Qualcomm considering getting into the Fab business

[Arachnotronic]

Step away from the RDF. Exxon has lower margins, but has about 160% of the profits that Apple has (with 4 times the revenue).

The top 4 for profit in the US are Exxon, Chevron, Apple, then MS. Exxon has over 150% of the profit of the second place slot, and Chevron, Apple, and MS are pretty close to one another. (amusingly, MS has higher margins than Apple, but less revenue)

I stand corrected. Thank you.

 

[ShintaiDK]

Apple is an epic IT bubble. Not exactly the first we had of those. People quickly forget after a burst.

It is also along time since the dotcom bubble bursted in 2000.

 

[pelov]

To maybe return to the original topic a bit: how much in trouble are both IBM and the ARM collective in the 10 years or so going forward? I would say, by the look of things, a lot of trouble.

Aren't both companies on the up and up? ARM is doing really well in the mobile space and IBM is doing great in enterprise. Why would you be worried?

 

[Arachnotronic]

Aren't both companies on the up and up? ARM is doing really well in the mobile space and IBM is doing great in enterprise. Why would you be worried?

ARM isn't doing "great". The winners of the ARM ecosystem are guys like Qualcomm, NVIDIA, and Texas Instruments. Oh, and TSMC.

 

[ShintaiDK]

That makes sense. I wonder if IBM's existing x86 license coupled with AMD's ISA extension-sharing agreements with Intel could somehow render IBM with an x86-64 license in the event of an IBM buyout of AMD, but I'm really speculating wildly at this point. I also think there is a clause in the Intel-AMD agreement saying that it is null and void if either party is purchased by another company.

To maybe return to the original topic a bit: how much in trouble are both IBM and the ARM collective in the 10 years or so going forward? I would say, by the look of things, a lot of trouble.

Why would IBM want to lose money and throw billions into a black hole? Without Intels fabs you basicly lost the x86 competition nomatter design. And when Intel is so far ahead and keeps the superb performance as they do know. Competing in any form would just be silly.

Plus even if IBM started today, it would take atleast 4 years before the first x86 proper design rolled off.

 

[ShintaiDK]

I can't really say that they do need AMD (whatever direction they take), but I figure that if demand for POWER systems really becomes so small that IBM can't justify continuing development on them (which, admittedly, is unlikely, if for no other reason that IBM in any case would have to continue supporting existing customers for at least 10 years), it would be better for IBM to make high-end x86-compatible processors and systems to compete with high-end Xeon than to cede the hardware side completely to Intel. The ISA doesn't really make that much of a difference in the enterprise space, but at least the added scale that is attainable in the x86 server market would help support development costs better than the RISC market does for POWER development today.


LOL, I strongly agree. HP is headed for complete irrelevance. Every stupid decision and every insane acquisition they make just puts them one step closer to complete disaster.

IBM would just sell Intel OEM systems like they do now. And what they make a fortune on servicing.

 

[pelov]

ARM isn't doing "great". The winners of the ARM ecosystem are guys like Qualcomm, NVIDIA, and Texas Instruments. Oh, and TSMC.

That's to do with the slim margins that ARM themselves set. They make pennies on a several hundred dollar device from licenses and they're okay with that. They have slim overhead and don't have to bother with things like fabs. ARM isn't going anywhere. In fact, ARM adoption is increasing on both mobile and server (outpaced AMD, actually).

 

[lol123]

Aren't both companies on the up and up? ARM is doing really well in the mobile space and IBM is doing great in enterprise. Why would you be worried?

The basic problem is that despite that ARM has a lot of wind in its back, TSMC (which supplies Apple, Qualcomm, Nvidia and TI (?)) can't even provide enough wafers for its customers. At the same time, Intel has released an at least somewhat competitive smartphone SoC at 32nm and will move to a new architecture on 22nm tri-gate (remember that this process would allow cuts in power consumption of up to 50% at low clock frequencies) in 2013, and then to 14nm tri-gate in 2014. With Intel's track record, there is no reason to believe that they won't be able to deliver wafers in volume, in time. TSMC on the other hand can't keep with up with demand even on 28nm bulk. So we might have a situation in a couple of years where ARM simply can't get enough silicon out to market to consolidate its predicted gains or even defend its current market position. You might recall that is similar to what happened to AMD at the height of their success.

