Is ARM the end of Intel's monopoly?

[Idontcare]

So you are saying ARM will suffer the same destiny as AMD?

Yes. But not just AMD. Look to everyone else who attempted to compete with Intel in the x86 space.

Intel will either do the same to everyone in the ARM space by virtue of continuing to push x86, or they will adopt ARM as a fall-back position themselves and then proceed to beat them at their own game.

Why do you think Intel alone accounts for nearly 1/3 of the entire semiconductor's annual R&D expenditure, and has a capex budget that dwarfs most company's annual revenue? They aren't playing around here.

 

[Fjodor2001]

ARM cores are much smaller than Intels Core and ATOM. Even ATOM at 22nm will be much bigger than Cortex 15. Even with a full node shrink ahead, Intel will not be in the same position they are now against AMD.

One more think, I see nobody talking about GPGPU, something that Intel is even more behind and will play a bigger role in the future.

http://forums.anandtech.com/showthread.php?t=2294334

Why is the Atom core so bloated?

About 10x larger than a Cortex-A7 (even after having compensated for 32 nm vs 28 nm). Is there an inherent difference in the ARM vs x86 instruction set resulting in x86 cores becoming larger?

Also, I notice that the Atom core is about 50% larger than the Jaguar core (after having compensated for 32 nm vs 28 nm), despite the Jaguar core being much faster.

 

[piesquared]

In the future, I don't think intel will have near the process advantage it has now. It's certainly not a foregone conclusion, look at ST-Ericsson's 28nm FD-SOI for example.

 

[Puppies04]

About 10x larger than a Cortex-A7 (even after having compensated for 32 nm vs 28 nm). Is there an inherent difference in the ARM vs x86 instruction set resulting in x86 cores becoming larger?

Yes, ask an x86 to emulate an arm core (all emulation suffers a performance hit) and it will happily abide. Ask an arm core to emulate an x86 core and you better put a pot of coffee on.

 

[AtenRa]

That specific portion of my post which you quoted was intentionally referring to the outcome of Intel's fall-back position which would be Intel producing their own ARM cores to compete with everyone elses ARM cores, at that point the primary difference would be process technology.

Unless TSMC manages to best Intel in the process node race, Intel ARM would have no excuse for not being cheaper to produce and lower power in operation compared to the ARM chips coming out of TSMC (or GloFo, or Samsung). But Intel would of course attempt to make x86 successful first and foremost for the same business strategy reasons Ellison speaks to.

Intel has two options (ARM or x86) and a serious lead in process tech. Everyone else they are competing with (except AMD) only has ARM to put forward and no process tech advantage. It is up to Intel to decide how they want to skin this cat, but obviously they are going to attempt to do it in a way that maximizes margins and profits first. If that fails, plan B is still practically guaranteed to beat out everyone anyways (because of the process tech advantage).

Hmmm, i have misunderstood you. Yes if they are going to produce ARM SoCs they are having the upper hand due to process, but they are still way off in GPGPU tech. I believe the GPGPU will be more important in the near future than CPU cores.

 

[Lonbjerg]

In the future, I don't think intel will have near the process advantage it has now. It's certainly not a foregone conclusion, look at ST-Ericsson's 28nm FD-SOI for example.

Keep dreaming...Intel just upped the money they spend on R&D...AFAIR they spend more on R&D than AMD's entire budget...fuzzy-warm-feelings won't change that.

 

[Lonbjerg]

Hmmm, i have misunderstood you. Yes if they are going to produce ARM SoCs they are having the upper hand due to process, but they are still way off in GPGPU tech. I believe the GPGPU will be more important in the near future than CPU cores.

Because AMD said so?

 

[GillyBillyDilly]

Is there an inherent difference in the ARM vs x86 instruction set resulting in x86 cores becoming larger?

Lots of legacy instructions. I don't know how much of the die space they make, but they are there.

 

[AtenRa]

Because AMD said so?

Because NV and AMD are pushing this way and i see more and more applications using GPU acceleration every day.

 

[GillyBillyDilly]

Keep dreaming...Intel just upped the money they spend on R&D...AFAIR they spend more on R&D than AMD's entire budget...fuzzy-warm-feelings won't change that.

He was talking about the future. We don't know if in the future Intel will be able to afford this level of R&D. There are actually quite realistic reasons which speak for Intel not being so cash rich in the future.

 

[2is]

Personally, I don't think so.

As a consumer of several mobile (ARM powered devices) I'm already starting to see stagnation in the mobile segment. Newer products are offering less and less breakthrough features and I'm often finding myself upgrading just because I'm bored and not so much because the next best thing is that much better, but I'm sure I'll be doing less and less of that. That said, the mobile segment is still strong because it has not yet been over-saturated like the PC market is, but with Samsung/Google flooding the market with products (both low and high cost) and Apple also shortening their product release timeline, saturation will set in pretty quick

 

[Red Hawk]

You have to separate that question per market. For example Intel may be dominant in the desktop space, but they are no big player in the mobile phone space which is ARM's domain.

