Oracle to stop all software development on the Itanium processor

[IntelUser2000]

http://www.oracle.com/us/corporate/press/346696

After multiple conversations with Intel senior management Oracle has decided to discontinue all software development on the Intel Itanium microprocessor. Intel management made it clear that their strategic focus is on their x86 microprocessor and that Itanium was nearing the end of its life.

That does not bode well for Poulson.

 

[darkswordsman17]

That's pretty big news, although I'm surprised Itanium fared even as well as it did.

 

[RobertPters77]

What's the point of Itanium anyway? I've heard so much about it but I've never understood the hype and obsession. It's like the vaporware processor of the cpu world.

 

[pm]

It's not so surprising given Oracles acquisition of Sun.

What's the point of Itanium anyway? I've heard so much about it but I've never understood the hype and obsession. It's like the vaporware processor of the cpu world.

I thought for something to be vaporware, you needed to talk about it but not actually have it for sale. You can buy Itaniums... There's been 5 generations of them (codenamed Merced, McKinley, Madison, Montecito, Tukwila so far with a roadmap going forward of future designs) As far as hype, yes, originally it was hyped a little over the top, but since then it's had solid sales in the very high end server arena, and, at least from my perspective, not very much hype at all.

(and, yes, for the record, I design Itaniums for a living, so my perspective is a bit skewed)

 

[bunnyfubbles]

What's the point of Itanium anyway? I've heard so much about it but I've never understood the hype and obsession. It's like the vaporware processor of the cpu world.

enterprise servers and high performance computing applications

it is in no way intended for the consumer market

 

[ViRGE]

Wow, if this is true...

I expected to see Itanium outlive everything - it's not like Intel to give up like that.

 

[IntelUser2000]

There's been 5 generations of them (codenamed Merced, McKinley, Madison, Montecito, Tukwila so far with a roadmap going forward of future designs)

The troubling thing is not Oracle ceasing future development. It's why they decided to do so. Or so the press release claims.

 

[Idontcare]

The troubling thing is not Oracle ceasing future development. It's why they decided to do so. Or so the press release claims.

IntelUser, surely you can see the conflict-of-interest in this situation.

Oracle benefits by spreading FUD regarding Itanium's future.

Oracle benefits from any advanced info they can secure from Intel regarding future Itaniums that are under development.

Why would Intel give Oracle any advanced info on future Itaniums just so Oracle can siphon that info over to their UltraSparc design and sales teams?

Intel obviously broke off the flow of information, Oracle obviously stands to benefit by spreading as much FUD as possible.

Ellison is not exactly known as being a softy in the tech sector, he plays to keep and this collision course between Intel and Oracle was inevitable the day Oracle's executive team decided they were going to make a move to buy SUN.

So I view Oracle's statements to be expected and little more than FUD in the purest IBM-sense of the invention of the marketing strategy.

 

[podspi]

IntelUser, surely you can see the conflict-of-interest in this situation.

Now, if a similar announcement came from HP...

 

[Edrick]

Now, if a similar announcement came from HP...

Once HP says that, it is game over for Itanium. Would be a shame since Paulson looks to be a really nice CPU that should beat Power7 in a lot of areas.


I always wanted to get one of these system for my home server. But now that MS stopped support for it, that pretty much dashed my dreams. :'(

 

[Edrick]

Here is HP's response (from FUD)

http://www.fudzilla.com/processors/item/22200-hp-shoves-itaniums-into-blades

 

[podspi]

How big is the market for heavy-iron systems like the Power7 and Itanium? It seems to me like x86 is slowly taking over that market by just throwing tons of cores at the problem.


It's almost unbelievable to me that Microsoft is dropping Itanium support and adding ARM support. Sign of the times...

 

[Idontcare]

How big is the market for heavy-iron systems like the Power7 and Itanium? It seems to me like x86 is slowly taking over that market by just throwing tons of cores at the problem.


It's almost unbelievable to me that Microsoft is dropping Itanium support and adding ARM support. Sign of the times...

MS doesn't support Power7 or UltraSparc either, it is not so surprising that they threw in the towel on Itanium.

Itanium is for mission-critical stuff where uptime is paramount...that is not exactly MS's forte.

That said, if HP does come to effectively be the sole distributor of Itanium products then you can bet Intel will be looking to spin off the division and sell it to HP with some sort of long-term foundry agreement in place.

(same way SUN worked with TI to have TI fab SUN's processors)

 

[Edrick]

And here is Intel's official response:

http://www.dvhardware.net/article49162.html

This Oracle announcement really got everyones attention.

 

[Mr. Pedantic]

What is the point of Itanium? Intel already has a very highly performing chip design in x86, why don't they just focus on development of that?

 

[Idontcare]

Itanium was supposed to take us beyond x86, enable programmers and compilers to escape the baggage that is a legacy architecture which is now 40 yrs old.

I really don't care what CPU I will be using 10yrs from now, be it x86 or ARM or EPIC derived...but I am quite certain that the price tiers will continue to collapse for the same reasons they have in the HDD sector.

