Intel set for server chip blitz
From Tukwila to Nehalem-EX
High-end server chip rivals Intel and IBM have picked the same day - next Monday, February 8 - to launch their respective quad-core "Tukwila" Itanium and eight-core Power7 processors.
As El Reg previously reported, IBM is getting ready to launch the initial Power7-based servers in New York next Monday. The Power7 chips will have up to eight cores and Big Blue is promising to offer from two to three times the performance of systems its dual-core Power6 and Power6+ chips with the Power7.
IBM is expected to roll out some of the updated Power Systems line on Monday, with a rolling thunder gradual release through 2010. The word on the street was that the initial Power7 machines were expected in May, but IBM seems to have moved them up to better compete for news and dollars with impending chips from Intel and Advanced Micro Devices.
According to sources at Intel, the chip maker was unaware that IBM was planning its Power7 launch for February 8 when it scheduled the Tukwila launch in San Francisco for the same day. At the event, Kirk Skaugen, vice president of Intel's Architecture Group, will be joined by Martin Fink, general manager for Business Critical Servers within Hewlett-Packard's Enterprise Storage, Servers, and Networking Group. HP, of course, accounts for the vast majority of sales and shipments of Itanium-based systems and is technically a co-development partner for the Itanium design.
Here's the statement put out by Intel:
"Tukwila," the code name for the newest Itanium processor, has begun revenue shipments. The most advanced Itanium processor yet, "Tukwila" more than doubles the performance of its predecessor and adds a range of new scalability, reliability, and virtualization features. Eighty percent of the Global 100 companies already use Itanium-based servers. The launch of this Itanium mission-critical processor is part of a major push Intel is making into the server processor arena, with several announcements slated for the first half of the year.
As 2010 was winding up, Intel sources told El Reg that the Tukwila Itaniums would have a formal launch in the first quarter of 2009. The much-delayed Tukwilas were expected by the end of 2008, but in February 2009 Intel copped to the fact that they had been delayed to mid-2009. And in May last year, these chips - presumably to be called the Itanium 9500s - were pushed out again to the first quarter of 2010. This was done when, during final chip testing, Intel identified a means to optimize the performance on high-end systems by tweaking the chip in some manner that was never really explained.
Not to drag it all up again, but the original Tukwilas were due in late 2006 or early 2007 - and then the memory controllers and other features were changed. The Tukwila processor socket was also made compatible with future "Kittson" and "Poulson" Itaniums rather than with the "Montecito" Itanium 9000s and "Montvale" Itanium 9100s, pushing the chips out to 2008.
The Tukwila chip will weigh in at nearly 2 billion transistors and sports 30 MB of on-chip L3 cache memory. Clock speeds for the chip are expected to range from 1.2 GHz to 2 GHz, with top-end parts burning at 170 watts. Generally speaking, Intel is promising about twice the performance of current dual-core Itanium 9100s, which themselves are a bit of a disappointment in terms of clock speeds with their top-end 1.66 GHz clock speeds compared to the 2 GHz and higher clocks that Intel was whispering about before the Montecito and kicker Montvale chips came to market.
By the way, the official word from Intel as 2009 wound down was that the eight-core "Becton" Nehalem-EX chips - which are aimed at midrange and high-end servers like the Tukwilas and which use the same "Boxboro" chipset (but different processor sockets) - went into production at the end of last year. This was as planned, according to Intel. The company said that parts were shipped to OEMs and that it was "seeding end users." The timing of the Nehalem-EX chip launch remains a bit vague, and it's scheduled for some time during the first half of this year.
As El Reg told you last week, Intel will also next week give out some of the details about the power management features of its "Westmere-EP" processors for two-socket servers and workstations at the International Solid State Circuits Conference in San Francisco.
The Westmere-EP chips, which are tweaked versions of the current "Nehalem-EP" Xeon 5500s that have saved Intel's financial cookies and which will no doubt be called the Xeon 5600s, are expected to be launched at the end of March. Intel has not confirmed this date, but motherboard maker Super Micro has. The Westmeres are expected to have six cores, rather than the four in the Nehalems.
Intel Labs will also present a paper at ISSCC on a research chip that crams 48 32-bit x86 cores in a 2D mesh on a single chip. ®
Tukwila is boring
A chip that is only used for HP-UX and Non-Stop OS. I hope they will be taking questions.
1) Will Unisys, SGI or Groupe Bull have Tukwila systems or did they drop support?
2) Why only four cores when Nehalem and POWER7 are both 8 cores.
3) Why less than 2GHz when Nehalem is 3+ and POWER7 is 4+
4) Why announce a chip and not announce actual systems, what is the point?
5) Wasn't tukwila announced two years ago?
6) Why is HP requiring a fork lift upgrade to Tukwila systems
7) Why is Tukwila finally getting to 65nm when all other chips are 45nm or better?
8) Why is there no core performance increase when software costs are so high and priced per core?
9) How are the Linux sales on Itanium? I hear they are so bad RedHat has dropped support.
10) Will Tukwila finally have hardware virtualization? Itanium seems to be the only processor besides SPARC64 to not provide at least some hardware virtualization.
11) I hear Oracle is going to increase the core price to .75 just like SPARC or higher what is the factor?
12) Is HP going to have a BI offering now that Exadata was canceled and Neoview is not selling?
13) Besides having twice the cores in the same footprint what is the value of Tukwila? The same core and software license performance.
14) Oracle still does not have support for the eBiz suite applications on Itanium..will that ever change?
15) I hear HP is giving away free upgrades to Tukwila, but I don't see the point since applications will not get any more performance.
