The case of an Ultra chip using two M2 Pro chips

[Rickroller]

I think there is a pretty good chance that there will be a second Ultra version for the M2 family of chips. The Pro chip isn’t a cut down version of the Max chip this time around, it’s designed as a unique full unit, and is probably capable of being used in an Ultra Fusion assembly when scaled up. The upcoming Mac Pro would benefit from having more than one option powering its chassis, and the more balanced M2 Pro could do the trick.

there is also another reason to add this chip to the family, and that’s to allow Apple to add an Ultra addition to the MacBook lineup. When you look at what Inten and AMD are planning to release as soon as next month for the X86 laptop market, it would be an opportunistic marketing coup that would at least place some competitive CPU performance pressure where they may have thought there was none.

As Apple generally designs for a specific thermal power envelope, I think a possible M2 Pro Ultra should be fine in the 16 inch Pro chassis. Doubling the wattage of the Pro chip to 120W is of course more than the M2 Max 90W limit, but we already have the Max chip in the 14 inch MacBook Pro, and that also has a relatively
small chassis and battery for the Chip it’s running.

Also adding this less beastly Ultra would help to close the gap in the Mac Studio lineup. At the moment the base model starts at $1999 for the Max and Jumps to the M1 Ultra at $3999, which isn’t ideal. Another good thing about the Pro Ultra is that for some users the extra cores are more important than the extra GPU cores.

The M2 Pro Ultra

24 CPU cores
38 GPU cores
Double neural
double double all the codecs
and 144GB RAM @ 600 GB/s Bandwidth

 

[Pressure]

Those particular specifications do not make much sense but I agree that there would be a place for a larger chip than the M1 Max (432mm²) / M2 Max but smaller than the UltraFusion M1 Ultra (~864mm²) in the line-up.

However it would only be for the Mac Studio and Mac Pro.

 

[vanc]

to add an Ultra addition to the MacBook lineup.

Very unlikely. It's not just about thermals, but also the product positioning. Apple definitely wouldn't want too much overlap between different segments. Ultra chip would be too overpowered for a laptop.

 

[leman]

I think there is a pretty good chance that there will be a second Ultra version for the M2 family of chips. The Pro chip isn’t a cut down version of the Max chip this time around, it’s designed as a unique full unit

What do you mean “this time around”? M1 Pro wasn’t a “cut down” version of Max either.

and is probably capable of being used in an Ultra Fusion assembly when scaled up.

What are you basing this on?

 

[Rickroller]

Very unlikely. It's not just about thermals, but also the product positioning. Apple definitely wouldn't want too much overlap between different segments. Ultra chip would be too overpowered for a laptop.

Compared to the Intel 13th gen, doesn’t seem overpowered at all. The CPU part itself would only be 60w, which is a lot less wattage than what other mobile chips can consume.

 

[Rickroller]

What do you mean “this time around”? M1 Pro wasn’t a “cut down” version of Max either.



What are you basing this on?

Sorry, not cut down per se, but the chip looked a lot more similar to a Max that was cut down at the middle of the GPU. As for whether the ultra fusion can be added, I’m not sure the chip shown on stage wasn’t a distraction from what they actually plan to do.

 

[Sydde]

Sorry, not cut down per se, but the chip looked a lot more similar to a Max that was cut down at the middle of the GPU.

In what way does the M2 Pro die not look like a shaved M2 Max die?

 

[Rickroller]

In what way does the M2 Pro die not look like a shaved M2 Max die?

The layout looks different, as in where the parts are placed. For example the 3 core GPU tile in the Pro is placed higher up closer to the CPU cores.

 

[leman]

Sorry, not cut down per se, but the chip looked a lot more similar to a Max that was cut down at the middle of the GPU. As for whether the ultra fusion can be added, I’m not sure the chip shown on stage wasn’t a distraction from what they actually plan to do.

I cant say I can see what you seem to be seeing. And sure, the published pictures could have been edited, but that posibility alone doesn’t lend credibility to your argument .

