IBM to Drop Itanium

Hack Jandy writes "Xbitlabs is reporting that IBM chose not to persue Itanium in their next generation server lineup because of the "market acceptance issues" of the platform. They will still continue with new revisions of Xeon servers, however. With IBM's investments in Power, I can't help but think the writing was already on the wall. The article also hints that IBM might start using Power in their high end server products."

Comments

Stick a fork in it, it's done

[drinkypoo]

No one outside of certain specialized environments that demand loads of floating point is interested in itanic. Flush the damn thing already and work on EM64T, will you intel?

WTF does "The article also hints that IBM might start using Power in their high end server products" mean anyway? The processor is called POWER, and IBM already uses it in their high-end server products, like the ones that used to be called RS/6000. As for Power, well, show me a transistor that works without it.

Re:Stick a fork in it, it's done

[IGnatius T Foobar]

IBM already uses it in their high-end server products, like the ones that used to be called RS/6000.

Actually, that hardly does it justice. pSeries (formerly RS/6000), yes, but also iSeries (formerly AS/400) is now POWER. The new OpenPower line of systems from IBM can run AIX, i5/OS (formerly OS/400), and Linux. In fact, it can run them simultaneously thanks to IBM's really good server partitioning technology (you can partition down to 1/10 of a CPU!).

I'm currently doing some development work on one of these boxes (running Linux on POWER) and let me tell you, it just smokes. Runs circles around Itanium, even before you start parallelizing (which is usually the case, since you're always going to have a dual-core chip, maybe even several of them).

IBM has absolutely no reason to continue supporting Itanium. It doesn't buy them anything. Itanic is an architecture nobody wants. If Intel hadn't sank so much R&D into it while still being able to live off the revenue from their 32-bit processors (and now, their AMD64 clones), Itanium would have been shelved a year ago.

From the summary:

[Noose For A Neck]

"The article also hints that IBM might start using Power in their high end server products."

What? IBM already uses POWER in it's high-end server products. What do you think they develop it for, anyway?

Re:From the summary:

[superpulpsicle]

http://www.llnl.gov/computing/tutorials/ibm_sp/

Here's a link to the history of IBM's processors POWER. This is one of the best sites out there IMHO, and it still seems mighty confusing.

IBM never had a good history of marketing their processors like Intel and AMD. They fight competition with raw numbers.

Re:From the summary:

[drinkypoo]

IBM doesn't market IBM products. They market IBM. You buy into IBM, you do everything their way, and everything will work. These days, they add "...and you will get the best possible performance" to that, too. Nothing out there has the performance of the latest-generation POWER processors. As IBM has been busy proving building supercomputer after supercomputer, it scales pretty well too :) Granted, highly parallel opteron processors are pretty slick, but given a level playing field I know what I'd pick. Intel introduced itanium too late and at too high a cost to make enough inroads before opteron started to take off, and now itanic has no chance to proliferate enough to ever become inexpensive. IBM has been putting work specifically into making their cores as modular as possible so they can easily turn them into other versions ever since the PowerPC 601, which is why we new PowerPC cores so closely follow the release of new POWER cores.

Re:From the summary:

[Henriok]

What? IBM already uses POWER in it's high-end server products. What do you think they develop it for, anyway?

No they don't!
the pSeries and iSeries isn't considered "high end" by IBM, they are considred low end and midrange servers. The high end is the zSeries and they doesn't use POWER/PowerPC processors just yes. Word has it that the future POWER6 processor will converge the three server lines on one processor platform. The eClipz project is tied very closely to this. "e" as in eServer, "l" as in Linux, "i" as in iSeries, "p" as in pSeries and "z" as in zSeries will.. eclips the Sun.

Re:From the summary:

[macshit]

The eClipz project is tied very closely to this. "e" as in eServer, "l" as in Linux, "i" as in iSeries, "p" as in pSeries and "z" as in zSeries will.. eclips the Sun.

I knew IBM had good research people, but I never realized their acronym technology was that far advanced! My god, with an acronym like that ... their competitors might as well just give up now and save themselves the embarrassment.

I'll miss it

[m50d]

Say what you like, but Itanium was a nice architecture. The compiler is the proper place for the optimisations, the processor should be left to do the actual processing. It's still the most efficient way I know to do raytracing or anything multimediay, and I predict there will still be a market for them for some time.

On second thought, maybe they'll start appearing cheaply on ebay. That'd be nice.

Re:I'll miss it

[Schweg]

With Itanium, Intel attempted to tackle a set of issues that isn't new, and other companies have tried (and failed at) before. It's hard to shift almost all of the burden of optimization, because it's a case of "early optimization" (the root of all evil). Optimization by the processor at run-time allows one to deal with data-dependent issues, and base decisions on statistics gathered by modern processors (such as branch history, caching behavior, etc). Intel made a good try at it, but ended up making a very power-hungry processor that exposed a lot of complexity to the programmer, and whose advantages compared to other processors on the market were not very clear.

Re:I'll miss it

[Waffle Iron]

The compiler is the proper place for the optimisations, the processor should be left to do the actual processing.

On the contrary, the compiler has no insight into the actual run-time behavior of the current dataset, and compiler development can lag updates in CPU features by many years.

Nobody knows how to optimize for the exact version of the CPU that a program is being run on than the CPU itself.

Cell ?

[polyp2000]

lets face it when Cell arrives formally theres going to be little point in ploughing resources into something thats effectively headed for obsolesence

Getting leaner, IBM?