IBM have also had problems with new manufacturing processes (remember the gate-first debacle) and whether they will be able to keep up with Intel going forward is anyone's guess. It's extremely unlikely, but we might even see a reversal in the performance race between POWER and Itanium when Kittson supposedly arrives on 22nm tri-gate in 2014, depending on whether POWER8 disappoints or is a no-show on 22nm in 2013. But even if that doesn't happen, Xeon is rapidly gaining performance and RAS/scaling capabilities and will be additionally supported by Intel's growing manufacturing advantage in the future.

 

[pelov]

The basic problem is that despite that ARM has a lot of wind in its back, TSMC (which supplies Apple, Qualcomm, Nvidia and TI (?)) can't even provide enough wafers for its customers. At the same time, Intel has released an at least somewhat competitive smartphone SoC at 32nm and will move to a new architecture on 22nm tri-gate (remember that this process would allow cuts in power consumption of up to 50% at low clock frequencies) in 2013, and then to 14nm tri-gate in 2014. With Intel's track record, there is no reason to believe that they won't be able to deliver wafers in volume, in time. TSMC on the other hand can't keep with up with demand even on 28nm bulk. So we might have a situation in a couple of years where ARM simply can't get enough silicon out to market to consolidate its predicted gains or even defend its current market position. You might recall that is similar to what happened to AMD at the height of their success.

But doesn't this favor ARM? They sell their chips incredibly cheap and that's important for devices that sell in incredibly high volume. You're also not restricted to a single fab and a single chip maker. The x86 tax, despite what Intel claims, is still alive and well. Most consumers are still picking up 45nm ARM chips and they don't have any issues with it. Performance on a phone isn't as important as it is on a laptop or desktop where that fab advantage might play a larger role. You're not doing any extensive calculations and using processor intensive productivity apps on a smartphone and tablet. It's a media consumption device and as such, a cheap chip with ARM cores, whether 28nm or 45nm, doesn't matter all that much.

IBM have also had problems with new manufacturing processes (remember the gate-first debacle) and whether they will be able to keep up with Intel going forward is anyone's guess. It's extremely unlikely, but we might even see a reversal in the performance race between POWER and Itanium when Kittson supposedly arrives on 22nm tri-gate in 2014, depending on whether POWER8 disappoints or is a no-show on 22nm in 2013. But even if that doesn't happen, Xeon is rapidly gaining performance and RAS/scaling capabilities and will be additionally supported by Intel's growing manufacturing advantage in the future.

Intel wants nothing to do with Itanium. Ever. That's HP paying Intel money to keep going. Itanium is an HP baby not an Intel product. If it were up to Intel they'd have aborted that years ago but HP kept paying them money and begging them to keep at it. It still hasn't developed into anything but more lawsuits between Oracle, Intel and HP. Because unlike Itanium, you can actually still buy Power, it's doing really well and the software support doesn't hinge on how the courts decide what will happen between HP and Oracle.

The Xeon/x86 chips are a different story, though

 

[lol123]

But doesn't this favor ARM? They sell their chips incredibly cheap and that's important for devices that sell in incredibly high volume. You're also not restricted to a single fab and a single chip maker. The x86 tax, despite what Intel claims, is still alive and well. Most consumers are still picking up 45nm ARM chips and they don't have any issues with it. Performance on a phone isn't as important as it is on a laptop or desktop where that fab advantage might play a larger role. You're not doing any extensive calculations and using processor intensive productivity apps on a smartphone and tablet. It's a media consumption device and as such, a cheap chip with ARM cores, whether 28nm or 45nm, doesn't matter all that much.

It doesn't favor ARM, but I wouldn't say that it specifically disfavors them either. The problem isn't ARM's business model, but rather that going forward, no one but Intel (and to some extent Samsung and IBM) will have access to modern fabs with sufficient capacity. And the chip sizes don't matter when you can't get enough wafers from the foundry to make your business work. As we could see in the article, Qualcomm have even been forced to adjust their earnings projections because they are unable to keep up with demand. As long as ARM-based chips are to be sold very cheap, which they will have to be to compete with Intel, the ROI for TSMC from constructing new fabs on smaller (and increasingly expensive) manufacturing processes will also be limited.

Intel wants nothing to do with Itanium. Ever. That's HP paying Intel money to keep going. Itanium is an HP baby not an Intel product. If it were up to Intel they'd have aborted that years ago but HP kept paying them money and begging them to keep at it. It still hasn't developed into anything but more lawsuits between Oracle, Intel and HP. Because unlike Itanium, you can actually still buy Power, it's doing really well and the software support doesn't hinge on how the courts decide what will happen between HP and Oracle.