That said, I too am curious to see where Haswell/Rockwell based XEONs take Intel's x86 architecture. Rockwell will be 16nm, and will probably debut 2-3yrs before anyone else's 16nm process.

 

[Martimus]

It's not so surprising given Oracles acquisition of Sun.




I thought for something to be vaporware, you needed to talk about it but not actually have it for sale. You can buy Itaniums... There's been 5 generations of them (codenamed Merced, McKinley, Madison, Montecito, Tukwila so far with a roadmap going forward of future designs) As far as hype, yes, originally it was hyped a little over the top, but since then it's had solid sales in the very high end server arena, and, at least from my perspective, not very much hype at all.

(and, yes, for the record, I design Itaniums for a living, so my perspective is a bit skewed)

Still anouncements like this are stressful for people like you who are working on the project. While you are much closer to it, and feel like you have a better feel as to what is going on with planning, the truth is that oftentimes these outside sources have the correct end result, even if their reasoning is flawed.

I saw that happen in my last major program, where most of the issues that were reported were already corrected, and it seemed that the outside estimates of cancelation were just conjecture based on old information, but they were correct. Even though the reasoning was no longer accurate, the end result happened.

It is amazing what perception does to higher management. It is often far more important that reality, since the complexity of the actual situation isn't easy for higher management to truly grasp.

Anyway, I don't expect Itanium to go anywhere immediately, but it wouldn't be surprising to see Intel try to sell it off within the next 2-3 years. I am sure you are relatively secure yet in your job (or at least I hope so).

 

[pm]

I completely agree with Idontcare's assessment of the situation - although to be honest the roadmap aspect of it didn't occur to me but that makes sense too.


Quoting from the Intel Press release:

As a result of recent announcements from Oracle, Intel is taking this opportunity to directly reiterate its plans for the Itanium processor. “Intel’s work on Intel Itanium processors and platforms continues unabated with multiple generations of chips currently in development and on schedule,” said Paul Otellini, president and CEO of Intel Corporation “We remain firmly committed to delivering a competitive, multi-generational roadmap for HP-UX and other operating system customers that run the Itanium architecture.” Poulson is Intel’s next generation 32nm 8 core based Itanium chip, and is on track to more than double the performance of the existing Tukwila architecture. Kittson is an officially committed roadmap product for Itanium beyond Poulson and is also in active development. Intel Itanium processor industry momentum will be highlighted in a keynote at the upcoming Beijing Intel Developer’s Forum.

What is the point of Itanium? Intel already has a very highly performing chip design in x86, why don't they just focus on development of that?

It's not just about performance - with a lot of high-end servers extremely high reliability and scalability are more important than raw performance. The main point of Itanium nowadays is very high reliability, high scalability (lots of CPUs in one box) and high transactional throughput. In terms of reliability in particular, Itanium has a set of data-hardening and error recovery features that are not available in the Xeon line - features such as register file parity, data poisoning containment, core-level lockstep and others. So if you want to build a very large, high transactional throughput, extremely high reliability server - such as the HP Nonstop line of servers - then Itanium can make more sense than Xeon.

 

[Idontcare]

It's not just about performance - with a lot of high-end servers extremely high reliability and scalability are more important than raw performance. The main point of Itanium nowadays is very high reliability, high scalability (lots of CPUs in one box) and high transactional throughput. In terms of reliability in particular, Itanium has a set of data-hardening and error recovery features that are not available in the Xeon line - features such as register file parity, data poisoning containment, core-level lockstep and others. So if you want to build a very large, high transactional throughput, extremely high reliability server - such as the HP Nonstop line of servers - then Itanium can make more sense than Xeon.

It's also a beast of a workstation with some HPC apps.

I use Gaussian for ab initio calcs and I was quite jealous of the guys who sported Itanium workstations because their runtimes were generally 4-5x faster than mine for comparable jobs...me being $-limited, I opted for the much cheaper x86-based workstation at the time.

That was a few years ago though, not sure if the Itanium performance delta for this app has been sustained.

 

[Idontcare]

EETimes has an article up too, I definitely find the following excerpt disconcerting if true:

''HP CEO Leo Apotheker made no mention of Itanium in his long and detailed presentation on the future strategic direction of HP.''

If Itanium is not a high enough internal priority so as to make it into the CEO's strategy rollout then I'd say dark sky's are swirling.

I remember when TI decided to ax their CMOS R&D at the 45nm, one business quarter the CEO was telling EETimes that the future of TI was critically dependent on our process tech advantage, the next business quarter he is telling EETimes that going fabless for CMOS was the only viable path.

Hmmm...

 

[Mr. Pedantic]

It's not just about performance - with a lot of high-end servers extremely high reliability and scalability are more important than raw performance. The main point of Itanium nowadays is very high reliability, high scalability (lots of CPUs in one box) and high transactional throughput. In terms of reliability in particular, Itanium has a set of data-hardening and error recovery features that are not available in the Xeon line - features such as register file parity, data poisoning containment, core-level lockstep and others. So if you want to build a very large, high transactional throughput, extremely high reliability server - such as the HP Nonstop line of servers - then Itanium can make more sense than Xeon.