 

[Sydde]

To me, the M2 Pro

Spoiler: that I have seen

looks exactly like the

Spoiler: M2 Max that I have seen

with half the GPU array missing (perhaps not "cut off", as such, but just using part of the same reticle). There are some small differences, but those look more like an effect of the photo exposure/angle.

 

[Rickroller]

I cant say I can see what you seem to be seeing. And sure, the published pictures could have been edited, but that posibility alone doesn’t lend credibility to your argument .

The argument isn’t actually technical. I’m not at all knowledgeable about the merits of actually making this part, more whether or not it makes sense to exist in the lineup. Before the Ultra was presented, it was only the rumors from leakers like German that even gave us an idea of how Apple would scale their silicon stack upwards in the desktop.

Maybe I should have included my thinking was the increased volume of Pro chips they’re making by adding the Mac mini Pro, and maybe the larger MacBook Air, would make sense as Ultra chips as well. Especially as I think the pro chip is very well balanced if your workflow is based on CPU workloads.

 

[Rickroller]

To me, the M2 Pro

Spoiler: that I have seen

looks exactly like the

Spoiler: M2 Max that I have seen

with half the GPU array missing (perhaps not "cut off", as such, but just using part of the same reticle). There are some small differences, but those look more like an effect of the photo exposure/angle.

These images are why I got the idea they should/could make an Ultra version of the smaller chip. isn’t it funny that its a given that the M2 Max Ultra is a certainty, but we can’t imagine a smaller version…?

 

[leman]

The argument isn’t actually technical. I’m not at all knowledgeable about the merits of actually making this part, more whether or not it makes sense to exist in the lineup. Before the Ultra was presented, it was only the rumors from leakers like German that even gave us an idea of how Apple would scale their silicon stack upwards in the desktop.

Maybe I should have included my thinking was the increased volume of Pro chips they’re making by adding the Mac mini Pro, and maybe the larger MacBook Air, would make sense as Ultra chips as well. Especially as I think the pro chip is very well balanced if your workflow is based on CPU workloads.

I think I understand where you are coming from - you are primarily interested in the CPU and not the GPU. Until now, Apple always delivered balanced designs, we will have to wait and see if hey will also build some more asymmetrical ones.

BTW, it’s entirely possible that there won’t be an M2 Ultra. It could be that M2 family is just a stopgap until N3 is in mass production, which will come with a whole new set of technologies we’ve seen in recently published Apple patents (new on-chip network, new memory protocols, new GPU cores etc.).

 

[theorist9]

Another good thing about the Pro Ultra is that for some users the extra cores are more important than the extra GPU cores.

The M2 Pro Ultra

24 CPU cores
38 GPU cores
Double neural
double double all the codecs
and 144GB RAM @ 600 GB/s Bandwidth

I would think there are some for whom the gap lies in the opposite direction—those who don't need more than the 12 CPU cores on the Pro, but do want more GPU power, particularly given the gap in GPU performance between AS and the NVIDIA 4000-series. And for the future, that would certainly be the case if Apple AAA gaming ever becomes a thing.

 

[Sydde]

Apple always delivered balanced designs, we will have to wait and see if hey will also build some more asymmetrical ones

I get the feeling that modern CPUs are reaching a wall of sorts. Not that they cannot get better, but, at least in the consumer-level market, there is just not that much to be gained from increases in CPU performance. Apple has been decorating the SoC with other circuitry that offsets the computational lifting that the CPU has had to handle in the past, and it is fast approaching the point where SPEC/GB5 scoring will be effetively meaningless WRT M-series devices.

 

[Rickroller]

I think I understand where you are coming from - you are primarily interested in the CPU and not the GPU. Until now, Apple always delivered balanced designs, we will have to wait and see if hey will also build some more asymmetrical ones.

BTW, it’s entirely possible that there won’t be an M2 Ultra. It could be that M2 family is just a stopgap until N3 is in mass production, which will come with a whole new set of technologies we’ve seen in recently published Apple patents (new on-chip network, new memory protocols, new GPU cores etc.).

