ARM vs x86 in an iPhone world

[Voltron]

Now that the iPhone is opening up development and with Android on the horizon, there are likely to be a LOT of developers programming for ARM devices. Many, many, many more during the past decade of ARM powered phones.

How will this affect the relevance of x86 in the computing world? There are those, such as Charlie from The Inq, that seem to want to kiss Intel's x86 butt all over the place (though it may be just misplaced NVIDIA aggression).

If ARM runs the iPhone why not desktop Leopard? Or why not in the next Apple TV? Or is the next Apple TV an iPod? Perhaps for some people, just as some people use an iPod as a primary stereo. Or maybe there is an ARM chip in the rumored Apple tablet (if it exists).

 

[Arun]

I think the executive summary is this: anyone who thinks ARM vs x86 matters outside of their respective home turf is crazy. Intel's reasoning for why anyone would want to use their chips in handhelds is that it matters. Thus, they are crazy. And mostly hopeless, unless they somehow manage to beat all expectations and come out with an outstanding product that beats everything else on the market.

Anyway, just like there's no magical reason why x86 makes sense in handhelds, there's no magical reason why ARM makes sense in desktops or laptops, even in Apple's case where the OS is not a true concern. ARM is very possibly a nicer ISA, but in the end it doesn't really matter because nobody is going to make a sufficiently high performance core out of it. Even for a Tablet, it doesn't really seem to make much sense in practice.

As for 'mid-way' products such as the Apple TV, who knows. Right now it's a 3-chips architecture (Intel CPU - Intel chipset - NVIDIA GPU). I think we'll see that migrate to a 2-chips architecture first with an IGP, and then eventually to a single-chip SoC with a x86 CPU. All of the above is IMO, of course.

 

[Voltron]

I pretty much agree.

But with Fusion and and Intel's plan to integrate CPUs, how much more powerful are x86 CPUs as we know them going to be? Are 8 and 16 cores going to make it in consumer devices? What are the incremental benefits of additional cache?

By the same token, are ARM cores so small is there room for them to improve performance at a faster rate than the x86 processors with process migrations?

So an ARM tablet right now looks iffy, but I bet you could get some pretty decent performance, especially with a good GPU, and amazing battery life in the not so distant future.

 

[nutball]

ARM-powered desktop computers have been tried. Erm... about fifteen years ago. Twenty in fact I believe was when the first one came out. I've owned three, they were great. Fifteen years ago. I now own a PC and much as I loved the ARM ISA when programming at a low-level, there is just no sense in which that justifies an ARM-based PC. It's not really necessary to program at that level these days, and if you program at a high level, who cares what's underneath? The one company who had the opportunity to make ARM truly great on the desktop was Intel, and they had zero motive to do so. So they didn't.

As far as I can see ARM are where they are today precisely because they got beaten out by 486/586/P6 on the desktop. They decided to go another route, and that route was a one-way street as far as I can see. Conversely Intel ignored the niche that ARM moved into (otherwise ARM arguably wouldn't exist today), and I'm guessing they have a long way to go to catch up with ARMs tech lead in that area.

 

[Arun]

But with Fusion and and Intel's plan to integrate CPUs, how much more powerful are x86 CPUs as we know them going to be? Are 8 and 16 cores going to make it in consumer devices? What are the incremental benefits of additional cache?

I'm sure 8 and 16 cores will make it into consumer devices. The real question is whether it will ever enter the mainstream. And my tentative answer to that is 'No'. Or at least not before the price premium is basically zero.

After quad-core, what you're likely going to see for the mainstream (i.e. anyone not using PCs for games or as a 3D workstaiton) is complete and utter commoditization, IMO. There is just no workload that could benefit from more outside the gaming market, and the next-gen Windows won't tax hardware as heavily as Vista did. Accelerators are also mostly useless. The mainstream should be able to understand this trend in a couple of years.

I think the fundamental problem you've got is there still *are* users who need more performance, but they aren't the majority. So there are good reasons to keep up with Moore's Law in one market, but it's hurting you in the other. Your only chance then is to reduce the price of the low-end PC and hope you can expand the market that way, compensating the other loss. This is certainly the direction most major players are taking.

However, it is also worth not being too aggressive in terms of timeframe for that transition. We are still talking many years out here, let alone because a high-IPC x86 quad-core won't really become very cheap until 22nm. However, I could easily imagine single-chip integration of lower-IPC solutions as early as 45nm...

By the same token, are ARM cores so small is there room for them to improve performance at a faster rate than the x86 processors with process migrations?

Yes, they're really really small. The Cortex-A9 cores you'll see used in chips around 2009 (and final products in 2010, presumably) are probably slower on an IPC basis than Pentium 3s, and I doubt you could clock them much higher than the latter on 65nm even if you optimized the layout around that! So you're literally talking about 10 years old tech here...

So an ARM tablet right now looks iffy, but I bet you could get some pretty decent performance, especially with a good GPU, and amazing battery life in the not so distant future.

Why not just use Intel's Silverthorne or something from VIA and be done with it? You don't need hundreds of hours of battery life in a Tablet... ARMs are just too slow and 'too' cheap for that market, IMO.

The problem is if you want an ARM core to be viable for that market, you'd need to develop it from scratch for that. It's just not worth the effort IMO.

 

[MfA]

Even for a CMP processor with a greater emphasis on parallel performance I don't think the ARM architecture would make much of a difference ... it's not a huge improvement over x86. Something like Cell's SPU instruction set (ignoring local memory constraints) is better than either IMO (mainly for how it handles branches). Without Thumb-2 I'd rank x86 higher than ARM in fact for the simple reason it makes better use of instruction caches, caches and multipliers are more of a factor than core logic in the area those cores take.

 

[roninja]

ARM or x86 the GPU will be SGX.

 

[NickDjinn]

Are 8 and 16 cores going to make it in consumer devices? What are the incremental benefits of additional cache?

By the same token, are ARM cores so small is there room for them to improve performance at a faster rate than the x86 processors with process migrations?

So an ARM tablet right now looks iffy, but I bet you could get some pretty decent performance, especially with a good GPU, and amazing battery life in the not so distant future.

Just 3 years later we are seeing hex and octo cores. Arm CPUs are making their way into Tablets and doing a pretty good job. We are even about to see a dual core high performance Arm laptop....and devices like the entourage edge. They perform better with less power and space even than an intel Atom. Arm is taking off, and smart phones probably had a lot to do with their success.

I'm sure 8 and 16 cores will make it into consumer devices. The real question is whether it will ever enter the mainstream. And my tentative answer to that is 'No'.
The problem is if you want an ARM core to be viable for that market, you'd need to develop it from scratch for that. It's just not worth the effort IMO.

Octocores already hit the market and its barely been a few years. They are being used in servers as well as in gaming computers.

This is an outdated article, but besides this commentary India is talking about producing a subsidized linux alternative to the Ipad costing only $100 and having favorable specs in comparison....Arm is all around superior for mobile computing, especially in cost but it can compete performance with with intel at a fraction of its power consumption.

This should be good for linux, unless windows finds a way to make itself compatible with arm processors. Linux supports arm, likely the near future of mobile computing, and windows doesnt.