Intel drags feet on Itanium quad-core (again)
No Tukwila this Tuesday. Or even this year
Are you one of the IT shops with Itanium-based servers at the heart of your data center expecting a substantial speed boost this year from Intel with the quad-core "Tukwila" Itaniums? You'd better forget about it, because the processor won't be making an appearance until the first quarter of 2010.
As we reported earlier this week, Intel is hosting a media event on May 26 after the Memorial Day weekend and has said that it would talk about "the next evolution in high-end server architecture", and that it thinks this architecture "raises the standard in cost-effective RISC replacement solutions".
Considering that Intel always pitches Itanium as a mainframe and RISC/Unix alternative - and also taking into account that back in February Intel delayed the Tukwila chip so it could be given a new socket common to the future "Poulson" and "Kittson", and said it would have this (once again revised) Tukwila to market in mid-2009 - you would infer from Intel's terse invitation to the press that Tukwila was at least one of the topics it would discuss ahead of a product launch.
As it turns out, while Intel will no doubt be fielding questions about Tukwila next Tuesday, what it really wants to talk about is the eight-core "Nehalem EX" processors for high-end servers. The rumor is that this chip will not be available until the first quarter of 2010, either.
What all of this means is that Intel is trying to get people to think about its future high-end Itanium and Xeon chips just ahead of when Advanced Micro Devices is getting set to roll out its six-core "Istanbul" Opteron processors months early. This would be laughable were it not for the fact that no one is building a big scalable Opteron server that can compete with the Integrity-class SMP servers from Hewlett-Packard, which is where most Itanium processors end up.
So AMD can get powerful chips into the field, but has no big SMP partners; and Itanium has big SMP partners, but they will have to make do with dual-core "Montvale" Itanium 9100s. By even generous estimates, Tukwila has been completely redesigned at least once and is going to be at least three years late - that's almost as bad a reputation as the original "Merced" Itaniums had. The difference is that most people believe that Intel will be able to deliver at least twice the performance with Tukwilas as it has delivered with Montvales. Merced was a total dog in terms of performance, and basically unusable.
According to Gordon Haff, an analyst at Illuminata who broke the Tukwila delay story, Intel has issued a statement to his fellow analysts. It doesn't say much:
During final system-level testing, we identified an opportunity to further enhance application scalability best optimized for high-end systems. This will result in a change to the Tukwila shipping schedule to Q1 2010. In addition to better meeting the needs of our current Itanium customers, we believe this change will allow Tukwila systems a greater opportunity to gain share versus proprietary RISC solutions including SPARC and IBM POWER. Tukwila is tracking to 2X performance vs its predecessor chip. This change is about delivering even further application scalability for mission critical workloads.
At press time, Intel's home office in Santa Clara was not yet awake, and Intel UK's press office did not return requests for comment.
In effect, the Itanium chip that was supposed to take over the server space and replace myriad RISC and CISC processors - particularly HP's PA-RISC and Compaq's Alpha - has become a more or less proprietary processor for HP manufactured by Intel. Many vendors that had (perhaps fearfully) embraced Itanium in the late 1990s when it looked pretty impressive on paper have dropped it over the years.
"A special cause for optimism"
Sun Microsystems was first, then Dell and IBM, and now Unisys is on the fence and emphasizing Xeons and mainframe iron. NEC and Hitachi are awfully quiet about their Itanium plans as they are suffering huge losses, and Fujitsu has put it plain as it ate the Siemens systems business in Europe at the end of March that Xeon servers would be its preferred platform, with Sparc and mainframes getting honorable mentions and Itanium getting less than that.
No wonder, then, that a few weeks ago the Itanium Solutions Alliance, the club of Itanium hardware and software suppliers that acts as the official cheerleader for the Itanium chip, was saying that the Itanium was "a special cause for optimism". This was the ISA doing damage control ahead of the delay they knew about.
I don't think Itanium is going away, but I do think that continuing delays with Tukwila are hurting HP. As did the delays in IBM's Power5+ and Power6 processors, and the anemic performance boost that IBM got from Power6+, as did Sun's delays and then killing off of the "Millennium" UltraSparc-V and the delays with the "Rock" UltraSparc-RK processors.
Fujitsu's Sparc64 roadmap has been rewritten lots of times, too, and that hurt both Sun and Fujitsu. The good news for Intel and HP is that everybody seems to be screwing up their chips at the high-end; this, of course, is situation normal in the chip business. Companies are always tempted to add too many features, and then the chips don't work.
By Haff's back-of-the-envelope estimates, Intel is generating somewhere between $1.5bn and $2.5bn in Itanium processor sales a year, with about 100,000 chips being pumped out each quarter and chips ranging in price from between $2,000 and $3,000 as they ship to OEM server makers. And the Itanium Solutions Alliance is pegging Itanium system sales at around $5bn per year (these being dominated by HP's Integrity line, of course).
"If we were talking about Xeon, this would be an unmitigated disaster," says Haff, quite correctly. "But Itanium is not a hellacious money pit in terms of investment and they will have to soldier on."
This is particularly true of HP, whose HP-UX customers have gone through an eight-year ordeal getting their applications ported to Itanium, whose OpenVMS customers are in the process of moving to Itanium, and whose NonStop servers actually need hardware features relating to system reliability that are on Itanium but are not on Xeon chips. HP has said that it has not ported HP-UX to Xeon chips, and while this may sound stupid in terms of the long-term survival of the HP-UX Unix distro, I actually believe HP when it says it doesn't have a skunkworks somewhere with HP-UX running on Xeons.
How smart or stupid this strategy is, time will tell. ®