Gurman: Apple Testing 'M3 Pro' Chip for MacBook Pro With 12-Core CPU and 18-Core GPU

[deconstruct60]

We know the Mac Pro is coming, Apple has told us. We also know it is going to be M3 since TSMC confirmed this with their financial disclosure requirements that require them to disclose information to their shareholders. No way Apple can prevent this disclosure though I am sure their respective legal departments have a tug of war over what gets disclosed.

Where did TSMC disclose the specifics of the Apple M3 implementation? If you are trying to claim that TSMC has "HPC" N3 products coming soon therefore if-so-facto-presto-changeo that has to be the Mac Pro M3 SoC, then you need to go back to check what TSMC throws in the "HPC" category. Plain old A-series goes in the "HPC" category. They are really talking about anything that isn't a very low power embedded controller. Not the super rarified air of "supercomputers".

What Mark Gurman may have uncovered was a compute module for the upcoming Mac Pro. The module will have to have ECC memory. The ECC controller will be within the AS. So, externally, it looks like there is 36 GB of RAM but 4GB is used for ECC. So, this also suggests there will be a unique Mac Pro SiP. I have come to realize this is necessary if they use ECC. If they ditch the ECC then they could use the same modules. Any reason they would ditch the ECC? Does the memory being tightly coupled give them the same reliability?

Pretty good chance Apple doesn't touch ECC with a ten foot pole. (at least "at rest" ECC that isn't supported by the baseline LPDDR5 standard. )

 

[chucker23n1]

We also know it is going to be M3 since TSMC confirmed this

TSMC wouldn’t know this kind of stuff to begin with.

What Mark Gurman may have uncovered was a compute module for the upcoming Mac Pro. The module will have to have ECC memory. The ECC controller will be within the AS. So, externally, it looks like there is 36 GB of RAM but 4GB is used for ECC. So, this also suggests there will be a unique Mac Pro SiP.

I doubt 36 GiB is even in the ballpark of what matters for the Mac Pro, and they also aren’t going to use a SiP for it.

 

[hans1972]

Playing 4K videos in the background and a bunch of tabs open. I love Firefox Mozilla browser. It's my 2nd go-to web browser. Lately, Safari has been acting up.

Next: Blind people demand 4K. They're tired of not seeing 1080p. Now they want to not see 4K.

 

[koyoot]

We know the Mac Pro is coming, Apple has told us. We also know it is going to be M3 since TSMC confirmed this with their financial disclosure requirements that require them to disclose information to their shareholders. No way Apple can prevent this disclosure though I am sure their respective legal departments have a tug of war over what gets disclosed.

What Mark Gurman may have uncovered was a compute module for the upcoming Mac Pro. The module will have to have ECC memory. The ECC controller will be within the AS. So, externally, it looks like there is 36 GB of RAM but 4GB is used for ECC. So, this also suggests there will be a unique Mac Pro SiP. I have come to realize this is necessary if they use ECC. If they ditch the ECC then they could use the same modules. Any reason they would ditch the ECC? Does the memory being tightly coupled give them the same reliability?

M3 Max has 20 CPU cores: 12 Performance and 8 Efficiency, and 48 GPU cores, while M3 Ultra: 40 CPU cores, 24 Performance and 16 Efficiency and 96 GPU cores.

Thats what will land in Mac Pro.

 

[tipoo]

TSMC wouldn’t know this kind of stuff to begin with.

Not saying they leaked it or not, but uh, they absolutely would, you know they work closely with their clients on bringing their intended designs to TSMC transistor libraries, they know what Apple has coming

 

[steve123]

M3 Max has 20 CPU cores: 12 Performance and 8 Efficiency, and 48 GPU cores, while M3 Ultra: 40 CPU cores, 24 Performance and 16 Efficiency and 96 GPU cores.

Thats what will land in Mac Pro.

Very true. Apple would have prototyped various configurations of the M3 silicon along the way. Some of these configurations might not be end use products but rather they may be test vehicles along the way to various end use products.

