Mac Silicon - What do you expect from the first Mac SoCs?

[aaronage]

Hey everyone!

Pretty new here! I'm excited to see what the Apple silicon team will deliver for the first Mac SoC. It's a chance for Apple's silicon team to take the Mac far beyond the capabilities and performance of generic x86 competition in the PC space. I feel many tech enthusiasts are underestimating just how transformative this transition could be.

I haven't seen a thread dedicated to speculation around the performance and features of the first Mac SoCs, so wanted to get a discussion going ? (if a thread does exist, sorry! link to it and I'll move there)

Leaks suggest that MacBook Pro will be the first Mac to get Apple Silicon, so I've had that on my mind since the announcement.

What might a MacBook Pro SoC look like?

Based on what we have already seen from iPad SoCs, rumours, and statements from the keynote, my conservative guesstimate is:
- ~25W TDP
- 5nm >200mm2 die size (monolithic)
- >3GHz peak boost
- 8P/4E CPU cores (as rumoured)
- ~4 TFLOP GPU
- 32MB L2
- Quad-channel LPDDR4X for >120GB/s
- All the other goodness like Neural Engine, custom video encode/decode, custom crypto etc.

I say conservative because I really don't think itt's a stretch based on what we've already seen from A12Z. The above is "just" iPad SoC "scaled up", so it's the bare minimum I would expect.

Things could get spicier than that, though! "Advanced Silicon Packaging" was called out during the Apple Silicon reveal. That could mean so many things - could be something exotic like 3D stacking, something like HBM memory on package, or it could be something relatively new like chiplets.

If I had to guess, I wouldn't expect the first generation SoC to be super exotic. It makes sense for Apple to ease into the transition and not go all out with novel packaging tech that hasn't been validated at scale. But then... why call out "Advanced Silicon Packaging" so early? Hmm ? Exotic packaging might make sense on lower volume Mac's first which would rule out anything MacBook.

Some bonus things we could see:
- USB 4
- PCI-E 4 for storage
- HBM instead of LPDDR4X? I'm doubtful since it would be very expensive, especially to offer a 32GB model.

My personal view is that Apple will be ahead of Intel Tiger Lake and AMD Renoir/Cézanne in traditional CPU and GPU workloads... buuuuut that's not the most exciting part! What will really separate Apple from Intel and AMD will be the combination of Apple's hardware accelerators (AMX, Neural Engine) and Apple's software frameworks (like the accelerate framework). If Apple has nailed the frameworks, and that's a big if, they could be many times faster than PC competition in many workloads. That's just... huge.

(sidenote: was anyone else excited about heterogeneous compute many years back? When AMD introduced Fusion? I was, it was never realised, but Apple might have just solved that problem ? )

What does everyone think? ?

Think I'm overestimating, underestimating or have just totally lost my mind? Let's discuss ??

Edit: Clarify the post refers to MacBook Pro SoCs and not other models.

 

[Tommytomtom]

I (probably over excitedly) expect that they will do a “no floppy drive” type release - not hold back and just get as much new tech into the package as possible.

iMac rumours are making me lean this way, and not just because of that “no floppy” history.

If a MacBook Air was the the first hardware then I’d think they might try to blow everyone’s socks off with a huge evolution in battery life and small form factor.

With an iMac as the first device they will want to make it as disruptive as the original iMac - and that means massive revolutionary changes in what a computer is; not just more of what we already understand a computer to be.

 

[Zackmd1]

Hey everyone!

Pretty new here! I'm excited to see what the Apple silicon team will deliver for the first Mac SoC. It's a chance for Apple's silicon team to take the Mac far beyond the capabilities and performance of generic x86 competition in the PC space. I feel many tech enthusiasts are underestimating just how transformative this transition could be.

I haven't seen a thread dedicated to speculation around the performance and features of the first Mac SoCs, so wanted to get a discussion going ? (if a thread does exist, sorry! link to it and I'll move there)

Leaks suggest that MacBook Pro will be the first Mac to get Apple Silicon, so I've had that on my mind since the announcement.