[osewa77]

First they drop a PC line tha was not making them money. Then they drop a server line that's clearly not the future of that space. I think they're making some right decisions here. If the POWER platform succeeds, as it more likely would when resources are focussed on it, and it is accepted as a viable alternative to the PC platform, the ensuing competition would probably be good for all of us.

Re:Getting leaner, IBM?

[Anonymous Coward]

No, the reason that Power(PC) never made a dent in x86 is that IBM promised everyone that it would scale better and it simply did not. Furthermore, IBM themselves quashed cheap PowerPC workstations due to internal politics surrounding OS/2, never provided good chipsets to third parties, etc.

Hey "Blame Microsoft For Everything" is fun, but IBM never seriously attempted to position PowerPC in the mainstream x86 market.

Re:Itaniums are the worst chips ever

[QuickFord]

True, but they make a real nice keychain!

AMD64

[bstadil]

work on EM64T, will you intel?

baloney, call it by its proper name AMD64

Re:AMD64

[Anarke_Incarnate]

AMD64 is the real deal. EM64T is a kludge that is mostly compatible with AMD64. AMD64 has better performance and better handling of >4GB RAM.

Re:AMD64

[mczak]

In what way exactly?

I suspect the parent poster is refering to the IOMMU of the Athlon64/Opteron chips, which intels EMT64 chips lack. This might have consequences for some PCI (or other I/O) devices, the OS might need to use copy buffers if those devices want to do dma transfers to/from address space above 4GB.

Re:AMD64

[Hal_Porter]

Yeah, I read about that, it's a cool hack, in the most positive sense of using something in a way that it was not designed for. Intel failed to spot this extra use for the AGP GART when they cloned the chip I suppose.

It's an interesting question though. AMD got to design the chip and the architecture at the same time. Intel had to retrofit AMD's 64 bit stuff to the P4. There are all kinds of reasons why this would be hard - the P4 had a dual speed ALU which needed to be widened to 64 bit for instance.

I wouldn't be surprised at all if the resulting chip had some performance issues, but I haven't seen any benchmarks as to comparative 64 bit performance though. I don't particular like P4's even for 32 bit stuff - it looks like they ultrapipelined the CPU to get higher clock frequencies in a way that reduces performance compared to a similar priced AMD part for instance, whereas AMD seems to have worried about real benchmarks like Spec and ( Doom3 :-) )

Did retrofitting AMD^H^H^HEMT64 make it worse?

IBM's High end

[MagnusDredd]

Article has some strange ideas about what constitutes a High-end server. I'd imagine a IBM P595 which supports up to 64 Processors would be high end... IBM seems to think so too. But then again what do they know about high end. I mean, they are only #2 in the High end server market (over $1,000,000 per server), and #1 in the mid-range server market (between $100,00 and $,1,00,000 per server).

Not accurate

[Anonymous Coward]

TFA and moreso the summary and headline is not fully accurate according to what I've heard.

What IBM has decided not to do is support the Montecito IA64 chips. Apparently Intel initially approached IBM about licensing the X3 technology for an chipset to support Montecito, IBM agreed and shut down their own program to develop a chipset and redeployed the resources, Intel came back a few weeks later and said they had changed their mind, would IBM build an X3 chipset for Montecito but by this point they had also announced that the next post-Montecito Itanium chip would be plug compatible with Xeon. Hence the market opportunity for Montecito is about 18 months so it's not worth IBM's effort to build a chipset for only that time.

IBM has therefore decided to continue to sell the existing x455 servers through this year, skip Montecito and support Itanium again with X3 when it becomes plug compatible with Xeon. That means that for about a year they will have no server that will support Itanium.

Two years is a long time in this business so who knows if anyone outside of the HP/UX install base will care about Itanium by then but IBM does have a plan for continued IA64 support if current trends continue.

This is not good news for Itanium but it's also not a complete cancellation.

Power vs Itanium vs Xeon vs Opteron

[SpamMonkey]

Ok, so we all know the various CPU names and who makes them etc but do we actually know how they compare? Me and the team I work for have total ownership of 7 SAP Application servers and 1 database server, total ram in the DB server is 48GB and the App servers have been 4 and 12GB's each (normal compared to batch processing). They're all running on either IBM P630 to P670's. What does that mean? I have NO idea except that they are able to comfortable deal with 1200 active users at any given time.

now, if someone can tell me that Itanium will give us better performance for more we'll look into it, if it's Xeon then it's Xeon (pah but you get the idea). What I fail to see is why it's important what hardware is being used as long as it does the job it needs to do!

Thanks.

They could have been popular

[extra the woos]

If Intel released them at a low price and with desktop motherboards that were affordable. If an average geek could build the latest itanium system for $200 more than the latest athlon system, well, people would buy it because it's something different, it performs well, and because they want to mess with the architecture. It shuld have been marketed like the P-Pro. Too much for the desktop user, but if you want one you can afford it and you can build it/buy it!

"IBM to Drop Itanium" is proof!!!

[Kymermosst]

Proof that the best way to accelerate an Itanium is at 9.8 m/s^2.

(That's 32 ft/s^2 in ye olde units.)

Re:IBM using Power based CELL CPUs

[pmonje]

OK, I'm not supposed to talk about this, but I have two friends who work for the illuminati and all the new mind control chips are going to be based on the CELL. Unfortunately they couldn't give me to many details because of the NDAs that their reptilian overlords made them sign. Anyone wanna bet me?

seriously slashdot needs a "-1 talking out your ass rating"