M2 Platform Wish List and Predictions

[Macative]

The M1 is based on the A14 cores. It made sense for the M2 to be based on the A15 cores. There’s probably a development gap between the M2 and its iterations with the A16. Apple would not only do a regular M2, but also M2 Pro, Max, and Ultra. The A16 would probably be too early in the stages for Apple to develop the M2 ultra.

I could be wrong. We’ll find out.

You are. There is no development gap. These things are developed in parallel.

First M2 Macs Expected to Launch Later This Year Using TSMC's 4nm Chip Process

Apple plans to launch a series of Macs with M2 chips based on TSMC's 4nm process later this year, according to Taiwanese publication DigiTimes. This advancement should allow for continued performance and power efficiency improvements. "TSMC's 4nm process will also build Apple's M2 chip that will...

www.macrumors.com

Regardless of source, it is expected that M2 will be based on 4nm process.

 

[playtech1]

Apple makes all of its Mac money on the BTO upgrades. Can anyone find another $1000 laptop on the market that comes with only 8 GB of RAM?

It really isn't an argument about whether or not 8 GB is enough for base model users (it is). It is a question of whether the price is appropriate for it (its not). It is all about charging you that extra $200 for 16 GB.

I agree - but... I think a spec bump to 16GB will be used as justification to bump the price of the entry level MacBook Pro models.

Moving to 16GB and charging an extra $100 on every MacBook sale should better for Apple than $200 on those that configure them through the Apple Store and with inflation being what it is and they might even be thanked for it!

IIRC the move to 4GB from 2GB was 2010 and the move from 4GB to 8GB as the base tier came in 6 years later with the 2016 models. I know these things aren't written in stone, but it's definitely approaching a time when 8GB should get dropped in the Pro models. To be honest it's a joke that it's not already the standard RAM size (as is the 256GB SSD, which is a capacity so low that it is hard to even find for sale at retail these days).

 

[JPack]

It just seems odd to me that only the Studio Display gets this feature on the Mac (perhaps it gets it because it's basically an iPad internally), plus the MBP notch is massive relative to the actual camera, so there seems to be space for a more sophisticated camera setup (albeit not much depth).

The problem with Center Stage isn't space, but rather the feature doesn't fit the needs of MacBook Pro users. This has been discussed before.

To get Center Stage, an ultrawide camera is needed. This UW lens creates distortion. To get a proper selfie, the camera has to crop and center. This results in a lower quality photo. This is why the current iPad Pro M1 has a lower quality selfie than the previous gen.

 

[darngooddesign]

...but it's definitely approaching a time when 8GB should get dropped in the Pro models.

Apple already did that. You can't buy a Mac Studio or MacBook Pro with 8GB.

 

[Macative]

Moving to 16GB and charging an extra $100 on every MacBook sale should better for Apple than $200 on those that configure them through the Apple Store and with inflation being what it is and they might even be thanked for it!

Eh, I don't know. If 1000/1200 is replaced with one 1100 model, it works out to be about the same for Apple in terms of overall revenue, but only if they don't lose customers by raising the entry level price. So there isn't much to gain for them, and every potential to lose. And no real benefit to customers as many are more interested in that $999 price point than 16 GB of RAM.

 

[JPack]

I agree - but... I think a spec bump to 16GB will be used as justification to bump the price of the entry level MacBook Pro models.

Moving to 16GB and charging an extra $100 on every MacBook sale should better for Apple than $200 on those that configure them through the Apple Store and with inflation being what it is and they might even be thanked for it!

IIRC the move to 4GB from 2GB was 2010 and the move from 4GB to 8GB as the base tier came in 6 years later with the 2016 models. I know these things aren't written in stone, but it's definitely approaching a time when 8GB should get dropped in the Pro models. To be honest it's a joke that it's not already the standard RAM size (as is the 256GB SSD, which is a capacity so low that it is hard to even find for sale at retail these days).

It's easier to attract customers with a low base price. Once customers are hooked, they're often willing to upgrade memory and storage.

The entry level MacBook Pro is used by a lot of business professionals who don't need 16GB memory but still want the CPU for number crunching spreadsheets, better display, and battery life.

 

[JMacHack]

Being realistic I think we’ll see:

Slightly improved single thread speed. (Maybe slower than Raptor Lake because nobody cares if a nuclear warhead can beat it if it fries itself) possibly more P-cores on higher end models, more e-cores on lower end. More cores = better Multi thread perf.

More improved, possibly more numerous gpu cores. Currently that’s where ASi falls short and the A15 saw the largest improvement there.

More than two displays max on the base model (big IF, partially based on my assumption of more gpu cores)

Improved memory bandwidth (maybe, would provide more performance uplift than just improving the cores)

Higher memory capacity. (Apple isn’t gonna charge less, keep dreaming)

 

[cnnyy20p]

Apple always increase clock speed very iteration. It’s going be faster as you wish.
The less unlikely one is an option for 2 efficiency cores to 2 performance cores due to how they design their SoC.

 

[PsykX]

All I want is 16GB RAM as one of the pre-configured base model.

It would make sense.
A computer with 8GB today is a cheap computer. 8GB in a luxury product is not working well. Yes, a Mac Mini, be it low entry model, is considered a luxury product because of the Mac branding.

Now, I know and I can confirm by experience that 8GB in Apple Silicon is much better than 8GB in a PC, but I'd say it's worth around 12GB, we're still 33% short.

What I'm expecting with M2 :
- 20% faster CPU (like iPhone 13)
- 50% faster GPU (like iPhone 13 Pro)
- 16GB base models
- 3 monitors (which means a MacBook Air will be able to have 2 monitors)

If they do all of that, with the Mac Mini and MBA redesigns, these things are going to sell like hot breads.

 

[Zdigital2015]

It made sense before.

But keep in mind that the base M series chip now goes into the Macbook Air, Mac Mini, iMac, iPad Pro 10, iPad Pro 13, iPad Air.

In the past, it just went into the iPad. There are now 5 reasons to refresh at least the base M SoC once a year.

I would like to see Apple refresh the M series each year, and the M Ultra & 2x M Ultra every 2 years.

Apple is setting the precedent now for the M-Series SoC to be on a two-year cadence even while the iPhone Ax SoC is refreshed on an annual basis. I don’t see this cadence changing especially with chip process changes running at a fairly consistent 18-24 month cadence.

 

[exoticSpice]

Apple makes all of its Mac money on the BTO upgrades. Can anyone find another $1000 laptop on the market that comes with only 8 GB of RAM?

It really isn't an argument about whether or not 8 GB is enough for base model users (it is). It is a question of whether the price is appropriate for it (its not). It is all about charging you that extra $200 for 16 GB.

Dell - XPS 13" Touch-Screen Laptop - Intel Core i7 - 8GB Memory - 512GB Solid State Drive - Platinum Silver​

there's more but here's one and this one is >$1500 and has 8GB RAM

 

[dieselm]

Apple is setting the precedent now for the M-Series SoC to be on a two-year cadence even while the iPhone Ax SoC is refreshed on an annual basis. I don’t see this cadence changing especially with chip process changes running at a fairly consistent 18-24 month cadence.

They spun an Ax chip for iPads along with the A chip for phones nearly yearly from A5 to A12 (and M1, essentially a rebranded A14X). Things went a little different as the iPad line expanded and the A chips got more powerful.

But yearly is certainly within their past and present capabilities. The consumer lines like iMac, iPad Air and Macbook Air benefit much from "freshness".

The Pro/Max/Ultra lines are more about additional cores/memory/co-processors. We wouldn't miss too much going every other year, though traditionally Apple did refresh the Macbooks and iMacs at least with spec bumps each year too.

Process technology (7nm, 5nm, 4nm, etc) is slowing to a 24-36 month cadence with less notable gains.
Guess we'll see.

 

[sunny5]

- Faster SD Card reader

There is no faster SD card standard. SD Express is not even available to major vendors.

 

[exoticSpice]

There is no faster SD card standard. SD Express is not even available to major vendors.

there is UHS III

 

[sunny5]

there is UHS III

I said who is using UHS III? None, especially for all cameras.

 

[jdb8167]

They spun an Ax chip for iPads along with the A chip for phones nearly yearly from A5 to A12 (and M1, essentially a rebranded A14X). Things went a little different as the iPad line expanded and the A chips got more powerful.

Not really. The first Ax chip was the A8X which came out in October 2014. Then just about one year later the A9X came out in September 2015–slightly less than one year. A10X was June 2017 or about 20 months after the A9X. The A12X came out in end of October 2018. Another 16 months later. The A12Z is probably irrelevant to the discussion since it appears to be an A12X with one GPU core turned back on. Finally the M1 was the longest iteration coming in on November 2020, a little over two years after the A12X.

It is pretty clear from that timeline, the AX SoCs started off annually but quickly went to 16 months to two years and with the M1 it was over two years. I don’t think we can gain much insight over that irregular cadence except that over time the time between Ax SoCs generally got longer.

 

[Malus120]

Something I'd really like to see with the M2 that I haven't seen mentioned yet is:

-A move to the ARMv9-A instruction set

Back with the A7, Apple basically pioneered AArch64 (being the first implementation by almost a year.) While Apple Silicon is certainly pushing ARM in completely new ways on the desktop (their interconnect technology is insane for example) it'd be nice to see them stay ahead of the curve as far as the base ARM feature-set is concerned.
(Not a software dev but the new Scalable Vector Extension 2 (SVE2) and Scalable Matrix Extension (SME) both sound nice, as do some of the other features.)
One of the difficulties I feel we're already seeing with Apple Silicon is that Apple has a bunch of proprietary tech that, while really cool, is going to take a lot of work for devs to get the most out of. Putting some energy into having the latest ARM features would hopefully take some of the pressure off both by ensuring that we get to take advantage of the latest optimizations devs are making for use across the entire ARM ecosystem, while also putting Apple in a position where it can leverage its position to push for faster adoption of those features.

Obviously for this to happen I'd imagine the A16 would have to be based on v9, and the M2 based on that. I don't know that the likelihood is super high but hey, I can dream right

Also just to reiterate what's already been said by others the M2 really needs to have support for at least 2 (external) displays (3 including the internal) on the base configurations (or, even better decouple # of displays supported from the SOC config.)

 

[leman]

Something I'd really like to see with the M2 that I haven't seen mentioned yet is:

-A move to the ARMv9-A instruction set

Back with the A7, Apple basically pioneered AArch64 (being the first implementation by almost a year.) While Apple Silicon is certainly pushing ARM in completely new ways on the desktop (their interconnect technology is insane for example) it'd be nice to see them stay ahead of the curve as far as the base ARM feature-set is concerned.
(Not a software dev but the new Scalable Vector Extension 2 (SVE2) and Scalable Matrix Extension (SME) both sound nice, as do some of the other features.)
One of the difficulties I feel we're already seeing with Apple Silicon is that Apple has a bunch of proprietary tech that, while really cool, is going to take a lot of work for devs to get the most out of. Putting some energy into having the latest ARM features would hopefully take some of the pressure off both by ensuring that we get to take advantage of the latest optimizations devs are making for use across the entire ARM ecosystem, while also putting Apple in a position where it can leverage its position to push for faster adoption of those features.

Obviously for this to happen I'd imagine the A16 would have to be based on v9, and the M2 based on that. I don't know that the likelihood is super high but hey, I can dream right

Also just to reiterate what's already been said by others the M2 really needs to have support for at least 2 (external) displays (3 including the internal) on the base configurations (or, even better decouple # of displays supported from the SOC config.)