Re: RE: Tukzila is no new news
I really shouldn't... but Matt makes it too easy. I hat defending others... It seems funny that Matt is using Sun's arguments for CMT now. It's about the system not the core speed! LMOA! Intel admits that the core speed will remain the same and they are just increasing the core count and Matt can defend this? They promised much much more. Come on Matt, even a Fanboy such as you can scream foul.
"Well, if Tukzilla is "no new news", maybe that's because Intel and hp have been keeping the market well-informed of what was coming, whereas with IBM it's all up-in-the-air guesswork."
Over 2 years late, lower MHz, and fewer cores than promised... Hmmm... Who's keeping who up to date?
""...SGI....bla,bla,bla....Unisys....bla,bla,bla...." Erm, so what? I don't have any of either of those vendors' systems. I don't have any Bull boxes either. I doubt if many hp customers do, so why should hp or their customers worry?"
Who cares? Maybe Intel? Intel sells volume. One vendor does not equal volume, or HP would have just developed the chips themselves. Intel cares very much that no one besides HP uses Itanic. That's why HP has to keep pumping bribes into Intel to keep developing this stillborn chip.
""All virtualisation has overheads, honey, even hardware virtualisation. It's kind of a law of reality that you can't get anything for nothing. Otherwise it wouldn't be virtualisation, it would be magic! Please do not try and tell us that hardware virtualisation has zero overheads as such amusing statements will only get you the moniker of Mrs Kebabfart."
The original poster was reaching, really... However, HP does have more overhead in their virtualization than all of the other vendors, and IBM has a LOT of overhead, so that's saying something. Of course, x86 virtualization does not have much to be proud of either. Really,
when it comes down to it, who cares. As long as you can squeeze more out of less, the better.
"Sorry to disappoint, but we have it in production, with VMs in clusters, running biz crit apps. The only time I have seen a similar level of belief in any IBM virtualisation product it was on mainframes, not pSeries."
I almost didn't comment here 'cuz it was such nonsense. Saying that HP's virtualization was trusted more than IBM's? Come now... You're really serious? They're really about the same in my opinion...
"Maybe, but then hp has a history of innovating where others said it couldn't be done. "
HA HA HA HA HA!!! Are you serious? HP hasn't innovated anything since selling off PA-RISC. HPUX has some advantages, but come on now... Your example of HP innovating is an Intel chip being put on a card?
""....Let's not confuse Xeon with Itanium....." Ah, but then you don't want to admit the two are linked. For years, Intel have been taking RAS features from Itanium and putting them into Xeon, and the economies of scale of Xeon have allowed Intel to fund Itanium and cut the costs of Itanium systems."
You insist on saying that Itanic will innovate because Intel innovates with Xeon... They are linked in that Intel wants Xeon to take over the Itanium market. That's it... That's the only way that they are linked. When Intel drops Itanic and HP is forced to move their customers once again, don't say that we all didn't say we told you so. Even SPARC has a bigger market than Itanic. Intel does not have as much patience as you think.
"Goes rather nicely with MS SQL, especially on hp ProLiant (SAP's most common platform)."
Who runs SAP on MS SQL? Now you're just being silly. Note to Matt - SAP is for running real companies, not the company you seem to want us to believe you work for. Once again, you're saying that Itanium is great because Proliant is successful. Interesting how your mind works.
HEY MATT!!! This subject is about ITANIC! Stop changing the subject!
""....eBiz will never be supported on Itanium and is fully supported on all Power systems...." Considering Adam's statement from above, it would seem that you are far more aware of Power than Itanium. Maybe you should try reading up on your opposition before jumping in with both feet."
Avoiding the subject are we Matt? Hmmm... Really I don't think it matters if eBiz runs on Itanic or not, but your avoidance of the subject speaks volumes. Itanic has fewer and fewer apps. Note to Matt - Applications matter. If you don't run 'em, then no one will run on your hardware. HP and Intel will have to pump more money than they are now into vendors to keep developing for their platform.
Moving forward there will be three platforms - X64 (Intel volume) - Power (IBM) - SPARC (Oracle)
Of course, SPARC definitely has less legs than x64 and Power, but they all have more chance than Itanic.
There is a lot of supposition going on here. I can help a little on the Power 7 stuff:
- IBM had advised that Power 7 servers were a March intro for several months. Moving the announcement up to February is a surprise.
- The allegation the IBM will bring out small machines with defective cores FIRST does not match with IBM open announcements. IBM set expectations that mid-high servers would come earliest.
- IBM's 4 GHz RISC core will well-outperform INTC's 2 GHz RISC core. Since IBM packs 8 on a chip, it would be reasonable to expect that 8 *4GHz is four times the throughput of 2*2GHz from an Itanium chip. We will see.
- IBM is announcing SYSTEMS and INTC is announcing CHIPs. There isn't much doubt about that. Itanium built-out Systems should be announced quickly after INTC's announcement, though.
- 32nm or 25nm Itaniums have not yet been announced by INTC. There are IBM and INTC announcements for 32nm chips. INTC will be first (6-15 months). IBM will be later (24-36 months). This is only relevant if it affects high-core offerings or high-GHz offerings that deliver important performance advantages.
- Itanium hardware virtualization hasn't appreciably improved in over six years. There needs to be INTC guidance for that. IBM and INTC 32-bit hardware virtualization has increased significantly/dramatically in the past five years.
- Oracle matters. We will have to see what they do. They are 'playing defence' with all the Sun stuff that they bought, though.
- Even Itanium fans are on edge with this delayed intro. Itanium needs a smooth launch to have any chance at a further 5 year's existence in the CPU-wars.