And there will not be a one single "super duper chip" that will rule all markets.

Whether Intel will successfully compete in ARM territory and vice versa is hard to tell.

Exactly. Here's how I see it:

Performance dekstops/all-in-ones: ARM doesn't have a chance. In this environment, Intel is virtually unshackled from heat, size, and power concerns of mobile platforms. They have the best performance currently and the experience and resources to keep growing that performance. ARM has never even attempted a desktop CPU, and for good reason. Intel will likely continue to virtually monopolize the desktop market, with AMD's x86 processors taking up a small portion in the budget sector.

Laptops/Ultrabooks: Possibly. This market has the lest heat/power/size concerns of the mobile platforms. Intel is already firmly entrenched though and only getting better, so if ARM wants to make headway they really need to improve performance.

Tablets: The main battleground. Large enough that Intel can stretch its performance legs but small enough that power is enough of a concern to go with ARM. Both sides need to bring their A game in order to dominate this market.

Phones: This is sort of a similar situation to desktops, only the other way around. It's not quite so bad for Intel, as they do have working, semi-competitive CPUs in this market already. As it stands, Intel could work its way up, but it will be a very tough fight and Intel may just fail altogether (at least in bringing x86 to this market; they could still just adopt ARM, as Idontcare said).

 

[Lonbjerg]

Because NV and AMD are pushing this way and i see more and more applications using GPU acceleration every day.

You know why they are pushing it?
It's really simple.
It's their only chance to compete...Intel has locked down all other sectors for them.

And just because they NEED it...dosn't mean it will happen.

 

[piesquared]

Keep dreaming...Intel just upped the money they spend on R&D...AFAIR they spend more on R&D than AMD's entire budget...fuzzy-warm-feelings won't change that.

Neither will sticking your head in the sand. I thought you were aware that AMD spun off it's fabs quite a few years ago so i'm not sure how AMD's R&D budget has anything to do with process tech. Time to pull that head out and come up for air. lol You might consider TSMC, Samsung, IBM, GLOBALFOUNDRIES

 

[2is]

Neither will sticking your head in the sand. I thought you were aware that AMD spun off it's fabs quite a few years ago so i'm not sure how AMD's R&D budget has anything to do with process tech. Time to pull that head out and come up for air. lol You might consider TSMC, Samsung, IBM, GLOBALFOUNDRIES

You should recognize the difference between processor design and fabrication. AMD is still responsible for one of those (I'll let you try and figure out which one) and it requires R&D.

 

[Exophase]

Yes, ask an x86 to emulate an arm core (all emulation suffers a performance hit) and it will happily abide. Ask an arm core to emulate an x86 core and you better put a pot of coffee on.

Not if we're talking about two cores that have similar performance levels like Atom and Cortex-A9.

x86 doesn't have emulation magic built into its ISA. Chips that are locked out of x86-64 like Medfield are at a distinct disadvantage in emulating chips that have more registers. On the flip-side, it means 32-bit x86 will be an available (if not the only) Android NDK target for some time to come and that eases emulation going the other way.

Sure, there's x86 stuff that takes multiple instructions to emulate on ARM. load-op and RMWs, 8 and 16-bit operations (although you will typically not have to keep them coherent with their larger registers after every instruction, if ever at all), large immediates, push/pop, call/ret, SSE stuff that doesn't fit, etc. But the same goes for x86 - three address arithmetic, folded shifts, address adjust, block memory, predication, NEON stuff that doesn't fit, etc. Someone would have to do very optimized translators to get a good feel for which one can do better; it's probably not an easy question to answer before that.

I do know that ARM emulation on x86 suffers because all ALU instructions squash flags, meaning you will often need relatively expensive flags save/restore or other strategies to get around it. It's to the extent that the researchers writing the translation software wanted a flags nullifying instruction prefix added to Atom, but no such instruction was added. ARM doesn't have the same kind of problem because it can already nullify flags, and x86's more eager flags destruction means flags are more likely to be dead and less work to emulate.

I also know that if you tried to do this right now, where the ARM emulation is running on the best available Atoms like Clover Trail, and the x86 emulation on top ARMs (Swift, Krait, A15) the Atom is going to have a big disadvantage by being in-order. Statically re-scheduling code in translation is hard to do, especially when you have too few registers.

 

[Hulk]

In the future, I don't think intel will have near the process advantage it has now. It's certainly not a foregone conclusion, look at ST-Ericsson's 28nm FD-SOI for example.