Why can they not merge the feature sets?

On a related note, I notice that IBM's POWER CPUs are huge, with large caches and core counts causing CPUs of 500mm^2 or more. Itanium has prodigious amounts of cache, causing similarly large CPUs. So why are most x86 CPUs so small? I'm sure Xeons and Opterons in particular could benefit from dumping a bit of Itanium's cache and producing a processor with double the core count in a 500mm^2 die. I thought AMD already did that in part with Magny Cours, and I'm surprised Intel isn't following, since their CPUs would probably do better with 32nm and a more efficient architecture anyway.

 

[Voo]

It's also a beast of a workstation with some HPC apps.
[...]
That was a few years ago though, not sure if the Itanium performance delta for this app has been sustained.

Well that's the question. Even (especially?) looking at Intel x86 is getting, better and better from a performance point of view, as well as RAS and looking at the price we don't even have to talk about it.
And then look at the timeline and nodes for Itanium. Tukwila was released when and was built on what node? Beginning 2010 and 65nm? Commitment to a product is something else (Yes I'm aware that you really don't want a large and expensive chip at the cutting edge of a node, but still - compare that to their Xenons)
The problem is the niche market where you don't want x86 CPUs is quite small and there's lots of competition there. And as time goes on their extremely cheap x86 CPUs become "good enough" both in terms of features and speed.
So the question is, why invest billions of R&D into Itanium, a project that so far has cost them billions of money and does not show much growth for the future. Also consider that their contenders in that area don't really make much money with their CPUs but are more interested in the all around contracts, software, etc. (hi IBM)

Or asked from another perspective, if you were a Intel manager and had to decide between investing a billion into Itanium or a low energy ARM contender, what would you choose?


@Mr. Pedantic: Mostly because HPC stuff is especially memory dependent (and much larger data sets!) compared to the usual consumer stuff. Also ever looked at what they can ask for those CPUs?

 

[pm]

Why can they not merge the feature sets?

To some extent that's been happening over time. Reliability features have waterfalled from Itanium into Xeon over time. But the cores used on Xeons come from desktop/mobile and reliability features use precious space and precious power (both of which affect the performance of the CPU - in a design you want everything close together - for example adding 8-bit parity to a 64-bit register file would grow it by 13% which slows down circuit performance and adds >13% more power). So you add these features for results that most consumers don't care about. So there's a limit to how much you can add reliability features into a consumer product before it doesn't fit in the desktop/mobile space any more.

On a related note, I notice that IBM's POWER CPUs are huge, with large caches and core counts causing CPUs of 500mm^2 or more. Itanium has prodigious amounts of cache, causing similarly large CPUs. So why are most x86 CPUs so small? I'm sure Xeons and Opterons in particular could benefit from dumping a bit of Itanium's cache and producing a processor with double the core count in a 500mm^2 die. I thought AMD already did that in part with Magny Cours, and I'm surprised Intel isn't following, since their CPUs would probably do better with 32nm and a more efficient architecture anyway.

This has been happening to Xeon too. The Xeon 65xx and 75xx series CPU's (formerly "Beckton") have 24MB of cache which is definitely edging on "prodigious" by my definition of the word. Still determining cache size depends on lots of things and while more is always better, it adds cost. Determining "optimal" is difficult to determine and the answer that you will get depends a lot on the work load and scalability of the system. Intel has positioned Itanium at the top of it's server matrix and so it makes sense that it would have the most cache.

 

[taltamir]

What's the point of Itanium anyway? I've heard so much about it but I've never understood the hype and obsession. It's like the vaporware processor of the cpu world.

Itanium is a superior architecture to x86. x86 is an architecture from 1978 and is severely outdated. It has momentum but it is far behind.
However, intel is focusing on maintaining the x86 monopoly above all else. This has benefited them for a number of years but its time has come. Other architectures are set to dethrone it very soon, their attempt to extend x86 into the GPU market was an abysmal failure.
Itanium was meant to replace x86 in the server market only (not for home usage) so that intel may wring as much money from the market as possible, it debuted with ridiculously high prices too.
AMD undermined intel's market segmentation efforts with their AMD64 architecture; but mostly it was intel itself who undermined itanium by simply not being so poor about supporting it.

Going forward intel will do all it can to hold onto the dying and ancient x86 architecture. I don't predict it lasting another 10 years.

 

[Voo]

Itanium is a superior architecture to x86. x86 is an architecture from 1978 and is severely outdated.

Well if you really equal the modern x86 architecture with the 8086, then by all means Itanium has existed since the mid 80s when we had the multiflow VLIW machines. The same goes for ARM as well then.

I somehow doubt that we'll replace x86 anytime soon (or at all), we'll integrate GPUs and extend the ISA a bit more, but who would be mad enough to pour billions into R&D after how Itanium turned out?
The only contender atm is ARM and they're not especially interested in consumer markets and have quite the uphill battle in the server market against Intel.

The quite minor advantages of replacing x86 with something else, are hardly worth the involved cost and all the REAL problems we'll face in the future (cache coherency protocolls for manycores, memory latencies,..) won't be solved magically by any new architecture.