Intel Confirms: Macs to switch to ARM by 2020.

[DanielMDA]

On 3/2/2019 at 9:18 PM, DrMacintosh said:

Remember people, that Apple is going to test out ARM based MacBooks with the fanless 12" MacBook. The MacBook Airs and Pros will continue to ship with x86 until Apple feels that their ARM implementation has more to offer than tradition x86. 

 

We might also see hybrid MacBook Pros with ARM and x86 capabilities on the same system. That technically already exists with the current TouchBar Macs and the T2 chip. 

Question, the T2 chip is 28nm right?

[DrMacintosh]

6 hours ago, Uttamattamakin said:

I wouldn't do this because I and other academics (I teach college and do theoretical physics research) have workflows that depend on a mix of really old software and really new (often custom coded that day in Matlab) software.

While that’s a valid concern, it’s nobodies fault but the ones who continue to use it. “If it ain’t broke don’t fix it” is only acceptable for so long. Relying on antiquated software shouldn’t be encouraged. 

 

6 hours ago, Uttamattamakin said:

On the plus side... maybe now we'll get a proper Macintosh computer in a convertible tablet form factor.

Alsost certainly never going to happen. The iPad Pro serves that purpose in the product stack. A Surface like Mac does not fit into the Mac lineup. It’s a formfactor of compromises for the neat gimmick of being able to remove the “required” keyboard that sacrifices lapability and usability. 

 

Also remember that Apple is against touchscreen Macs altogether, so that definitely means a tablet Mac is out of the question, yet alone a ARM based one. 

[DrMacintosh]

2 hours ago, DanielMDA said:

Question, the T2 chip is 28nm right?

I don’t know. Why would it matter? 

[DrMacintosh]

7 hours ago, descendency said:

so the best case scenario is losing a lot of performance in virtual workloads - I'd have to switch back to Windows for an upgrade

That’s actually your worst case scenario. The reality of it all is that the MacBook Pros will still be running on x86 based chips for the foresable future. Only the entry level MacBook will adopt ARM as a testbed. 

[DrMacintosh]

8 hours ago, Lukyp said:

I don't think people would be getting that much improved battery life for example for using arm instead of x86, and the effort and money required to transitioning would be excessive considering how cheap and power efficient some chips are (and more will be in the future),

While x86 chips can be extremely power efficient, ARM can and does far exceed it in an optimized worload with significantly less power consumption. 

 

Running Geekbench on a x86 based PC and on an iOS device shows that to be true. They run the same instructions across both platforms and mobile ARM chips have an architectural advantage from having cut out all the debilitating excess from x86. 

 

 

[Guest]

45 minutes ago, DrMacintosh said:

While x86 chips can be extremely power efficient, ARM can and does far exceed it in an optimized worload with significantly less power consumption. 

 

Running Geekbench on a x86 based PC and on an iOS device shows that to be true. They run the same instructions across both platforms and mobile ARM chips have an architectural advantage from having cut out all the debilitating excess from x86. 

 

 

Don't get me wrong I would prefer a CISC architecture over everything, no doubt for already existing stable cross-architecture available software it's the case, as I mentioned before I would definitely getting one since the toolchain for compiling on mobile are already available and I could work the same I do on x86 but faster and less efficient, but I'm more concerned about the support from the market on other things

Example: Why would adobe port their suite to ARM? I mean, ok there already are some for iOS and Android, but that's only because it's a different market and phones are going to be more widespread than computers at this point, and I don't think adobe is going to spend all those money on this kind of development in the sake of efficiency, no one did that, that could be possible and already happened for open source software where the cost would be hypothetically "zero" if you get what I mean
Unless a lot of people are going buy it for some reason, but if you tell them they do not have the software what would they do? It's the same issue as unsupported operating systems, they could be better faster and power efficient but why are going people to use it if there is no software available?
Imho they could start supporting iOS apps in some way already to have some things but that would need some sort of desktop integration, but that would not be a properly a "desktop" experience

That's why I hope Apple find a way either to support x86 binaries (but It's unlikely to happen as Mira Yurizaki said Intel was ready to sue microsoft for that) or to push companies some way, because they are the only one since the other alternatives have no big company behind them

[LAwLz]

22 hours ago, BuckGup said:

1) My dad works at Hewlett Packard Enterprise as an Engineer, my friends Dad works at Global Foundries

Are you really using "my dad works at X" as an argument? What are you, 12? Sorry but I don't believe you and even if I did, your dad saying something, or your friend's dad doesn't really mean anything either. I want facts, not hearsay

 

22 hours ago, BuckGup said:

2) It's not that niche as the government buys a ton, so do colleges, and research companies. ARM is making it so they can be smaller, more energy efficient, and cheaper while being faster at exactly what they want to do

It is a niche though. Just because you can mention 3 customers who might buy ARM servers (as well as x86) does not mean ARM is taking over or is selling more than x86.

