M6 How Big of a Jump Are We Looking At?

[BenRacicot]

So the M6 rumors have me pretty hyped. From what I'm reading, we're getting three major changes all at once:

1. TSMC's 2nm node (vs current 3nm)
2. WMCM packaging (replaces InFo - allows side-by-side or stacked chip components)
3. New architecture (rumors of modular design)

TSMC claims 2nm gives around 15% performance boost, but when you stack all three of these changes together... could we be looking at >50% performance improvement over M5? That seems almost too good to be true but the packaging + architecture changes could add a lot on top of the node shrink.

The big question though:

Could WMCM packaging enable Ultra-style configurations in the MacBook Pro?

Right now Ultra is two Max dies connected together, which generates a ton of heat - fine for a desktop but impossible in a laptop. But if WMCM lets Apple arrange components more efficiently with better thermals, could we actually see something like an M6 "Ultra" variant in a 16" MBP? Or at least some kind of beefed-up configuration that wasn't thermally feasible before?

I'm probably being too optimistic here, but the new packaging seems like it opens doors that didn't exist with the current approach.

And, what kind of performance boost can we realistically be targeting with 3 major updates to the M series?

A dedicated GPU chiplet could mean 60-80 GPU cores in an M6 Max!

What do you think - realistic or just wishful thinking?

 

[BenRacicot]

If you work the math backwards from what we “have” in the M4 GPU…

M4 Max Current Specs:
• 40 GPU cores
• Total package power: ~90-95W under full load (CPU + GPU)
• GPU portion: roughly 50-60W at full tilt (educated guess based on teardowns/reviews)
The 2nm Efficiency Gain
TSMC claims 2nm delivers same performance at 25-30% less power (or 15% more performance at same power).
So if M4 Max GPU uses ~55W for 40 cores:
• M6 GPU at 2nm: 40 cores would use ~38-40W for same performance
• That frees up 15-17W of thermal/power headroom
How Many More Cores Can We Fit?
If we assume linear scaling (which isn’t perfect but close enough):
• 40 cores = 38W (2nm efficient)
• Each core ≈ 0.95W
• With 15-17W freed up: +16-18 more cores
• Total: 56-58 GPU cores at the same power envelope
But Wait - Chiplet Advantages
With a dedicated GPU chiplet and better WMCM thermal management:
• Better heat dissipation (not sharing die space)
• Could potentially push 5-10W higher on GPU without thermal issues
• That’s another 5-10 cores
Realistic ceiling: 60-65 GPU cores in M6 Max MBP

 

[maflynn]

So the M6 rumors have me pretty hyped.

Really? While there's been some discussions, I don't think its reached "hyped" status

I'm probably being too optimistic here,

Sounds like it

just wishful thinking?

Yes.

At this stage, the M5 hasn't been released yet, any talk of the M6 is rumor, conjecture and wishful thinking.

 

[pshufd]

The vast majority of users are fine, or better than fine with M4. And even the chips that Intel and AMD are putting out. Power users may want more but a lot of customers want other comfort features from laptops and desktops which is why Apple is rumored to be adding those kinds of features. Apple won the CPU war and they are working on other aspects of their products going forward. They're still going to improve their processors but they're going to have to do the other stuff too - and I find the other stuff more interesting than the CPU these days.

 

[maflynn]

fine with M4

You can make a case that the majority of consumers will be fine with an M2. I see on reddit, people buying M2 studios, are incredibly happy with saving a lot of money and getting a fast high performing machine

 

[pshufd]

You can make a case that the majority of consumers will be fine with an M2. I see on reddit, people buying M2 studios, are incredibly happy with saving a lot of money and getting a fast high performing machine

We're running M1 so that's true but the M4 comes standard with 16 GB of RAM which is overall more comfortable for everyone. Apple has won the CPU race and it's resulted in competition to get AMD, Intel and Qualcomm to up their game.

 

[maflynn]

but the M4 comes standard

The M4 is a game changer, no question. I'm not saying anything negative towards that. The M4 mini is so powerful, at a low price - that's a game changer. My M4 Max Studio is the best desktop I've owned in years.