The Xeon/x86 chips are a different story, though

Yeah, you're right that the Itanium is dead, really no matter how Poulson and Kittson perform. It would be fun to see it go out in a blaze of glory though.

 

[ShintaiDK]

Itanium is anything but dead. Huawei and Inspur also picked up Itanium last year. Inspur is the biggest server vendor in China. And Huawei we all know.

 

[pelov]

Are you sure?

http://arstechnica.com/information-...th-court-make-oracle-support-itanium-forever/

But Oracle’s move to drop Itanium development can hardly be seen as revenge—the move was recognition of cold, hard numbers. Oracle is just one in a series of software providers who have cut and run from the Itanium platform based on its anemic market performance. By going after Oracle for its alleged breach of agreement, HP has been forced to expose much of the hand-wringing at both HP and Intel over the future of Itanium. All that has led to ever-increasing speculation in the Itanium dead pool.

That trajectory was certainly evident to software vendors supporting Itanium. Red Hat bailed on Itanium support in Enterprise Linux in late 2009. Microsoft Windows Server Product Manager Dan Reger announced in April of 2010 that Microsoft would phase out support for Itanium and stop developing products for it after Windows Server 2008 R2. Reger cited the ever-improving performance of Intel and AMD 64-bit x86 systems thanks to multi-core architecture.

In March 2011, Oracle announced it was ending all development in support of Itanium. Oracle executives explained the decision as justifiable euthanasia—they contended that Intel officials had made it clear that Itanium development was approaching "end of life." Oracle also pointed to the fact that since-canned HP CEO Leo Apotheker had failed to say one word about Itanium in his discussions of HP’s strategic roadmap.

It's dead.

 

[ShintaiDK]

Yes.

Also you run Unix on Itanium. Not Windows. Windows and mission critical aint going well hand in hand.

Itanium outsells both Sparc and Opterons for that matter. Only POWER and Xeons sells more of any server CPU.

 

[alyarb]

http://www.microsoft.com/en-us/server-cloud/windows-server/2008-r2-itanium.aspx

Windows has existed for Itanium since the beginning. Not as popular of course.

 

[ShintaiDK]

In 2009 Q3, the IDC numbers showed only 6% used Windows.

MS didnt ditch Itanium support due to hardware. But because nobody wanted MS and Windows when they didnt have to.

Windows 8 is also heading to be the biggest Windows flop ever.

 

[ShintaiDK]

Only way for Itanium to die is if Intel outcompete itself. Aka x86 Xeons can offer the same features. But if that happens POWER is dead too.

 

[blckgrffn]

Are either of those companies going pay Intel to keep making Itanium? HP paid Intel hundreds of millions of dollars just to keep it moving.

If you can't run Oracle, Sybase, MS SQL (ha ha), etc on Itanium you are going to really narrow the customer base incredibly quickly.

How many RAS features does the EX line really lack compared to Itanium?

IBM themselves allows you to make a 4 node, fault tolerant system out of the x3950 x5 with Terabytes of RAM and all the goodies.

There are plenty of other good platforms for Unix. Why bother with Itanium?

*we have a row of Itanium, but AFAIK we are not going to be buying any new ones, and haven't bought any new ones for some time. That Ars article says that is a very common theme for HP, who sells ~90% of all Itanium CPUs built.

 

[pelov]

Are either of those companies going pay Intel to keep making Itanium? HP paid Intel hundreds of millions of dollars just to keep it moving.

If you can't run Oracle, Sybase, MS SQL (ha ha), etc on Itanium you are going to really narrow the customer base incredibly quickly.

You can't even scrape the bottom of the barrel since Microsoft dropped support as well. Throw in that Intel themselves have stated privately to HP that they don't want to keep it alive and what life is there left for Itanium? I don't understand HP. They're throwing around money they don't have to keep it on life support.

Itanium is like a really cool dog they've had for a long time that has gotten old and sick. HP loved that dog. It was HP's dog. It taught it tricks, spent great summers with it and watched it grow but sooner or later you've got to face reality and see that euthanasia might be the most compassionate and considerate option. The medical bills are piling up and he isn't getting any younger

 

[ShintaiDK]

How many RAS features does the EX line really lack compared to Itanium?

http://www.intel.com/content/dam/do...n-critical-converged-infrastructure-paper.pdf

 

[blckgrffn]

http://www.intel.com/content/dam/do...n-critical-converged-infrastructure-paper.pdf

Half an answer.