 

And there is very little evidence that ARM servers actually are smaller, more efficient, cheaper and faster than x86 right now. I don't doubt that they are better at some of those areas, but I would like to see evidence that they are better in all areas like you claim.

I looked at the pricing and performance of Amazon's A1 instances and they are not exactly that impressive. They cost about 40% less per core, but the performance is also not even close to comparable. For example in PHPBench a Xeon Platinum M5 instances performs 3 times as well per core, which means the ARM instances will be far more expensive if you want the same performance. You pay 60% of the cost, but only get 33% of the performance.

Here are some benchmarks if you want to look at them: Phoronix

[bcredeur97]

still wanna know what arm can do when you give it say, 95 watts to play with. Still have yet to see that.

[Mira Yurizaki]

5 hours ago, DrMacintosh said:

While x86 chips can be extremely power efficient, ARM can and does far exceed it in an optimized worload with significantly less power consumption. 

 

Running Geekbench on a x86 based PC and on an iOS device shows that to be true. They run the same instructions across both platforms and mobile ARM chips have an architectural advantage from having cut out all the debilitating excess from x86.

That has little to do with the ISA and mostly to do with the implementation of it. The best a Snapdragon 845 I've seen on there is 2405 @ 1.76GHz. The iPhone XS got 4797 @ 2.49GHz. Normalizing the Snapdragon to 2.49 GHz would cause the score to theoretically bump up to about 3400. Same ISA, but vastly different scores.

[Guest]

6 minutes ago, Mira Yurizaki said:

That has little to do with the ISA and mostly to do with the implementation of it. The best a Snapdragon 845 I've seen on there is 2405 @ 1.76GHz. The iPhone XS got 4797 @ 2.49GHz. Normalizing the Snapdragon to 2.49 GHz would cause the score to theoretically bump up to about 3400. Same ISA, but vastly different scores.

So far Apple has takes the lead to mobile devices, I gotta admit

[Blademaster91]

5 minutes ago, Mira Yurizaki said:

That has little to do with the ISA and mostly to do with the implementation of it. The best a Snapdragon 845 I've seen on there is 2405 @ 1.76GHz. The iPhone XS got 4797 @ 2.49GHz. Normalizing the Snapdragon to 2.49 GHz would cause the score to theoretically bump up to about 3400. Same ISA, but vastly different scores.

Wouldn't the OS also affect scores?  Apple has iOS running efficiently on their CPU's so it doesn't seem accurate to compare a synthetic benchmark to an x86 CPU even though its using the same benchmark.

[BuckGup]

32 minutes ago, LAwLz said:

Are you really using "my dad works at X" as an argument? What are you, 12? Sorry but I don't believe you and even if I did, your dad saying something, or your friend's dad doesn't really mean anything either. I want facts, not hearsay

 

It is a niche though. Just because you can mention 3 customers who might buy ARM servers (as well as x86) does not mean ARM is taking over or is selling more than x86.

 

And there is very little evidence that ARM servers actually are smaller, more efficient, cheaper and faster than x86 right now. I don't doubt that they are better at some of those areas, but I would like to see evidence that they are better in all areas like you claim.

I looked at the pricing and performance of Amazon's A1 instances and they are not exactly that impressive. They cost about 40% less per core, but the performance is also not even close to comparable. For example in PHPBench a Xeon Platinum M5 instances performs 3 times as well per core, which means the ARM instances will be far more expensive if you want the same performance. You pay 60% of the cost, but only get 33% of the performance.

Here are some benchmarks if you want to look at them: Phoronix

lol I mean he actually does. Do you want his LinkedIn or something? What’s happening now is smaller companies will manufacture an ARM CPU specific to what the customer wants. So there really isn’t any benchmarks outside of the internal ones so you’d have to take my word until a couple years when it’s more mainstream and a customer allows benchmarks

[Bananasplit_00]

On 3/3/2019 at 3:45 PM, yian88 said:

I said very clearly that when i said stability and resource usage i did not meant kernel, did i not? the kernel is stable and very good for all sorts of purposes.

Read where i said that the entire software stack built by random people xorg, wayland, DE's, compositors etc are not stable or low resource usage and they will never be. And when i mean stable i dont mean you boot to desktop and it works, i mean switching multi windows, high refresh/high resolution multi-monitors, games, graphical apps and so on they behave like shit, lock up freeze or need restart quite often on all distros on all hardware.

 

Yes linux kernel is very good, and can be very stable as desktop OS but it needs a platform developed as one not random components that dont work togheter well, we need something like android where all the API, UI, security etc built on top of the kernel is unified and works flawless.

Games work pretty damn well now with Lutris, DXVK and Proton but the rest of that is very true. Having three monitors is a glitchy mess sadly... I still vastly prefer my Linux install over Windows 10 as it is just quicker, eats less resources and doesn't send back fuck loads of telemetry. Nvidia seems to be to blame for most of my issues though. 

[LAwLz]

6 minutes ago, BuckGup said:

lol I mean he actually does. Do you want his LinkedIn or something? What’s happening now is smaller companies will manufacture an ARM CPU specific to what the customer wants. So there really isn’t any benchmarks outside of the internal ones so you’d have to take my word until a couple years when it’s more mainstream and a customer allows benchmarks

I don't really care what your dad works with or what he says. Again, it's at best hearsay. 

 

What do you mean by "smaller companies will manufacture an ARM CPU specific to what the customer wants"? ARM processors are generic CPUs. They are not like ASICS. 

 

And I don't have to take your word for anything. 1) because I don't trust your word. I have no reason to. 

2) we already have benchmarks such as the AWS A1 instances I linked to earlier, and the numbers are not that good. 

[ScratchCat]

16 hours ago, Lurick said:

Because then it will run away >.>

It was a runaway success.

 

15 hours ago, Mira Yurizaki said:

Which doing this on say the normal MacBook first would make sense. Which if Apple can build up a base there means the rest may come.

Further the base MacBook is a lower volume product and aims at users who mostly use web browsers and built in apps. This means for most of the users a swap to ARM would not be detrimental and should the concept fail it wouldn't devastate overall Mac sales.

[Stefan Payne]

21 hours ago, Mira Yurizaki said:

The ISA has little to no bearing in whether or not a processor performs better.

Well, when they were developed, back in the 70s and 80s, it was different but things were different.

21 hours ago, Mira Yurizaki said:

how the second-gen Core processors beat the FX series.

I call bullshit on that in every way,

Especially 4 Core Core 2 were just crap and didn't scale very well because of the FSB.

 

21 hours ago, Mira Yurizaki said:

The only thing that would make this transition hairy is anyone who doesn't have an iOS compatible version of their app. I don't think Apple could make an x86 emulator like they did making a PPC emulator during the Intel transition considering Intel tried to sue Microsoft for doing it. I mean, Apple has a lot of money but my armchair business sense would rather put that money to helping others transition only rather than fight a court battle and make an emulator on top of the transition

You can probably use some kind of on the fly compiler...

[Ashley MLP Fangirl]

11 minutes ago, ScratchCat said:

Rosetta didn't seem to provide brilliant performance and despite improvements in translators like Windows on ARM there still is going to be a performance penalty. The concern is how much of the efficiency advantage ARM will loose when having to translate non-native applications.

no, you misunderstand. 

 

universal binaries were apps coded for both powerpc and intel. so they would perform good on both platforms. 

 

Rosetta was made for apps that were only coded for intel to work on powerpc. 

[Drak3]

1 hour ago, bcredeur97 said:

still wanna know what arm can do when you give it say, 95 watts to play with. Still have yet to see that.

Not much. AMD tried scaling up ARM, to find that ARM doesn't scale up well.

 

ARM can only achieve better power consumption numbers because it's a low performance design that revolves around running instruction sets that don't require much power, and ignoring performance enhancing features like multithreading.

 

Try to scale up the design and and power draw increases far faster than power draw, because x86 CPUs are designed for high performance and ARM CPUs are designed for low power consumption,

[Guest]

1 hour ago, LAwLz said:

What do you mean by "smaller companies will manufacture an ARM CPU specific to what the customer wants"

Maybe is talking about custom ARM implementations? Or other cheaper RISC ones, which could be since he was talking about niche supercomputers

Edited March 4, 2019 by Guest

[Mira Yurizaki]

1 hour ago, Blademaster91 said:

Wouldn't the OS also affect scores?  Apple has iOS running efficiently on their CPU's so it doesn't seem accurate to compare a synthetic benchmark to an x86 CPU even though its using the same benchmark.

