.

The problem is that even Westmere-EX is a questionable proposition.  Given the availability of 12-core Opteron 6168 chips for $750/USD, the much more expensive 10-core E7-2870 becomes a tenuous buy.

And even among Intel processors, the E7-2870's higher cost per core count is cause for pause, given that there are far cheaper Sandy Bridge-based quad-core Xeon chips (the recently released E3 Series).  The situation is complicated slightly by the fact that some software is still licensed on a per-chip basis, meaning that the E7 chips would yield a net savings if you look exclusively at Intel's offerings.  

Given the switch of many software vendors to per-core (rather than per-chip) licensing, the E7 may find less traction, though, than its E3 series brethren.

And the situation is even worse for Tukwila, which is even more expensive and by Intel's own admission trails the E3 and E7 series in performance.

III. Intel and HP -- Resigned to go Down With the Ship?

Given the poor outlook, it seems pretty incredible that Intel would continue to devote resources to the eternally disappointing Itanium project.  The project recently observed its 10-year anniversary, but given how Intel's ever-shrinking sales predictions continue to overestimate the Itanium 's dismal real-world sales, it was hardly cause for the chipmaker to celebrate.

With the SPARC and POWER RISC-style architectures deeply entrenched and a hungry ARM coalition waiting in the ranks Itanium's future is not bright.

But Intel has clearly stated that it remains loyal to its post and will stand by the platform for years to come -- even if it continues to sink.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: What?
By snakeInTheGrass on 4/11/2011 8:58:47 PM , Rating: 2
The E7-2870's cost per core causes pipeline stalls across all TLBs due to its inability to pre-flight enough instructions when dealing with too many dollars. Unless some per-chip software remains exclusive to Intel, not ARM. Look.

What? Hey, I tried, OK?


RE: What?
By SilthDraeth on 4/11/2011 11:36:08 PM , Rating: 2
LMFAO!!!

Great translation.


"I'd be pissed too, but you didn't have to go all Minority Report on his ass!" -- Jon Stewart on police raiding Gizmodo editor Jason Chen's home














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki