Apple Expected to Announce Arm-Based Mac Plans at WWDC, Transition Away From Intel to Begin in 2021

[richmlow]

MikeZTM,


I did not make the assumption that you assert that I made. I provided a benchmark graph as was given by Mathematica 12.

I merely stated that as it stands (today), I cannot run Mathematica (effectively) today.

As to running WolframMark on a Raspberry Pi 4, I cannot do this since I don’t have a RPi 4 (and I don’t have any intention of getting one).


richmlow

Please run that benchmark on a Pi4 with reasonable ram.
My 8GB Pi 4 isn't coming till next month but I will be happy to test it when it arrives. If you could do it now then we can have a result sooner.

You did made an assumption -- your assumption is a ARM11 Pi 1 with 256MB(or 512MB with model B) ram is a good representation of today's mobile CPU performance.

I guess a Xeon Platinum with 256MB ram will run as slow as that Pi in most situations.

[automerge]1591895111[/automerge]
@theorist,


Thank God that you have chimed into this thread!! What you have posted is exactly my point.

Are you a mathematician (like me)?

All the best,
richmlow

Richmlow is right to show healtyh skepticism. But I agree the picture he posted isn't meaningful, since it references an ARM without specifying its provenance. I did some Googling, and to the best I can tell, that processor was used in the early Raspberry Pi's, specifically the A, B, and B+.

Much more meaningful, as suggested by MikeZTM, would be to perform the Mathematica (MMA) benchmark on both a known Intel processor and the current Raspberry Pi 4+. According to

Mathematica 12 Available on the New Raspberry Pi 4—Wolfram Blog

Includes new features in Mathematica, including significant expansion of numerical, mathematics and geometric computation, audio and signal processing, text and language processing, machine learning, neural networks.

blog.wolfram.com

with MMA 12.0, the Raspberry Pi 4+ takes 77.8 seconds to complete the benchmark (they didn't give the score, just the individual times, which I summed). By comparison, it takes my Mid-2014 BTO MBP w/ 2.8 Ghz i7 (4980HQ Haswell/Crystalwell) (also running MMA 12.0) 5.50 seconds*, which is 77.8/5.50 = 14.1 x faster.

*This corresponds to a score of 2.52.

N.B: Richmlow's benchmark was run with MMA 12.1. On 12.1, my MBP scores a 2.91 (apparently MMA 12.1 is better optimized than 12.0), so it's somewhat faster than Richmlow's older MacPro, which scores 2.60. However, since Wolfram benchmarked their Raspberry Pi 4+ with MMA 12.0, I also used 12.0 in my comparison.

 

[Unregistered 4U]

Not disagreeing in general but I think you'll agree that designing the best in class chips (whatever class it is) is a challenge. There is a reason why we have only two major GPU vendors.

That’s more down to the barrier to entry, not that anyone else couldn’t do it. When you look at “GPU’s actually running in devices” as a metric, Apple’s the number 1 GPU vendor. However, they don’t have to try to sell their GPU’s to a wary public, folks just get them as part of the purchase.

 

[Sydde]

... the Raspberry Pi 4+ takes 77.8 seconds to complete the benchmark (they didn't give the score, just the individual times, which I summed). By comparison, it takes my Mid-2014 BTO MBP w/ 2.8 Ghz i7 (4980HQ Haswell/Crystalwell) (also running MMA 12.0) 5.50 seconds*, which is 77.8/5.50 = 14.1 x faster.

Raspberry Pi 4 uses a Cortex-A72 processor at 1.5Ghz (just over half the speed of your i7); the current iteration, 4 years later, is Cortex A78, which, at a minimum, is the architecture that Apple would be putting into ARM Macs. So, the comparison is probably a little flawed. ARM architecture appears to be advancing somewhat faster than Intel's.

 

[cmaier]

Raspberry Pi 4 uses a Cortex-A72 processor at 1.5Ghz (just over half the speed of your i7); the current iteration, 4 years later, is Cortex A78, which, at a minimum, is the architecture that Apple would be putting into ARM Macs. So, the comparison is probably a little flawed. ARM architecture appears to be advancing somewhat faster than Intel's.

Plus Apple doesn't even use ARM's architectures (they support ISA's corresponding to ARM's architectures, of course, but they decide the chip architecture on their own).

 

[MikeZTM]

MikeZTM,


I did not make the assumption that you assert that I made. I provided a benchmark graph as was given by Mathematica 12.

I merely stated that as it stands (today), I cannot run Mathematica (effectively) today.

As to running WolframMark on a Raspberry Pi 4, I cannot do this since I don’t have a RPi 4 (and I don’t have any intention of getting one).


richmlow




[automerge]1591895111[/automerge]
@theorist,


Thank God that you have chimed into this thread!! What you have posted is exactly my point.

Are you a mathematician (like me)?

All the best,
richmlow

His number greatly matches my assumption.
A Pi 4 improve the performance by more than 10 times using same bench as you did.

Remember Pi 4 is still using a 4 years old architecture. There's a lot of efforts done during the past 4 years.

 

[Sydde]

Plus Apple doesn't even use ARM's architectures (they support ISA's corresponding to ARM's architectures, of course, but they decide the chip architecture on their own).

Right. The ARM specification has some features that Apple does not need, like 32-bit+Thumb compatibility, multiple memory page sizing options and probably a few other things that do not fit their usage case, so they trim out the cruft to make the cores simpler and more efficient. They do collaborate with ARM on design patterns and take the most valuable aspects to their own units, and at the user-mode level the A-series are almost certainly ARM object code compatible, but someone else using one out of the box to build a different system platform might prove challenging.

 

[theluggage]

I merely stated that as it stands (today), I cannot run Mathematica (effectively) today.

...but that's rather irrelevant when you can't buy an ARM-based Mac today, probably won't be able to buy one for nearly another year and won't be forced to buy one for a couple of years yet.

Today, AFAIK, the only ARM-based machine that people are running Mathematica on is the Raspberry Pi: a $35 "maker" board aggressively designed down to a "disposable" price, built around surplus set-top-box chips, using an older ARM core design, a cheap SD card as its only mass storage, limited RAM and riddled with horrible i/o bottlenecks which are justified because it only costs 35 freakin' bucks! That's most likely where that benchmark came from (...it clearly wasn't Amazon's 64-core ARM cloud computing chip!)

Two things are certain about the future ARM Mac: (a) it will have a far more powerful processor than the Raspberry Pi and (b) it will cost a lot more than a Raspberry Pi!

Even the current iPhone/iPad processors benchmark at i7 speeds and leave the Pi in the dust - and although it's right to take that with a slight pinch of salt it's a good sign that - 2 generations down the line and not thermally constrained in a tablet - ARM Macs could have very decent performance.

All you can tell from the Mathematica benchmark is that Wolfram didn't have to break a sweat getting it to run on the Pi (and give it away for free) so there's no obvious reason why they shouldn't have a native ARM Mac version for paying customers in a reasonable timeframe. Also, they're gonna have to change their per-core licensing formula - even Intel chips now have twice the number of cores for the money that they did 2 years back, and they also need to allow for some ARMs actually having a mix of high-performance and low-power cores.

 

[richmlow]

Yes.



There are two main issues here.

1. Implementation of ARM in the computing platform. This is a hardware issue which Apple needs to deal with.

2. The running of Mathematica on Apple's ARM. This is a software optimization issue which Wolfram Research needs to deal with.

It's quite likely that both of these issues can be addressed by Apple and Wolfram Research. I'm an optimist. However, I'm trained to be a skeptic as I'm a mathematician. I deal in facts.

As I said in my original post, I'm very happy to give Apple my money if and when their rumored ARM laptop/desktop
computers (running Mathematica) give the same (or better) performance than my current Intel systems.



richmlow

...but that's rather irrelevant when you can't buy an ARM-based Mac today, probably won't be able to buy one for nearly another year and won't be forced to buy one for a couple of years yet.

Today, AFAIK, the only ARM-based machine that people are running Mathematica on is the Raspberry Pi: a $35 "maker" board aggressively designed down to a "disposable" price, built around surplus set-top-box chips, using an older ARM core design, a cheap SD card as its only mass storage, limited RAM and riddled with horrible i/o bottlenecks which are justified because it only costs 35 freakin' bucks! That's most likely where that benchmark came from (...it clearly wasn't Amazon's 64-core ARM cloud computing chip!)

Two things are certain about the future ARM Mac: (a) it will have a far more powerful processor than the Raspberry Pi and (b) it will cost a lot more than a Raspberry Pi!

Even the current iPhone/iPad processors benchmark at i7 speeds and leave the Pi in the dust - and although it's right to take that with a slight pinch of salt it's a good sign that - 2 generations down the line and not thermally constrained in a tablet - ARM Macs could have very decent performance.

All you can tell from the Mathematica benchmark is that Wolfram didn't have to break a sweat getting it to run on the Pi (and give it away for free) so there's no obvious reason why they shouldn't have a native ARM Mac version for paying customers in a reasonable timeframe. Also, they're gonna have to change their per-core licensing formula - even Intel chips now have twice the number of cores for the money that they did 2 years back, and they also need to allow for some ARMs actually having a mix of high-performance and low-power cores.

[automerge]1591906442[/automerge]

 

[Unregistered 4U]

Hell, there’s rumors going around that a new iMac with a new design, and new Intel processors and AMD graphics is supposed to come out later this summer.