The process advantage Intel has did not appear over night. They have been working at it for years and years. It is not something easily overcome. Besides AMD, which has been ground nearly into dust, remember Cyrix, NeXT and all of the others with grand plans to compete with Intel? All gone. Destroyed. And each vanquished competitor made Intel even stronger. Designing a new processor architecture and actually having the fabs to make it work is a very, very complicated process. I have been observing this industry for over 20 years and the only company that is always responding to market demands, whether it be AMD, Cyrix, or PowerPC, and coming out on top is Intel.

Now this is a little different I will admit because they have some truly powerful competitors this time round, with giants like Samsung and Apple breathing down their necks, but Intel is still Intel.

 

[Ferzerp]

The process advantage Intel has did not appear over night. They have been working at it for years and years. It is not something easily overcome. Besides AMD, which has been ground nearly into dust, remember Cyrix, NeXT and all of the others with grand plans to compete with Intel? All gone. Destroyed. And each vanquished competitor made Intel even stronger. Designing a new processor architecture and actually having the fabs to make it work is a very, very complicated process. I have been observing this industry for over 20 years and the only company that is always responding to market demands, whether it be AMD, Cyrix, or PowerPC, and coming out on top is Intel.

Now this is a little different I will admit because they have some truly powerful competitors this time round, with giants like Samsung and Apple breathing down their necks, but Intel is still Intel.

Your mistake was making a serious reply to someone who only posts with one goal (make AMD look good through any means. That includes talking up AMD, or talking down anyone competing with AMD). Reality never enters in to the content of that guy's posts.

 

[Roland00Address]

ARM cores are much smaller than Intels Core and ATOM. Even ATOM at 22nm will be much bigger than Cortex 15. Even with a full node shrink ahead, Intel will not be in the same position they are now against AMD.

You seem to be missing the bigger picture. It doesn't really matter the sizes of the arm cpu core vs atom cpu core what matters is the size of the entire atom soc vs the entire arm soc.

The current generation atom soc is 5.60 mm^2 at 32nm assumming perfecting scaling that is 2.80mm^2 if it was designed on 22nm.

But it isn't just the cpu that gets smaller when you due a die shrink it is also all the other components of the soc. The l2 cache gets smaller, the graphics get smaller, everything gets smaller.

------------------

Thus assuming equal transistor density (which intel and tmsc do not have) intel 22nm will allow intel to have larger cpu cores then arm at 28nm and 32nm while keeping the soc roughly the same die size. All the while benefiting from 3d transistors/finet and lower power consumption that you are also able to achieve by being at a better process node.

 

[tviceman]

That specific portion of my post which you quoted was intentionally referring to the outcome of Intel's fall-back position which would be Intel producing their own ARM cores to compete with everyone elses ARM cores, at that point the primary difference would be process technology.

Unless TSMC manages to best Intel in the process node race, Intel ARM would have no excuse for not being cheaper to produce and lower power in operation compared to the ARM chips coming out of TSMC (or GloFo, or Samsung). But Intel would of course attempt to make x86 successful first and foremost for the same business strategy reasons Ellison speaks to.

Intel has two options (ARM or x86) and a serious lead in process tech. Everyone else they are competing with (except AMD) only has ARM to put forward and no process tech advantage. It is up to Intel to decide how they want to skin this cat, but obviously they are going to attempt to do it in a way that maximizes margins and profits first. If that fails, plan B is still practically guaranteed to beat out everyone anyways (because of the process tech advantage).

Intel will in most likelihood make the fastest CPU (I'm not talking GPU though) regardless of whether they are making x86 or ARM, so long as they can leverage their most advanced fabs for making mobile chips. The immediate issues though is that they aren't utilizing their most advanced fabs for their mobile products AND they are currently absolutely sticking to their x86 guns. In the short-term, ARM is going to continue to grow insofar as the number of devices powered by ARM-based CPU's. Assuming Intel does not turn around and license ARM within the next two years, even with some products on the market, ARM-based mobile devices (cell phones, tablets, and hell lets include laptops) will likely continue to increasingly sell in larger volume than x86 devices.

Once a solidified ecosystem is in place, it's very hard to get rid of it. It can happen over time but not over night. And on top of the ecosystem, TSMC and Samsung are going to hopefully put their higher profits to good use to accelerate their node advancements, thus closing the gap with Intel. In the meantime, Intel still has the significant long term advantage but I think sticking to x86 is the hard road to penetrating the mobile consumer space.

 

[Idontcare]

You seem to be missing the bigger picture. It doesn't really matter the sizes of the arm cpu core vs atom cpu core what matters is the size of the entire atom soc vs the entire arm soc.

The current generation atom soc is 5.60 mm^2 at 32nm assumming perfecting scaling that is 2.80mm^2 if it was designed on 22nm.