What might a MacBook SoC look like?

Based on what we have already seen from iPad SoCs, rumours, and statements from the keynote, my conservative guesstimate is:
- ~25W TDP
- 5nm >200mm2 die size (monolithic)
- >3GHz peak boost
- 8P/4E CPU cores (as rumoured)
- ~4 TFLOP GPU
- 32MB L2
- Quad-channel LPDDR4X for >120GB/s
- All the other goodness like Neural Engine, custom video encode/decode, custom crypto etc.

I say conservative because I really don't think itt's a stretch based on what we've already seen from A12Z. The above is "just" iPad SoC "scaled up", so it's the bare minimum I would expect.

Things could get spicier than that, though! "Advanced Silicon Packaging" was called out during the Apple Silicon reveal. That could mean so many things - could be something exotic like 3D stacking, something like HBM memory on package, or it could be something relatively new like chiplets.

If I had to guess, I wouldn't expect the first generation SoC to be super exotic. It makes sense for Apple to ease into the transition and not go all out with novel packaging tech that hasn't been validated at scale. But then... why call out "Advanced Silicon Packaging" so early? Hmm ? Exotic packaging might make sense on lower volume Mac's first which would rule out anything MacBook.

Some bonus things we could see:
- USB 4
- PCI-E 4 for storage
- HBM instead of LPDDR4X? I'm doubtful since it would be very expensive, especially to offer a 32GB model.

My personal view is that Apple will be ahead of Intel Tiger Lake and AMD Renoir/Cézanne in traditional CPU and GPU workloads... buuuuut that's not the most exciting part! What will really separate Apple from Intel and AMD will be the combination of Apple's hardware accelerators (AMX, Neural Engine) and Apple's software frameworks (like the accelerate framework). If Apple has nailed the frameworks, and that's a big if, they could be many times faster than PC competition in many workloads. That's just... huge.

(sidenote: was anyone else excited about heterogeneous compute many years back? When AMD introduced Fusion? I was, it was never realised, but Apple might have just solved that problem ? )

What does everyone think? ?

Think I'm overestimating, underestimating or have just totally lost my mind? Let's discuss ??

I think you are pretty damn spot on. I would be curious to see if they include some type of higher performance ram as well. Since they will be using an SOC with unified memory, the higher the bandwidth the better for the GPU. HBM would be way to expensive IMO but possibly DDR5??? It is technically available, just no CPU or desktop system supports it yet.

 

[leman]

I wouldn’t be surprised to see LPDDR5 in their first laptop this year. Next year, high performance larger model with HBM2/3 as system RAM.

By the way, I think your conservative guess is very realistic!

 

[topdrawer]

i'm hoping for 18 hour battery life, unless they go for superthin then we will be in the 10 range forever

 

[iPadified]

I agree, spot-on. They need to differentiate it from the iPad chips to demonstrate scalability for later addressing the high end desktops. There is also a rumoured 24 inch iMac arriving in the autumn and your specs on the chip are probably good enough for that as well. However, Apple could simply overclock and laptop chip for an iMac to get more performance out. An iMac can easily cool 50W and being nearly completely silent.

 

[Yebubbleman]

Hey everyone!

Pretty new here! I'm excited to see what the Apple silicon team will deliver for the first Mac SoC. It's a chance for Apple's silicon team to take the Mac far beyond the capabilities and performance of generic x86 competition in the PC space. I feel many tech enthusiasts are underestimating just how transformative this transition could be.

I haven't seen a thread dedicated to speculation around the performance and features of the first Mac SoCs, so wanted to get a discussion going ? (if a thread does exist, sorry! link to it and I'll move there)

Leaks suggest that MacBook Pro will be the first Mac to get Apple Silicon, so I've had that on my mind since the announcement.

What might a MacBook SoC look like?