There’s still a rumor from last year that folks inside Intel were saying that Apple’s going to transition this year. Now, I took that to mean that someone inside Intel saw the latest Apple/Intel contract and was like... “that’s waaaaay below what we normally get from them. Either we gave them an INSANE deal, or they’re actually purchasing way fewer processors.”

So, if the iMac is indeed the last Intel Mac for this year (we’ll know on the 22nd) and sales fall off because folks are waiting for the next big thing, that would point to the basis of that particular rumor.

 

[smulji]

I know this not directly related to ARM Macs but given how there's always comparison between Apple's supposed ARM SoC for desktops and Intel processors, this is big news

Exclusive: Internal Memo Shows Murthy's Remarks Over Jim Keller's Departure, Details About New Structuring, Raja Koduri To Head Architectural Roadmap

You might have seen Intel's declaration that Jim Keller has left the company, effective immediately, and we have more details for you. Jim Keller was one of the masterminds behind Intel's new architectural efforts and had been with the company for two years when he announced his resignation due...

wccftech.com

@cmaier @mdriftmeyer

 

[cmaier]

I know this not directly related to ARM Macs but given how there's always comparison between Apple's supposed ARM SoC for desktops and Intel processors, this is big news

Exclusive: Internal Memo Shows Murthy's Remarks Over Jim Keller's Departure, Details About New Structuring, Raja Koduri To Head Architectural Roadmap

You might have seen Intel's declaration that Jim Keller has left the company, effective immediately, and we have more details for you. Jim Keller was one of the masterminds behind Intel's new architectural efforts and had been with the company for two years when he announced his resignation due...

wccftech.com

@cmaier @mdriftmeyer

I know both Jim and Sundari. I worked with Jim at AMD (I think he had another stint there after I left - can't keep track - he moves around a LOT).

I have theories about why he keeps moving

Sundari is great. I met her after I retired from my CPU career, and she started at Sun right after I left. I'll never forget when we first met - she goes "wait - are you THAT maier? the one who worked on millennium at sun? Oh my gosh - the stories about you are legendary..." (I didn't get along with management very well in my brief time at Sun

Great to see that she's moving up in the world.

As for Jim, in 2 years I guess we'll have another article about him moving someplace yet again.

 

[smulji]

I know both Jim and Sundari. I worked with Jim at AMD (I think he had another stint there after I left - can't keep track - he moves around a LOT).

I have theories about why he keeps moving

Sundari is great. I met her after I retired from my CPU career, and she started at Sun right after I left. I'll never forget when we first met - she goes "wait - are you THAT maier? the one who worked on millennium at sun? Oh my gosh - the stories about you are legendary..." (I didn't get along with management very well in my brief time at Sun

Great to see that she's moving up in the world.

As for Jim, in 2 years I guess we'll have another article about him moving someplace yet again.

I can only imagine the what it costs for a company to recruit a person of his caliber only to have him up and leave on a regular basis. At this point, I think the only chip design company he hasn't worked at is Qualcomm.

 

[cmaier]

I can only imagine the what it costs for a company to recruit a person of his caliber only to have him up and leave on a regular basis. At this point, I think the only chip design company he hasn't worked at is Qualcomm.

Makes me wonder what you mean by “of his caliber”

To be fair, when I met him and worked with him, he was only involved with the LDT bus (which got named hyper transport for the public.). He may have achieved great things after that - I don’t have any actual knowledge. I’ve seen reports that say he did other things in those days, but not to my recollection - Fred Weber was the guy who was the visionary behind those chips.

 

[diddl14]

My next HackingTosh: https://www.gigabyte.com/ARM-Server/R281-T91-rev-100

 

[firewood]

His number greatly matches my assumption.
A Pi 4 improve the performance by more than 10 times using same bench as you did.

Remember Pi 4 is still using a 4 years old architecture. There's a lot of efforts done during the past 4 years.

Note that Raspian on a Pi 4 runs ARM apps in 32-bit mode. A 64-bit version of Mathematica would be significantly faster than a 32-bit build, and Apple’s newer A series chips are all 64-bit only. A contemporary ARM CPU would faster than the 3 or 4 year old one in the Pi 4, and Apple’s A series FPUs tend to be faster than ARMs. So even if some hypothetical higher wattage Apple chip does not outright exceed the FP performance of an i7 or i9, the multipliers from the above factors would put Apple in the ballpark for any competently ported scientific app.

 

[smulji]

Makes me wonder what you mean by “of his caliber”

By that I mean someone who's among the best in his field.

 

[cmaier]

By that I mean someone who's among the best in his field.

What evidence do we have for that?

 

[smulji]

What evidence do we have for that?