Both technologies you mention (SVE and SME) are extensions over ARMv8, so strictly speaking there is need for them to support the latest standard for that. The only proprietary tech in Apple ARM I am aware of is AMX (matrix stuff), and Apple doesn’t expose it, instead hiding it behind libraries. I suppose that can be replaced with SME…

 

[ian87w]

Something I'd really like to see with the M2 that I haven't seen mentioned yet is:

-A move to the ARMv9-A instruction set

Back with the A7, Apple basically pioneered AArch64 (being the first implementation by almost a year.) While Apple Silicon is certainly pushing ARM in completely new ways on the desktop (their interconnect technology is insane for example) it'd be nice to see them stay ahead of the curve as far as the base ARM feature-set is concerned.
(Not a software dev but the new Scalable Vector Extension 2 (SVE2) and Scalable Matrix Extension (SME) both sound nice, as do some of the other features.)
One of the difficulties I feel we're already seeing with Apple Silicon is that Apple has a bunch of proprietary tech that, while really cool, is going to take a lot of work for devs to get the most out of. Putting some energy into having the latest ARM features would hopefully take some of the pressure off both by ensuring that we get to take advantage of the latest optimizations devs are making for use across the entire ARM ecosystem, while also putting Apple in a position where it can leverage its position to push for faster adoption of those features.

Obviously for this to happen I'd imagine the A16 would have to be based on v9, and the M2 based on that. I don't know that the likelihood is super high but hey, I can dream right

Also just to reiterate what's already been said by others the M2 really needs to have support for at least 2 (external) displays (3 including the internal) on the base configurations (or, even better decouple # of displays supported from the SOC config.)

Good point. And actually now I believe it will. Snapdragon 8gen1 already adopted ARMv9, so I can't think that Apple would want to be behind. So that could mean the M2 will be alongside the A16 instead of the A15.

 

[mr_roboto]

Both technologies you mention (SVE and SME) are extensions over ARMv8, so strictly speaking there is need for them to support the latest standard for that. The only proprietary tech in Apple ARM I am aware of is AMX (matrix stuff), and Apple doesn’t expose it, instead hiding it behind libraries. I suppose that can be replaced with SME…

No, Apple made more proprietary extensions to ARM than just AMX - unless you're only thinking of things which add additional instructions?

Apple Silicon Hardware Secrets: SPRR and Guarded Exception Levels (GXF) | Sven Peter

blog.svenpeter.dev

Scroll down to "tl;dr" first, then read the rest of it if you're interested in the details. SPRR and GXF are a really cool pair of features, imo, used both for security and to accelerate JIT engines (the latter by making it very fast to flip pages of JITted code between RW- and R-X permissions from userspace).

In addition to SPRR and GXF, there's also the special mode bits for Rosetta. There's one which makes a core obey x86 style memory ordering rules instead of more relaxed ARM-legal rules, and another which (iirc) makes it possible for plain Neon computational instructions to produce bit-exact results compared to their SSE equivalents. (The IEEE 754 standard has enough wiggle room that different compliant implementations are permitted to produce slightly different outputs from the same input data. The choices on how to use that wiggle room in x86 SSE are slightly different from Neon, so this was necessary to achieve x86 emulation which is both fast and exact.)

There's probably other stuff I haven't read about or don't remember.

 

[leman]

No, Apple made more proprietary extensions to ARM than just AMX - unless you're only thinking of things which add additional instructions?

Apple Silicon Hardware Secrets: SPRR and Guarded Exception Levels (GXF) | Sven Peter

blog.svenpeter.dev

Scroll down to "tl;dr" first, then read the rest of it if you're interested in the details. SPRR and GXF are a really cool pair of features, imo, used both for security and to accelerate JIT engines (the latter by making it very fast to flip pages of JITted code between RW- and R-X permissions from userspace).