Based on what we have already seen from iPad SoCs, rumours, and statements from the keynote, my conservative guesstimate is:
- ~25W TDP
- 5nm >200mm2 die size (monolithic)
- >3GHz peak boost
- 8P/4E CPU cores (as rumoured)
- ~4 TFLOP GPU
- 32MB L2
- Quad-channel LPDDR4X for >120GB/s
- All the other goodness like Neural Engine, custom video encode/decode, custom crypto etc.

I say conservative because I really don't think itt's a stretch based on what we've already seen from A12Z. The above is "just" iPad SoC "scaled up", so it's the bare minimum I would expect.

Things could get spicier than that, though! "Advanced Silicon Packaging" was called out during the Apple Silicon reveal. That could mean so many things - could be something exotic like 3D stacking, something like HBM memory on package, or it could be something relatively new like chiplets.

If I had to guess, I wouldn't expect the first generation SoC to be super exotic. It makes sense for Apple to ease into the transition and not go all out with novel packaging tech that hasn't been validated at scale. But then... why call out "Advanced Silicon Packaging" so early? Hmm ? Exotic packaging might make sense on lower volume Mac's first which would rule out anything MacBook.

Some bonus things we could see:
- USB 4
- PCI-E 4 for storage
- HBM instead of LPDDR4X? I'm doubtful since it would be very expensive, especially to offer a 32GB model.

My personal view is that Apple will be ahead of Intel Tiger Lake and AMD Renoir/Cézanne in traditional CPU and GPU workloads... buuuuut that's not the most exciting part! What will really separate Apple from Intel and AMD will be the combination of Apple's hardware accelerators (AMX, Neural Engine) and Apple's software frameworks (like the accelerate framework). If Apple has nailed the frameworks, and that's a big if, they could be many times faster than PC competition in many workloads. That's just... huge.

(sidenote: was anyone else excited about heterogeneous compute many years back? When AMD introduced Fusion? I was, it was never realised, but Apple might have just solved that problem ? )

What does everyone think? ?

Think I'm overestimating, underestimating or have just totally lost my mind? Let's discuss ??

I think you're pretty much on-point. I think if A14 supports LPDDR5X, then we'll see that if Apple's first Mac SoC is A14-based. If not, then we'll see LPDDR4X still.

Otherwise, I think the rumors about the 13" MacBook Pro being first makes total sense, especially since they are able to produce ARM chips that easily exceed that of the Intel 10th Generation U-series processors as of a year ago.

My prediction is that, while they could've done FaceID for the Mac the way Microsoft does Windows Hello facial recognition on current PCs, Apple would've wanted to wait until they had the Neural Engine to work with and that the first Mac portables with their SoC's will likely have FaceID as one of those "incidentally, here's what you're also getting on the new architecture" type of features much like Front Row and the built-in webcam was with the first MacBook Pros.

I (probably over excitedly) expect that they will do a “no floppy drive” type release - not hold back and just get as much new tech into the package as possible.

iMac rumours are making me lean this way, and not just because of that “no floppy” history.

If a MacBook Air was the the first hardware then I’d think they might try to blow everyone’s socks off with a huge evolution in battery life and small form factor.

With an iMac as the first device they will want to make it as disruptive as the original iMac - and that means massive revolutionary changes in what a computer is; not just more of what we already understand a computer to be.

What feature are you looking to drop here and why do you think it's necessary to drop a feature for the sake of dropping a feature?

People already hate the limited port selection on the current MacBook Pros and MacBook Airs. You could try to drop USB-A on the Mac Pro, iMacs, and Mac mini, but again, I ask you why?

I wouldn’t be surprised to see LPDDR5 in their first laptop this year. Next year, high performance larger model with HBM2/3 as system RAM.

By the way, I think your conservative guess is very realistic!

Most MacRumors predictions (mine included) are out there. A conservative guess is much more likely to be a realistic one.

That said, LPDDR5 already? The computing industry barely finished the move to DDR4. Some systems are still stuck on LPDDR3. Apple barely finished the move themselves for their laptops. I get we're talking a different architecture here, but it seems like it'd be at least a few years off.