I expect the Mac Pro will offer several versions of compute modules for the Mac Pro with various CPU and memory configurations. If Apple goes in this direction, users will be able to build and upgrade their Mac Pro over several generations of compute modules.

 

[steve123]

and they also aren’t going to use a SiP for it.

It is not clear what you are trying to say here? Every Mx is a System in Package that integrates a CPU die with discrete memory devices into a single package.

 

[CWallace]

Not saying they leaked it or not, but uh, they absolutely would, you know they work closely with their clients on bringing their intended designs to TSMC transistor libraries, they know what Apple has coming

Yes TSMC knows what they are manufacturing for Apple, and if they are making a CPU SoC on 3nm they can assume it is M3, but I expect Apple and TSMC are not using marketing names like "M3, M3 Pro, etc." on the contracts.

Is it pedantic? Perhaps. But it is not beyond the realm of possibility that this next class of Mac SoC could be the same as M2, just on a newer 3nm process, and Apple could call it "M2 Plus".

 

[tipoo]

Yes TSMC knows what they are manufacturing for Apple, and if they are making a CPU SoC on 3nm they can assume it is M3, but I expect Apple and TSMC are not using marketing names like "M3, M3 Pro, etc." on the contracts.

Is it pedantic? Perhaps. But it is not beyond the realm of possibility that this next class of Mac SoC could be the same as M2, just on a newer 3nm process, and Apple could call it "M2 Plus".

Ok, Apple could call it the Gazpacho Whizbang for some god unknown reason to be fair, but TSMC would certainly know the node, the transistor budget, and the design features as to make a fairly good guess it's the successor to the M2.

 

[koyoot]

Very true. Apple would have prototyped various configurations of the M3 silicon along the way. Some of these configurations might not be end use products but rather they may be test vehicles along the way to various end use products.

I expect the Mac Pro will offer several versions of compute modules for the Mac Pro with various CPU and memory configurations. If Apple goes in this direction, users will be able to build and upgrade their Mac Pro over several generations of compute modules.

The chips will be exactly the same way designed as M1 and M2 series.

The specs will be different from M1 and M2 series. Compute Module is something similar to Intel's Compute Element.

 

[steve123]

Pretty good chance Apple doesn't touch ECC with a ten foot pole.

I am not sure I understand why? There appear to be techniques to implement ECC with LPDDR5:

Error Correction Code (ECC) in DDR Memories | Synopsys IP

Explore how ECC memory enhances DDR reliability, preventing data corruption and system failures effectively.

www.synopsys.com

Of those, side-band ECC could be implemented into an Mx with "36 GB of memory" if Apple used a 2 x 128 Gb LPDDR5 devices for storage and one 32 Gb LODDR5 for the ECC codes?

 

[DeepIn2U]

Woo hoo! 36GB unified Memory will be insane 🔥. We need more RAM. I'm tired of seeing this message in 2023.

View attachment 2201846

Reading your post I thought WITAF aid going on! Seeing your update is even more shocking.

This has been happening with a number of apps in Ventura. For me, I end up somehow using over 300 GB of memory with the Music app. Odd to me, because it only has 16 GB of memory 😆 #jokes (memory leaks).

Has this ONLY been happening in Ventura or since macOS 11 as I think that rolled out with M1 originally.

Playing 4K videos in the background and a bunch of tabs open. I love Firefox Mozilla browser. It's my 2nd go-to web browser. Lately, Safari has been acting up.

Huh? Who? Yeh try to temp me to technology but it’s not working. /s

Agreed! Memory leak is real and it exists!

Is memory leak more of an issue to Apple Silicon? Hopefully Apple is aware and can release bug fixes to resolve this?

LOL. What in Frodo’s name are you doing in Firefox that eats up 79.2 Gb??
-edit- OK, a video and a few tabs… must be a memory leak I think…

But why is a memory leak seemingly this big and unresolved by Apple this far in Venture release cycle?!

That’s what scares me after the release of the M1 with these CPU’s bound to absolutely minimal RAM.