 

[theolderoldcoot]

I wouldn't believe all the hype that your seeing. It usually isn't as great as people like to write about when the product hits the ground.

 

[thenewperson]

I wouldn't believe all the hype that your seeing. It usually isn't as great as people like to write about when the product hits the ground.

Based on what?

 

[Fallforward]

So the M6 rumors have me pretty hyped. From what I'm reading, we're getting three major changes all at once:

1. TSMC's 2nm node (vs current 3nm)
2. WMCM packaging (replaces InFo - allows side-by-side or stacked chip components)
3. New architecture (rumors of modular design)

TSMC claims 2nm gives around 15% performance boost, but when you stack all three of these changes together... could we be looking at >50% performance improvement over M5? That seems almost too good to be true but the packaging + architecture changes could add a lot on top of the node shrink.

The big question though:

Could WMCM packaging enable Ultra-style configurations in the MacBook Pro?

Right now Ultra is two Max dies connected together, which generates a ton of heat - fine for a desktop but impossible in a laptop. But if WMCM lets Apple arrange components more efficiently with better thermals, could we actually see something like an M6 "Ultra" variant in a 16" MBP? Or at least some kind of beefed-up configuration that wasn't thermally feasible before?

I'm probably being too optimistic here, but the new packaging seems like it opens doors that didn't exist with the current approach.

And, what kind of performance boost can we realistically be targeting with 3 major updates to the M series?

A dedicated GPU chiplet could mean 60-80 GPU cores in an M6 Max!

What do you think - realistic or just wishful thinking?

Please define "node shrink" because it's not 3nm -> 2nm. For example, the 2nm contacted gate pitch is 45nm and metal pitch is 20nm.

 

[dmccloud]

The stacked components theory is an interesting one, because TSMC already uses similar technology with AMDs X3D chips. They even flipped the order of the CPU and cache on the 9000 series, which has made noticeable improvements in both performance and core temperatures. AMD also uses what is effectively a modular design with their use of CCDs in the Ryzen series. So we know TSMC has both the capability and the experience to build products in that fashion.

Please define "node shrink" because it's not 3nm -> 2nm. For example, the 2nm contacted gate pitch is 45nm and metal pitch is 20nm.

This is fairly simple. gate pitch is the minimum distance between parallel gates on the die. The node refers to the (relative) size of the transistors themselves rather than the actual measureable size. There is also no industry-wide standard for what constitutes an "(x)nm" process, which is why Intel basically made up their own naming scheme to appear like they were pulling ahead of Samsung and TSMC when the reality was they couldn't even get existing processes working right.

 

[Falhófnir]

I probably wouldn't expect anything unusually impressive. Apple being able to deliver consistent (if not earth shattering) performance and efficiency gains year over year is impressive enough for where the chip market currently is. For recent chips the node shrink hasn't necessarily meant an outsized increase in performance. As it now has to last 3 cycles it seems Apple likes to leave a bit on the table for the M7 and M8 rather than taking full advantage all at once. I'd expect the usual 10-20% improvement, and be pleasantly surprised if they deliver more. Graphics is a bit of a wildcard as they seem to have a bit more headroom there, though it seems the M5 will already be a fairly big jump so maybe next year they will focus on the CPU side, or other features of the chip (efficiency, neural engine) again?

 

[ls1dreams]

Hope a big one but I'm expecting more diminishing returns moving forward. I'm on M1 Max right now and intend to skip M5 since I don't expect much out of it. M6 + OLED + maybe cellular could be a nice generational bump.

We prob just see another 12-15% bump or so. The iPhone 16 pro ->17 pro single threaded bump was not that impressive. (~3400->3800 or 12% or so), though we're still on 3nm. A lot of that probably also just came from better thermals and I don't really expect macbooks to change that much in that regard.

 

[senttoschool]

Great thread OP. Ignore the miserable responses you've gotten so far.

I think M6 generation Macbooks will truly make many M1 holdouts finally upgrade. Between the all new design, OLED, N2, and hopefully 2nd generation matmul in GPU cores, this is going to be a monstrous upgrade for people like me who are still on M1 Pro MBP.

 

[thenewperson]

Great thread OP. Ignore the miserable responses you've gotten so far.