Yeah CPU is as important as GPU for a lot of people.

why is there an expectation that M3 is going to be available so soon…? I would guess that iPhone takes priority for any new chips considering how many units they sell of those vs MacBooks…

Personally I think that Apple isn’t really going for stopgap solutions as it’s not like their chips are underwhelming in an Apple silicon market that’s barely over two years old. The whole marketing around M2 is still in comparison to Intel systems, so they must have data telling them there are still a large pool of holdouts to convert.

Looking at the new MacBook Pros just released, they seem to have been on track for a 12 month update if something unexpected hadn’t happened.

 

[theorist9]

I get the feeling that modern CPUs are reaching a wall of sorts. Not that they cannot get better, but, at least in the consumer-level market, there is just not that much to be gained from increases in CPU performance. Apple has been decorating the SoC with other circuitry that offsets the computational lifting that the CPU has had to handle in the past, and it is fast approaching the point where SPEC/GB5 scoring will be effetively meaningless WRT M-series devices.

Not sure what you mean. Even with the M-series processors, users still report delays and lags when using MS Office and Adobe software. Faster SC speeds would help with that (as would, of course, better-written programs).

 

[theorist9]

why is there an expectation that M3 is going to be available so soon…? I would guess that iPhone takes priority for any new chips considering how many units they sell of those vs MacBooks…

OTOH, if initial availability of the N3 process is limited, then it's a lot easier to get enough N3 chips for the Mac Studio than it is for the iPhone. Note also that they only put the A16/N4P chips in the iPhone 14 Pros; the iPhone 14's got the the A15/N5. If they wanted to put the newest chips in the highest-volume products, then all the iPhones would have gotten A16's.

 

[Rickroller]

OTOH, if initial availability of the N3 process is limited, then it's a lot easier to get enough N3 chips for the Mac Studio than it is for the iPhone. Note also that they only put the A16/N4P chips in the iPhone 14 Pros; the iPhone 14's got the the A15/N5. If they wanted to put the newest chips in the highest-volume products, then all the iPhones would have gotten A16's.

I’m not sure if I remember correctly, but isn’t there something in hardware on the A16 needed to run the Dynamic Island…?

 

[sam_dean]

Very unlikely. It's not just about thermals, but also the product positioning. Apple definitely wouldn't want too much overlap between different segments. Ultra chip would be too overpowered for a laptop.

I think what prevents the 5nm M1/M2 Ultra being in a MBP 16" is the

- die shrink
- power consumption
- thermals
- price points

Ultra doubles the price of a Mac Studio. Is there a large enough market for a laptop of that price point?

3nm M3 Ultra in a MBP 16" may occur in mid 2024 to early 2025.

As for Mac chip with mostly CPU cores & lesser GPU cores or mostly GPU cores and lesser CPU cores it appears that Apple is focusing on volume use cases and not unique ones.

Unique ones like Wolfram Mathematica users who want as much CPU cores as possible with as few GPU cores. Odds are scientists do not outnumber gamers, gaming app developers or 8K video editors. As such most of the SoCs are for volume use case.

 

[senttoschool]

It would be a lot easier to cut down an Ultra chip from 2x Max than to make another Ultra from 2x Pro. They already have an M1 Ultra with 20CPU/48GPU cores. If you cut down 8 GPU cores, you'd have essentially the same SoC the OP is describing.

 

[Rickroller]

I think what prevents the 5nm M1/M2 Ultra being in a MBP 16" is the

- die shrink
- power consumption
- thermals
- price points

Ultra doubles the price of a Mac Studio. Is there a large enough market for a laptop of that price point?

3nm M3 Ultra in a MBP 16" may occur in mid 2024 to early 2025.

As for Mac chip with mostly CPU cores & lesser GPU cores or mostly GPU cores and lesser CPU cores it appears that Apple is focusing on volume use cases and not unique ones.