For the longest time, since iOS 7 and with macOS 11 Big Sur I wondered why Apple would dumb down both OS’ use of 3 dimensional icons and borders with drop shadows and depth if the cpu/gpu is so powerful above the competition. Now I realize the limitation was forced upon by minimal packaged ram!!

Can we also get the iconic glowing Apple logo back, please? I wish Apple had never eliminated it from the Mac's. 💻

View attachment 2201864

I would SERIOUSLY love to have this back. One would think Apple would have a unified presentation of the Apple logo across the Mac lineup or a dual stance : metal Apple logo on the pro’s (their about performance vs flash and style) and the non pro Mac’s (being of style) would have an optional backlight version - initiated by the end user.

Yet unfortunately just like in iOS, there’s too much going on that refinement and a consistent experience for features and changes mentioned during an iOS or Mac launch seem to not be so. e.g. modal pop-ups in iOS (still exists), but why?!

I live in Chrome and I have never seen that issue. This on a Mac Mini M1, with 16gig of RAM and a Parallels Win11 VM that is getting 8gigs running in the background.

FF has always been a stability mess for me. I get more broken websites on FF than I do with Safari.

I’m curious how Microsoft’s Edge browser performs (the one with Bing ChatGPT within), anyone try?!

I wonder why the "odd" size of 36GB of RAM. We know that RAM tends to revolve around eights or sixteens and neither of those goes smoothly into 36. I would assume 8-16-24-32 (as we have seen) 40-48-56-64, etc. options vs. an oddball like 36GB.

This is not a complaint- just noticing that "one of these is not like the others" and wondering why. Is this basically 5K (monitor) vs. 4K (or 8K) because... well... it's Apple?

I’m curious if the extra 4GB would be specifically reserved for macOS system preferences to deal with memory leaks? Just a thought.

 

[gank41]

Has this ONLY been happening in Ventura or since macOS 11 as I think that rolled out with M1 originally.

This has been happening to me off & on since I got this M1 MBP in late 2020. It's happened on Big Sur, Monterey (Mostly), and Ventura (not as much).

Is memory leak more of an issue to Apple Silicon? Hopefully Apple is aware and can release bug fixes to resolve this?

No, not M1 specific. Memory Leaks have been a thing for quite some time. Here's a good article that explains them in some great detail:

Memory leaks

What’s the difference between an app and kernel memory leak? How would you notice one, and how to investigate it, and (sometimes) work around the problem.

eclecticlight.co

But why is a memory leak seemingly this big and unresolved by Apple this far in Venture release cycle?!

That's the big question! In @TheYayAreaLiving 🎗️ 's issue, I don't think Mozilla can "fix" that. And in my issues, it's mostly been Apple's own Music app with the issue. If a first party app has this issue, STILL after all these years, I would expect a third party app to at some point, too.


Computers are hard I guess?

¯\_(ツ)_/¯

 

[chucker23n1]

It is not clear what you are trying to say here? Every Mx is a System in Package that integrates a CPU die with discrete memory devices into a single package.

They’re SoCs. The Apple Watch thing is an SiP.

 

[chucker23n1]

Not saying they leaked it or not, but uh, they absolutely would, you know they work closely with their clients on bringing their intended designs to TSMC transistor libraries, they know what Apple has coming

They’ll know Apple has requested a bigger design. They won’t know if that’s a BTO option on the Mac Studio, or an M2 Extreme, or an M3 Max. They can figure some of that out if they’re bored, but that isn’t their job.

 

[tipoo]

There's a twist to the RAM saga. Gurman has disqualified the idea of moving to 12GB modules. The base might start at 18GB.

So either ECC, or a fast HBM2 cache maybe? 2 for 16GB, 4 for 32GB

 

[DavidSchaub]

They’re SoCs. The Apple Watch thing is an SiP.

That is just how you cut the marketing? The Apple Watch motherboard is being given a name

 

[chucker23n1]

if they are making a CPU SoC on 3nm they can assume it is M3,

We can’t even assume that. Apple using a different process node for a higher-end variant isn’t unprecedented.

but I expect Apple and TSMC are not using marketing names like "M3, M3 Pro, etc." on the contracts.