I think M6 generation Macbooks will truly make many M1 holdouts finally upgrade. Between the all new design, OLED, N2, and hopefully 2nd generation matmul in GPU cores, this is going to be a monstrous upgrade for people like me who are still on M1 Pro MBP.

Yeah that’s my upgrade target from my M1 Pro as well.

 

[BenRacicot]

@senttoschool thanks so much for the positivity! Do keep in touch!

@ls1dreams the highest-end M6 series will not be a "diminishing returns" product unless Apple rug pulls it on us.

Here is another math'd outlook I put together after seeing Mistal 3's release.

Memory bandwidth = (MT/s × bus width) / 8
M4 Max uses a 512-bit bus:
8,533 MT/s × 512 / 8 = 546 GB/s

(Not really enough bandwidth for LLM throughput, the weights have to travel through this.)

Just upgrading to LPDDR6 (same 512-bit bus):

Speed	Bandwidth
10,000 MT/s	640 GB/s
12,000 MT/s	768 GB/s
14,400 MT/s	921 GB/s

So LPDDR6 alone gets us past 900 GB/s without any bus changes. My 800 GB/s estimate was too conservative.

And if Apple widens the bus (likely for Max/Ultra):

Bus Width	@ 14,400 MT/s
512-bit	921 GB/s
768-bit	1,382 GB/s
1024-bit	1,843 GB/s

An M6 Ultra with a 1024-bit bus could theoretically hit 1.8 TB/s—over half of an H100's bandwidth.

This would be enormous for running smaller LLMs locally.

 

[ls1dreams]

@senttoschool thanks so much for the positivity! Do keep in touch!

@ls1dreams the highest-end M6 series will not be a "diminishing returns" product unless Apple rug pulls it on us.

Here is another math'd outlook I put together after seeing Mistal 3's release.

Memory bandwidth = (MT/s × bus width) / 8
M4 Max uses a 512-bit bus:
8,533 MT/s × 512 / 8 = 546 GB/s

(Not really enough bandwidth for LLM throughput, the weights have to travel through this.)

Just upgrading to LPDDR6 (same 512-bit bus):

Speed	Bandwidth
10,000 MT/s	640 GB/s
12,000 MT/s	768 GB/s
14,400 MT/s	921 GB/s

So LPDDR6 alone gets us past 900 GB/s without any bus changes. My 800 GB/s estimate was too conservative.

And if Apple widens the bus (likely for Max/Ultra):

Bus Width	@ 14,400 MT/s
512-bit	921 GB/s
768-bit	1,382 GB/s
1024-bit	1,843 GB/s

An M6 Ultra with a 1024-bit bus could theoretically hit 1.8 TB/s—over half of an H100's bandwidth.

This would be enormous for running smaller LLMs locally.

Hope so, M5 was kind of underwhelming. I roughly plan to upgrade M6 or M7 time from my M1 Max (24core igpu).

Given how expensive everything has gotten though not sure if I'd even buy a max model anymore and definitely not an ultra. I find very little reason to use most local LLM's now given how much better the cloud stuff is unless I need some privacy.

 

[senttoschool]

Hope so, M5 was kind of underwhelming. I roughly plan to upgrade M6 or M7 time from my M1 Max (24core igpu).

Really? I think M5 was an outstanding upgrade.

4.3k GB6 ST score.
As much as 45% faster GPU.
Matmul in GPU makes AI workloads up to 4x faster.
Faster memory bandwidth too.

It's a fantastic gen over gen upgrade. Just 1 year.

 

[leman]

If you work the math backwards from what we “have” in the M4 GPU…

M4 Max Current Specs:
• 40 GPU cores
• Total package power: ~90-95W under full load (CPU + GPU)
• GPU portion: roughly 50-60W at full tilt (educated guess based on teardowns/reviews)
The 2nm Efficiency Gain
TSMC claims 2nm delivers same performance at 25-30% less power (or 15% more performance at same power).
So if M4 Max GPU uses ~55W for 40 cores:
• M6 GPU at 2nm: 40 cores would use ~38-40W for same performance
• That frees up 15-17W of thermal/power headroom
How Many More Cores Can We Fit?
If we assume linear scaling (which isn’t perfect but close enough):
• 40 cores = 38W (2nm efficient)
• Each core ≈ 0.95W
• With 15-17W freed up: +16-18 more cores
• Total: 56-58 GPU cores at the same power envelope
But Wait - Chiplet Advantages
With a dedicated GPU chiplet and better WMCM thermal management:
• Better heat dissipation (not sharing die space)
• Could potentially push 5-10W higher on GPU without thermal issues
• That’s another 5-10 cores
Realistic ceiling: 60-65 GPU cores in M6 Max MBP