In addition to SPRR and GXF, there's also the special mode bits for Rosetta. There's one which makes a core obey x86 style memory ordering rules instead of more relaxed ARM-legal rules, and another which (iirc) makes it possible for plain Neon computational instructions to produce bit-exact results compared to their SSE equivalents. (The IEEE 754 standard has enough wiggle room that different compliant implementations are permitted to produce slightly different outputs from the same input data. The choices on how to use that wiggle room in x86 SSE are slightly different from Neon, so this was necessary to achieve x86 emulation which is both fast and exact.)

There's probably other stuff I haven't read about or don't remember.

Yeah, I had user-facing functionality in mind. That Apple tweaked a lot of things on the low-level goes without saying. Already their custom interrupt handling is a huge change. Thanks for the link, cool stuff, haven't seen it together like this yet!

Speaking about SSE and Neon... I've been doing more SIMD work this past week and getting deeper into Neon, it's just incredible how much saner and more versatile ARM's approach is. I was able to cut down the amount of instruction needed by 30% on average for my geometry routines compared to SSE. But getting around all those intrinsics is a pain, there are hundreds...

 

[thenewperson]

Not really. The first Ax chip was the A8X which came out in October 2014. Then just about one year later the A9X came out in September 2015–slightly less than one year. A10X was June 2017 or about 20 months after the A9X. The A12X came out in end of October 2018. Another 16 months later. The A12Z is probably irrelevant to the discussion since it appears to be an A12X with one GPU core turned back on. Finally the M1 was the longest iteration coming in on November 2020, a little over two years after the A12X.

It is pretty clear from that timeline, the AX SoCs started off annually but quickly went to 16 months to two years and with the M1 it was over two years. I don’t we can gain much insight over that irregular cadence except that over time the time between Ax SoCs generally got longer.

The first AX chip was the A5X in 2012. Quickly replaced by the A6X the same year. Next was A8X 2 years after. In the timeline of AX chips we've mainly gotten the even numbered ones, with a few odd numbered ones sprinkled in:

A5X (45nm) - March 2012
A6X (32nm) - October 2012
A8X - (20nm; TSMC takes over here) - October 2014
A9X - (16nm) - September 2015
A10X - (10nm) - June 2017
A12X - (7nm) - October 2018
M1 (5nm) - November 2020

It's clear they coincide with die shrinks more than a yearly or dual-yearly schedule, although it seems from this that 2 year cycles are way more common.

I think it's sensible to make a guess that the M2 could be based on A16 cores given this list. I am curious if this report will come to fruition though. It wouldn't be the first time they've done something like this, with the A10 being 16nm but the A10X being 10nm. However it either means Apple would use risk-production chips, which seems unlikely.

 

[killawat]

Apple Silicon Hardware Secrets: SPRR and Guarded Exception Levels (GXF) | Sven Peter

blog.svenpeter.dev

Great link. I don't think I'm brainy enough to make it through.

 

[Macative]

Dell - XPS 13" Touch-Screen Laptop - Intel Core i7 - 8GB Memory - 512GB Solid State Drive - Platinum Silver​

there's more but here's one and this one is >$1500 and has 8GB RAM

Nice find! However after looking at it, I don't think it is really comparable. This is a touch screen, and has a 512 GB SSD standard. I don't consider the touch screen to be a benefit at all, but it certainly drives up the price of the laptop significantly. And with twice the storage of the MacBook Air, there is that to consider too.

So yeah, this is certainly a $1000+ laptop with 8 GB of RAM, but if you take away with touch screen and extra storage it is more like a $600 laptop.

 

[Macative]

I think it's sensible to make a guess that the M2 could be based on A16 cores given this list.

It is sensible to think that because:
- There was no M2 that followed the A15.
- It is already reported to be based on the A16.
- Apple doesn't update Macs quickly, and isn't going to update their chips until it is worthwhile. Like, the performances gains to be had from a new fab process.