It may, but if an OS is doing its job properly, then it's taking up a trivial amount of CPU time when someone else wants to use the CPU.

[Drak3]

1 minute ago, Mira Yurizaki said:

It may, but if an OS is doing its job properly, then it's taking up a trivial amount of CPU time when someone else wants to use the CPU.

Without being able to verify performance by running iOS on Snapdragon/Exynos and Android on A series chips, it's still something we can't be certain of.

 

And I'm not convinced that Google or the FOSS community is really doing a good job with Android.

[Stefan Payne]

19 minutes ago, Drak3 said:

Not much. AMD tried scaling up ARM, to find that ARM doesn't scale up well.

No, AMD never really tried ARM and stopped it.

Rumor has it that that Opteron Version might have been a custom Chip for Amazon - wich didn't go over too well...

19 minutes ago, Drak3 said:

ARM can only achieve better power consumption numbers because it's a low performance design that revolves around running instruction sets that don't require much power, and ignoring performance enhancing features like multithreading.

No, has nothing to do with ARM vs. something else. Its about low Power Designs vs. Performance ones.

If you want to compare ARM, you need to compare it to Kabini and other lower power designs.

 

19 minutes ago, Drak3 said:

Try to scale up the design and and power draw increases far faster than power draw, because x86 CPUs are designed for high performance and ARM CPUs are designed for low power consumption,

Its not that easy, you need a completely new architecture, optimized for completely different things.

You essentially have to start with a blank piece of paper...

It is said that the AMD 29000 CPU was the groundwork for their K5 and that they "slightly modified" it to x86...

[Mira Yurizaki]

41 minutes ago, Drak3 said:

Without being able to verify performance by running iOS on Snapdragon/Exynos and Android on A series chips, it's still something we can't be certain of.

 

And I'm not convinced that Google or the FOSS community is really doing a good job with Android.

While I've been keeping that in mind, the is another fundamental explanation for why Apple's SoC performs better than Qualcomm's. Apple's SoCs are wider in execution cores and can do the following:

• 6 integer operations (two of which can do load/store and two can do branch prediction)
• 3 FPU operations

The Cortex-A75 that the Snapdragon 845 uses can do the following:

• 2 integer execution operations
• 2 load/store operations
• 1 branch operation
• 2 FPU execution operations
• 1 FPU store operation (I'm not sure what this does)

It's very unlikely in benchmarking scenario like Geekbench or SPEC2006 that they'll be using memory based operations unless the test is specifically doing that or a mixed use test. The outcome would have to be weighted on what does the math on the actual data. Considering that at best, the A12 can do 6 integer operations means it will totally wreck the Cortex-A75 on a pure integer performance test. Even if the test is hitting execution, load/store, and branching, the A12 still has an advantage overall due to having one more branch operation capability. Otherwise, the A12's multiplexed integer slots can be configured as needed for the best performance.

 

And pinging an older post:

52 minutes ago, Drak3 said:

ARM can only achieve better power consumption numbers because it's a low performance design that revolves around running instruction sets that don't require much power, and ignoring performance enhancing features like multithreading.

The instruction set doesn't factor into power consumption at all. The only thing that would affect power consumption is the ease of decoding it, but you can optimize CISC ISAs for power consumption because chances are compilers only use a fraction of the ISA. So the front-end of modern x86 CPUs likely optimize around working with those instructions over working with the lesser used ones.

 

It's important to remember ARM started off as a desktop computer ISA.

[DanielMDA]

9 hours ago, DrMacintosh said:

I don’t know. Why would it matter? 

Just for fun... Most controllers and chipsets still use the 28nm planar node. I am asking if Apple´s T2 is different, since Apple likes bleeding edge stuff.

[Mira Yurizaki]

18 minutes ago, DanielMDA said:

Just for fun... Most controllers and chipsets still use the 28nm planar node. I am asking if Apple´s T2 is different, since Apple likes bleeding edge stuff.

I'm pretty sure they're only bleeding edge when it suits them. The expense of manufacturing tends to go up as process nodes go down. Not to mention this is an operation-critical chip. The fewer surprises there are, the better.