You are overly optimistic. Even if we assume that we get 2.5D/3D packaging with improved logic density, I don't see how that would allow 50% or more GPU cores. Maybe — and I am being very generous here — 48 GPU cores in a Max would be a possibility, but already that is stretching it.

On the other hand, increasing the number of cores is not the only way to improve the GPU performance. M5 sees a notable boost for example, and that without raising the number of cores and with only moderate frequency increase.

 

[ls1dreams]

You are overly optimistic. Even if we assume that we get 2.5D/3D packaging with improved logic density, I don't see how that would allow 50% or more GPU cores. Maybe — and I am being very generous here — 48 GPU cores in a Max would be a possibility, but already that is stretching it.

On the other hand, increasing the number of cores is not the only way to improve the GPU performance. M5 sees a notable boost for example, and that without raising the number of cores and with only moderate frequency increase.

I personally just want to see better single threaded performance. Multi-core doesn't matter that much.

More gpu cores and memory bandwidth mostly would help with AI workloads.

 

[ls1dreams]

Really? I think M5 was an outstanding upgrade.

4.3k GB6 ST score.
As much as 45% faster GPU.
Matmul in GPU makes AI workloads up to 4x faster.
Faster memory bandwidth too.

It's a fantastic gen over gen upgrade. Just 1 year.

GPU/AI stuff was good but only around 12% single threaded uplift. For no lithography change I guess it's good though.
Anything is better than the old intel days of 5-7% cpu annual bumps. I guess I just feel spoiled with some previous years of seeing 15-20% single threaded bumps.

 

[leman]

GPU/AI stuff was good but only around 12% single threaded uplift. For no lithography change I guess it's good though.
Anything is better than the old intel days of 5-7% cpu annual bumps. I guess I just feel spoiled with some previous years of seeing 15-20% single threaded bumps.

M5 is currently the fastest single-threaded CPU, period. We also got the biggest IPC increase since M1. You expect quite a lot, my friend.

 

[Croke]

You are overly optimistic. Even if we assume that we get 2.5D/3D packaging with improved logic density, I don't see how that would allow 50% or more GPU cores. Maybe — and I am being very generous here — 48 GPU cores in a Max would be a possibility, but already that is stretching it.

On the other hand, increasing the number of cores is not the only way to improve the GPU performance. M5 sees a notable boost for example, and that without raising the number of cores and with only moderate frequency increase.

What if the NPU was removed. With the M5's addition of neural acceleration to the GPU cores we're seeing a bit of redundancy here aren't we? That's a bit expensive at the chip level.

I feel like it was an always a bit of a bandaid solution anyway - to look like they had an active hardware roadmap for AI, especially on the A-series chips. But in my (limited) experience, you either used the gpu OR npu at the app level for acceleration, and when the gpu core counts increase withe the pro, max and ultra the npu benefit falls behind performance-wise to an almost irrelevance.

It's possible, if any of the above holds water, that the M5 is a transition to a full-fledged tilt at building a scalable gpu-centric response to pytorch/cuda.

 

[thenewperson]

I feel like it was an always a bit of a bandaid solution anyway - to look like they had an active hardware roadmap for AI

This is very silly. The ANE has been there since 2018, way before the “AI” craze happened. People need to get their timelines straight before making comments in here, seriously.

 

[Croke]

This is very silly. The ANE has been there since 2018, way before the “AI” craze happened. People need to get their timelines straight before making comments in here, seriously.

... and Apple was widely regarded as being flat-footed when the "AI craze" rolled in. Which only further makes the point that it was a band-aid solution for something that it was never designed for. If its machine learning functions can be readily accommodated by a more robust and capable solution, why hang on to it?