 

[leman]

That said, LPDDR5 already? The computing industry barely finished the move to DDR4. Some systems are still stuck on LPDDR3. Apple barely finished the move themselves for their laptops. I get we're talking a different architecture here, but it seems like it'd be at least a few years off.

If I understand it correctly, LPDDR5 is already available and in volume production. Until now, Apple was limited by the memory controllers provided by the third party. In their own chips, they are usually fairly aggressive about adopting new RAM standards.Thats why I think it might be a possibility.

 

[pttai]

Will the first Apple Silicon come in a MacBook or a regular Mac?
I doubt it will be in a regular Mac.

 

[Tommytomtom]

What feature are you looking to drop here and why do you think it's necessary to drop a feature for the sake of dropping a feature?

I don’t mean that a feature will be dropped; just that they will go all-in on the new arch as opposed to just a toe in the water. That they will make a change nobody thought they had the courage to make.

The OP framed the options as either “gentle transition” vs “super exotic”. Aaronage also states that they feel most tech enthusiasts are underestimating the transformative nature - I strongly agree - but expect the wildest of his predictions will be the case.

Then again; the loss of support for 32bit apps has been handled before it would have been forced by ARM. Had it not, that could have been the no-floppy moment. Boot camp is being forced out; and they’ve gotten ahead of that news too.

But those points aside; my point was more about going all-in on what’s new.

 

[leman]

I don’t mean that a feature will be dropped; just that they will go all-in on the new arch as opposed to just a toe in the water. That they will make a change nobody thought they had the courage to make.

I think that ARM Macs will bring some important paradigm shifts to the PC world. First, AI acceleration as a standard computer feature (Intel does ship similar stuff in their newest CPUs, but the software integration is lacking). Second, TBDR GPUs and unified memory in a desktop environment. This is something I am particularly excited about as a graphics programming enthusiast - this is the biggest change since the advent of programmable shaders and a really smart way to do GPUs. Third, I expect Apple to move from modularity to full integration, delivering big efficiency improvements.

 

[ksec]

Assuming you did no Typo and it was indeed referring to "MacBook" ( unless you mean MacBook as a category ) , while the "direction" is right, some of the detail are slightly off. And it is not even "conservative" by any means. But very aggressive.

1. Macbook dont have 25W TDP. The current MacBook Air CPU has 12W Max TDP.

2. >200mm2 is no way conservative on a leading edge node. Just to give you an idea, a scale up of A12Z to 5nm will give you DOUBLE the CPU ( "both" HP and LP ), Double the GPU, Double the NPU, Double the SRAM with >50mm2 die space to spare.

3. You cant do Quad Memory Channel on 25W TDP, or at least you shouldn't.

But yes the idea of A14X will likely be like a double A12Z + Additional IPC and GPU generation improvement. It will likely be "sub 200mm2". Or even Sub 150mm2. And it would be enough to power everything from a possible MacBook, MacBook Air, Mac Mini. ( I actually think Apple may use A12Z on some of these products for cost reason )

Theoretically you could Max the A14X at possible close to 100W if you have all transistor running. That is 40W for CPU, 30W for GPU, 20W NPU and memory controller. If you hit 3+Ghz at 4W per Core inclusive of SRAM, 8 Core already gives you 32W, there isn't a lot of room to play with. So it should work on MacBook Pro. But it wouldn't in the power efficient range. And there is the question of whether Apple wants dGPU inside MacBook Pro. ( You are fundamentally limited to a Sub 100W whole system Design on MacBook Pro inclusive of Display, Storage and everything else due to flight restriction and battery size.)

Advanced Silicon Packaging is referring to TSMC's version of CoWoS, (Chip-on-Wafer-on-Substrate) or an earlier version of it. It is already being used on Apple S Series, Apple calls it SiP ( System in Package ).

 

[Yebubbleman]

I don’t mean that a feature will be dropped; just that they will go all-in on the new arch as opposed to just a toe in the water. That they will make a change nobody thought they had the courage to make.