Looking at the competitive analysis, it leaves me thinking that IBM would like have a sheet with as many checkboxes not left filled in on the Itanium side. Something from Gartner would be more compelling.

Even then, it looks like x86 is not that far behind. The next iteration of EX should close the gap even further

 

[Cerb]

Apple is an epic IT bubble. Not exactly the first we had of those. People quickly forget after a burst.

It is also along time since the dotcom bubble bursted in 2000.

With Jobs gone, I predict that they'll be able to smoothly ride the ideas he had in the pipe for st least another 3 years. After 5 years, though, I expect it to look like MS w/ Ballmer.

Are either of those companies going pay Intel to keep making Itanium? HP paid Intel hundreds of millions of dollars just to keep it moving.

If you can't run Oracle, Sybase, MS SQL (ha ha), etc on Itanium you are going to really narrow the customer base incredibly quickly.

How many RAS features does the EX line really lack compared to Itanium?

That depends on how much legacy software you might need to support, and/or if you're stuck in an old mainframe ideology.

For fault tolerance of a small number of high-bandwidth servers (if the server count is going to be large anyway, this doesn't apply), when latency is also an important factor, you really need poisoning and mirroring support. Now, currently the application has to be able to make some use of it, but several Oracle and IBM products can do that.

Beyond that, if you're building new, it's basically marketing. Computers are cheap. High speed networking isn't cheap, but not it's too expensive, and software can do most of the logical redundancy work. Back when computers were expensive, hot swapping CPUs, RAM, and cards made sense. Today, that means you really need to think about a redesign, so that you can shut a whole server down if you need to, without system downtime.

Basically, there are two extremes of fault tolerance: (a) built it so it won't be likely to fail unexpectedly, or (b) assume it's going to fail, and prepare for likely failure modes. Old big computer systems were close to A, while communication systems have historically been closer to B. Fault detection on buses and in the core can't be handled externally, so at a very low level, there is no replacement for a quality chip (hardening, ECC everywhere, logic fault prevention where possible, etc.). Once beyond that basic level, though, it's all a matter of choosing what's more necessary, and what's more efficient. Today, computers are cheap, so redundancy by way of more servers, more PSUs, more drives, bigger badder SANs, etc., are generally the better way to go, except in a few cases where local bandwidth is simply the only way to get a lot of work done (IE, when more cores and memory bandwidth > more servers).

 

[Idontcare]

With Jobs gone, I predict that they'll be able to smoothly ride the ideas he had in the pipe for st least another 3 years. After 5 years, though, I expect it to look like MS w/ Ballmer.

Never underestimate the ability of someone to take something golden and totally trash it.

Hector Ruiz is a more recent example, but a more relevant one would be Apple and what happened when Jobs was replaced by John Sculley the first time around.

Unfortunately a "MS w/Ballmer" is the rosey picture, they should be so lucky. Time will tell. Craig Barret was no Andy Grove, but Paul Otellini picked up the ball and did great things at Intel.

Maybe Tim Cook is Apple's Otellini?

 

[nonameo]

Never underestimate the ability of someone to take something golden and totally trash it.

Hector Ruiz is a more recent example, but a more relevant one would be Apple and what happened when Jobs was replaced by John Sculley the first time around.

Unfortunately a "MS w/Ballmer" is the rosey picture, they should be so lucky. Time will tell. Craig Barret was no Andy Grove, but Paul Otellini picked up the ball and did great things at Intel.

Maybe Tim Cook is Apple's Otellini?

My understanding is that cook is pushing for better working conditions in the china factories. That's a + in my book.

 

[Cerb]

Never underestimate the ability of someone to take something golden and totally trash it.

Hector Ruiz is a more recent example, but a more relevant one would be Apple and what happened when Jobs was replaced by John Sculley the first time around.

Maybe, maybe. But, Cook has been there and handled Apple operations for a long time, and is credited with a great deal of Apple's non-creative success. It would be weird for him to drag the company down. But, Jobs' talents were fairly unique, and not likely to be followed well by a manager who started out by getting an engineering degree.

 

[ShintaiDK]

With Jobs gone, I predict that they'll be able to smoothly ride the ideas he had in the pipe for st least another 3 years. After 5 years, though, I expect it to look like MS w/ Ballmer.

I think it will go much faster. Iphones essentially only sell in USA and UK today. So add 1-2 years and its starting to look grim there.