The major CPU architectures he was in charge of at AMD and Apple.

 

[theorist9]

@theorist,
Are you a mathematician (like me)?

All the best,
richmlow

Not a mathematician . My training is in physical chemistry, but my research is in theoretical biophysics (hence the user name....).

 

[cmaier]

The major CPU architectures he was in charge of at AMD and Apple.

That doesn’t mean much if he’s not working at a place long enough to get anything out the door.

 

[Wilderness-1902]

I don’t see how this has any bearing on CPU architecture. Developers for iOS/iPadOS/watchOS/tvOS/macOS use Macs because Xcode, not because x86.

If you were a college student of computer science, you choose a x86 or arm computer?

If it was me, I choose x86 since x86 is widely used for now.

Developers for Xcode don't care about arm or x86. It is true. But macos (arm) will lose the opportune to attract more developers from the x86 world, or even lose some developers during the transition. This is the point.

 

[arkmannj]

If you were a college student of computer science, you choose a x86 or arm computer?

If it was me, I choose x86 since x86 is widely used for now.

Developers for Xcode don't care about arm or x86. It is true. But macos (arm) will lose the opportune to attract more developers from the x86 world, or even lose some developers during the transition. This is the point.

you know what would be a real twist. What if we are all wrong and Apple isn’t transitioning to ARM, and instead they are getting rid of x86 products. What if Apple just decided they didn’t want to do desktops anymore, and that they could expand iPad to cover the mobile space they cared about.

And as the big kicker, they either released xcode For linux or Windows, or Visi’ll Studio extensions so that visual studio on PC could with full comparability work with xcode projects to transition them to visual studio projects/solutions.

I’m sure this will NOT be what is happening, I’d have my head spinning for a week if it did. But at the same time I do sometimes feel like Apple may only be reluctantly keeping their desktop/laptop lines going.

 

[Gerdi]

But macos (arm) will lose the opportune to attract more developers from the x86 world

Not sure what a "developer from the x86 world" is, but developer do not care much about the CPU architecture. In fact if you want to categorize developers, separation would be more about OS and programming language.

 

[Wilderness-1902]

Not sure what a "developer from the x86 world" is, but developer do not care much about the CPU architecture. In fact if you want to categorize developers, separation would be more about OS and programming language.

developers don't care for CPU arch? It is good for you.
[automerge]1591934583[/automerge]

Not sure what a "developer from the x86 world" is, but developer do not care much about the CPU architecture. In fact if you want to categorize developers, separation would be more about OS and programming language.

Really? For example, let's say if you work on a web app, Chrome on Android(arm) or Chrome on Android(x86) might have different behavior, since the underlying chip provide different video encoders.
[automerge]1591934708[/automerge]

you know what would be a real twist. What if we are all wrong and Apple isn’t transitioning to ARM, and instead they are getting rid of x86 products. What if Apple just decided they didn’t want to do desktops anymore, and that they could expand iPad to cover the mobile space they cared about.

And as the big kicker, they either released xcode For linux or Windows, or Visi’ll Studio extensions so that visual studio on PC could with full comparability work with xcode projects to transition them to visual studio projects/solutions.

I’m sure this will NOT be what is happening, I’d have my head spinning for a week if it did. But at the same time I do sometimes feel like Apple may only be reluctantly keeping their desktop/laptop lines going.

This is what I really want apple to do. It is stupid to buy a new machine just can run Xcode only for me

 

[MyopicPaideia]

If you were a college student of computer science, you choose a x86 or arm computer?

If it was me, I choose x86 since x86 is widely used for now.

Developers for Xcode don't care about arm or x86. It is true. But macos (arm) will lose the opportune to attract more developers from the x86 world, or even lose some developers during the transition. This is the point.

If I were a student today and needed to complete coursework? Probably x86. In 2-3 years? Probably ARM. Mark these words. macOS may not be a first target for developers and not compelling enough to have an ARM Mac just to get Xcode for - but you yourself made the point that developers target the highest profit market first. That is by far and away the Apple ecosystem, iOS/iPadOS in particular. People on Apple devices in general spend WAY more money, and this isn’t just me making claims, there is a lot of market data out there.

So while macOS isn’t the real reason developers want to use Xcode, iOS/iPadOS is. Let’s be clear we are talking about consumer and prosumer software markets here, which is where the volume and scale is. iOS/iPadOS is a multi billion dollar industry by itself.

Very few developers making mobile software are making Android apps and then bringing them to iOS. It is the other way around these days. Apple is very willing to lose those developers who treat their ecosystem as an afterthought. They are often anti-apple anyway.

Developers making web-based software for use in browsers don’t care much about ISA either, as they are developing towards a browser (i.e. Chrome), not even a particular OS, it is even another layer of abstraction.