But it isn't just the cpu that gets smaller when you due a die shrink it is also all the other components of the soc. The l2 cache gets smaller, the graphics get smaller, everything gets smaller.

------------------

Thus assuming equal transistor density (which intel and tmsc do not have) intel 22nm will allow intel to have larger cpu cores then arm at 28nm and 32nm while keeping the soc roughly the same die size. All the while benefiting from 3d transistors/finet and lower power consumption that you are also able to achieve by being at a better process node.

I agree that Intel's process tech advantage looks formidable on paper. But it isn't really an advantage unless Intel prioritizes the production schedule of its ARM contenders (ATOM based SoC's) on a timeline that puts that process node advantage to work in a timely fashion.

If Intel wanted to dominate the likes of Qualcomm, Samsung and Apple with their 22nm process tech then the very first production wafer to come out of a 22nm fab should have been ValleyView (Silvermont)...instead Intel opted to use their now dated 32nm process to compete with existing 28nm products.

The process tech advantage, real or hypothetical, is irrelevant if it is not brought to bear in a timely fashion.

(yes I know Intel claims to be bringing in their atom-based SoC time schedule, moving to 14nm sooner instead of later, but until I see it happen I will consider this to be your usual garden variety marketing FUD campaign designed to undermine confidence in handset makers to shake their commitment to the likes of Nvidia(tegra) and Qualcomm(snapdragon+))

 

[Genx87]

Intel is planning to release chips into Arms world soon on the latest processes and using full on x86 chips. Will Arm survive that incursion into its market of such a higher performance design in comparison to its own combined with a massive process advantage?

I am more worried about ARM and its partners than Intel.

Intel has been doing this for years. The results speak for themselves so far.

 

[Ferzerp]

Intel has been doing this for years. The results speak for themselves so far.

Not sure they've ever taken it seriously until the last 18-24 months though.

 

[Roland00Address]

I agree that Intel's process tech advantage looks formidable on paper. But it isn't really an advantage unless Intel prioritizes the production schedule of its ARM contenders (ATOM based SoC's) on a timeline that puts that process node advantage to work in a timely fashion.

If Intel wanted to dominate the likes of Qualcomm, Samsung and Apple with their 22nm process tech then the very first production wafer to come out of a 22nm fab should have been ValleyView (Silvermont)...instead Intel opted to use their now dated 32nm process to compete with existing 28nm products.

The process tech advantage, real or hypothetical, is irrelevant if it is not brought to bear in a timely fashion.

(yes I know Intel claims to be bringing in their atom-based SoC time schedule, moving to 14nm sooner instead of later, but until I see it happen I will consider this to be your usual garden variety marketing FUD campaign designed to undermine confidence in handset makers to shake their commitment to the likes of Nvidia(tegra) and Qualcomm(snapdragon+))

I agree intel has purposefully not gone with using their new nm factories for atoms. The reason for that was due to intel utilizing a lower power process for atoms instead of their higher power process for serves, desktop, and laptop chips. In other words even though both factories were operating at 45 or 32 nm neither factory could produce the other type of chip without significant retooling. (This is not uncommon in the industry for example with 28nm tmsc has 4 different variants of tooling due to power consumption.)

Intel purposefully tooled all their 32nm for servers and notebooks for that was where the profit was, it took 2 years to get 32nm atoms. Same logic with 22nm, 22nm atoms are going to be out in late 2012 so probably nov or december while 22nm core chips were out late april so a difference of 19 months or so.

Intel is purposefully changing this with airmont which should be comming out in 2014. Airmont will be comming out the same time where intel switches from 22nm chips to 14nm chips

That said when silvermont comes out at the end of this year, intel will be pushing 22nm chips and to my understanding all the arms chips will be using 28nm, 32nm, or 40 nm (for the really cheap crap kind.) I do not know of anyone shipping a working product that is going to be on 20 or 22nm by the end of this year.

Intel will have the process lead even with silvermont (22nm) they will have even a larger process lead with airmont (14nm)

 

[IntelUser2000]

Even if you are really fast, you can't design a CPU from the ground up in less than 5 years. That is a fact. That means Silvermont would have been conceived back in 2008, and maybe even further back. That was when they weren't so rushed.

Also, everybody forgets one important thing with Smartphone designs. They take 6+ months to get to market, because makers take extra time optimizing it to their needs. Even if Atom and Core comes at same time, Atom based products will arrive 6+ months later. That's part of the reason Ultrabooks based on Haswell are coming few months later. The battery life improvements it will bring is only possible if they use similar tactics that they use in Smartphones.

Medfield was available since early 2011, but the earliest ever product would have been the now-cancelled Nokia phone in end of 2011. The cancellation meant it was delayed till early-mid 2012. Now, do you think Intel was sitting around reading Anandtech forums during that time?