The OP framed the options as either “gentle transition” vs “super exotic”. Aaronage also states that they feel most tech enthusiasts are underestimating the transformative nature - I strongly agree - but expect the wildest of his predictions will be the case.

Then again; the loss of support for 32bit apps has been handled before it would have been forced by ARM. Had it not, that could have been the no-floppy moment. Boot camp is being forced out; and they’ve gotten ahead of that news too.

But those points aside; my point was more about going all-in on what’s new.

I get your point. But my counter to that is that I don't know what else they can do at this point without upsetting users. Furthermore, I don't see how they're not likely to go all in on the architecture. It seems like that's the key intention and that the difference between the first Mac to ship with an Apple ARM processor and the DTK will be at least as substantial as the one between the Pentium 4 equipped Power Mac DTK and the iMac Core Duo. All that to say, I don't know how else you want Apple to go all in on the architecture. I doubt they'll hold back if that's what you're fearing.

 

[leman]

3. You cant do Quad Memory Channel on 25W TDP, or at least you shouldn't.

Intel does offer quad memory channel with their Ice Lake CPUs, but these are 32-bit LPDDR4x channels. Standard DDR is still Dual channel 64-bit.

 

[Zackmd1]

If I understand it correctly, LPDDR5 is already available and in volume production. Until now, Apple was limited by the memory controllers provided by the third party. In their own chips, they are usually fairly aggressive about adopting new RAM standards.Thats why I think it might be a possibility.

Samsung S20 series uses LPDDR5 so it should be readily available and I wouldn't doubt the A14 series chip will support it. From what I can tell, 6gb, 8gb, 12gb, and 16gb modules exist.

 

[aaronage]

Assuming you did no Typo and it was indeed referring to "MacBook" ( unless you mean MacBook as a category ) , while the "direction" is right, some of the detail are slightly off. And it is not even "conservative" by any means. But very aggressive.

1. Macbook dont have 25W TDP. The current MacBook Air CPU has 12W Max TDP.

2. >200mm2 is no way conservative on a leading edge node. Just to give you an idea, a scale up of A12Z to 5nm will give you DOUBLE the CPU ( "both" HP and LP ), Double the GPU, Double the NPU, Double the SRAM with >50mm2 die space to spare.

3. You cant do Quad Memory Channel on 25W TDP, or at least you shouldn't.

But yes the idea of A14X will likely be like a double A12Z + Additional IPC and GPU generation improvement. It will likely be "sub 200mm2". Or even Sub 150mm2. And it would be enough to power everything from a possible MacBook, MacBook Air, Mac Mini. ( I actually think Apple may use A12Z on some of these products for cost reason )

Theoretically you could Max the A14X at possible close to 100W if you have all transistor running. That is 40W for CPU, 30W for GPU, 20W NPU and memory controller. If you hit 3+Ghz at 4W per Core inclusive of SRAM, 8 Core already gives you 32W, there isn't a lot of room to play with. So it should work on MacBook Pro. But it wouldn't in the power efficient range. And there is the question of whether Apple wants dGPU inside MacBook Pro. ( You are fundamentally limited to a Sub 100W whole system Design on MacBook Pro inclusive of Display, Storage and everything else due to flight restriction and battery size.)

Advanced Silicon Packaging is referring to TSMC's version of CoWoS, (Chip-on-Wafer-on-Substrate) or an earlier version of it. It is already being used on Apple S Series, Apple calls it SiP ( System in Package ).

My bad, I meant MacBook Pro. I think MacBook Air class devices will end up sharing an SoC with iPad.

>200mm2 does seem a bit high now you mention it. The 5nm node is said to deliver an 80% reduction in area vs 7nm, that's pretty exciting.

Regarding quad-channel - I remember some of Apple's earlier X series SoCs having quad-channel memory controllers and achieved pretty high bandwidth. Is that something we could see return? (I'm a little ignorant about the finer details of memory controllers so maybe "quad-channel" doesn't mean what I think it means in the case of old SoCs ? )

CoWoS sounds pretty interesting, will have to do some reading on that ?