Unique ones like Wolfram Mathematica users who want as much CPU cores as possible with as few GPU cores. Odds are scientists do not outnumber gamers, gaming app developers or 8K video editors. As such most of the SoCs are for volume use case.

i think the 16 inch MacBook Pro should be able to handle a 120w chip, and I’m not sure that there are a lot of workloads that would pin both CPU and GPU to the limit for extended periods of time out side of benchmarking.

Pricing is something to consider of course, but what is a halo product such as super-powered MacBook Ultra worth…? Changing the logic board and beefing up cooling is doable I would think, and adding $1000 would sting for sure, but you can’t buy a Ferrari for Camry prices either.

A Mac Studio with this chip should be a reasonable tier at $2999…? Why is a user who only needs the CPU from the Ultra forced to pay for the extra GPU…? I bet the number in this boat isn’t insignificant when you look at the price jumps to the higher tier stuff.

2024 is when I think the next series of M chips arrive, but that’s a while from now. The die shrink is used to add cores, not reduce the the package power. As in the M3 chips on 3nm will still mean the M3 Pro will be a 60w part, and the M3 Max will be a 90w part, so still to power hungry in a MacBook in an Ultra with 180w.

A MacBook Pro is sold to customers who use them to make money. Especially the 14 and 16 inch models. It’s a very profitable section of the market, and unsurprisingly they built their silicon to cater directly to the specific workloads these user target. certainly gaming is not part of the equation.

 

[sam_dean]

i think the 16 inch MacBook Pro should be able to handle a 120w chip, and I’m not sure that there are a lot of workloads that would pin both CPU and GPU to the limit for extended periods of time out side of benchmarking.

Pricing is something to consider of course, but what is a halo product such as super-powered MacBook Ultra worth…? Changing the logic board and beefing up cooling is doable I would think, and adding $1000 would sting for sure, but you can’t buy a Ferrari for Camry prices either.

A Mac Studio with this chip should be a reasonable tier at $2999…? Why is a user who only needs the CPU from the Ultra forced to pay for the extra GPU…? I bet the number in this boat isn’t insignificant when you look at the price jumps to the higher tier stuff.

2024 is when I think the next series of M chips arrive, but that’s a while from now. The die shrink is used to add cores, not reduce the the package power. As in the M3 chips on 3nm will still mean the M3 Pro will be a 60w part, and the M3 Max will be a 90w part, so still to power hungry in a MacBook in an Ultra with 180w.

A MacBook Pro is sold to customers who use them to make money. Especially the 14 and 16 inch models. It’s a very profitable section of the market, and unsurprisingly they built their silicon to cater directly to the specific workloads these user target. certainly gaming is not part of the equation.

Another thing to consider is how the Max was designed to produce Ultras. To my understanding the Pro may not be setup in a way to create Pro-based Ultras.

Any supply chain person would limit their SKUs to the most saleable ones so inventory remains lean. Your use case does not appear to be in a volume that Apple sees worthwhile to the bottomline.

It is like asking for M2 Ultra with 48 CPU cores and just 1 GPU core & no media engine. What is the expect annual units shipped for that configuration? How much should be the MSRP?

Die shrinks do allow for further reduced power consumption. See the iPads & iPhones of late. Battery life extended for the MBA M1 5nm vs to Intel Mac 14nm.

 

[Rickroller]

It would be a lot easier to cut down an Ultra chip from 2x Max than to make another Ultra from 2x Pro. They already have an M1 Ultra with 20CPU/48GPU cores. If you cut down 8 GPU cores, you'd have essentially the same SoC the OP is describing.

But isn’t it just easier to choose from a pre selected set and reduce the time needed to sort chips…? The smaller dies also cost less to produce.

 

[sam_dean]

But isn’t it just easier to choose from a pre selected set and reduce the time needed to sort chips…? The smaller dies also cost less to produce.

Cheaper based on silicon wafer surface area (material) but not leading edge node fab equipment.