Exactly.

And even if they knew (Apple itself may not even have finalized such marketing names; they usually do that very late), that doesn’t tell the story of what product it’s for.

Is it pedantic? Perhaps. But it is not beyond the realm of possibility that this next class of Mac SoC could be the same as M2, just on a newer 3nm process, and Apple could call it "M2 Plus".

Yup.

 

[chucker23n1]

That is just how you cut the marketing? The Apple Watch motherboard is being given a name

Right, but the Mac’s motherboard is very much not a single package. Not even the iPhone’s is. But the Watch’s is.

 

[chucker23n1]

There's a twist to the RAM saga. Gurman has disqualified the idea of moving to 12GB modules. The base might start at 18GB.

So either ECC, or a fast HBM2 cache maybe? 2 for 16GB, 4 for 32GB

It’s just six modules that are 3 each or 6 each.

 

[HobeSoundDarryl]

I’m curious if the extra 4GB would be specifically reserved for macOS system preferences to deal with memory leaks? Just a thought.

As others have since pointed out, while we're pretty accustomed in 2023 to multiples of 8s or 16s for RAM, if we step back in time, there were times when that was multiples of 4, 2, 1, 512K, 256K, 128K, 64K, 32K and so on.

My mis-assumption or wrong logical leap was that in times where multiples of 8 or 16 seem "normal" for RAM increases, there's nothing stopping Apple from leaning of multiples of 4, 2, 1, etc. 36 made me wonder if multiples of 8 were becoming multiples of 9 but that didn't intuitively make sense to me.

So while I was guessing typo- that maybe 36 was actually net of 40 or even 48- looking backwards, any multiple of 2 can make sense... or possibly even seemingly oddities like 16GB+64K of RAM if Apple desired it. I guess to make that make ANY sense, marketing could spin "over 16GB" instead of (specifically) "16GB." More is more???

 

[DavidSchaub]

Yes TSMC knows what they are manufacturing for Apple, and if they are making a CPU SoC on 3nm they can assume it is M3, but I expect Apple and TSMC are not using marketing names like "M3, M3 Pro, etc." on the contracts.

But I'm sure TSMC could figure it out. Heck, with just knowing how many chips per wafer, I could figure it out. From TSMCs perspective Apple only makes 2.5 big chip designs (M* and M* Max and M* Max cut down).

Is it pedantic? Perhaps. But it is not beyond the realm of possibility that this next class of Mac SoC could be the same as M2, just on a newer 3nm process, and Apple could call it "M2 Plus".

Moving to a new process node still takes lots of engineering and is really expensive. If you are doing that anyway, why in the world would you not also move to a newer/better architecture?

 

[koyoot]

There's a twist to the RAM saga. Gurman has disqualified the idea of moving to 12GB modules. The base might start at 18GB.

So either ECC, or a fast HBM2 cache maybe? 2 for 16GB, 4 for 32GB

There are possible two explenations.

The package is indeed 384 bit for M3 Pro, and 36 GB is indeed 36 GB, but that means that base config is 24 GB of RAM.

18 GB of RAM is impossible on 384 bit bus, its possible only on 192 bit bus.

Second possibility - its standard 256 Bit LPDDR5 memory controller plus 1024 HBM2 memory controller. The complexity of the package increases dramatically however, without any benefit to the performance.

Im still leaning towards 384 bit for M3 Pro because of simplicity of the design required.

 

[chucker23n1]

As others have since pointed out, while we're pretty accustomed in 2023 to multiples of 8s or 16s for RAM, if we step back in time, there were times when that was multiples of 4, 2, 1, 512K, 256K, 128K, 64K, 32K and so on.

Yes, but those are all powers or two. 36 instead has a prime factor of 3 in it.

 

[chucker23n1]

Moving to a new process node still takes lots of engineering and is really expensive. If you are doing that anyway, why in the world would you not also move to a newer/better architecture?

Maybe the newer microarchitecture isn’t ready.

For example, the A10 is 16nm, but the A10X 10nm. Same cores, just more of them.