M2 Ultra in a Macbook Pro will be possible at 3nm or 2nm

[Wokis]

You think? I have no way to know either way, but I suspect between the cost, the size, weight and likliehood of throttling - the appeal would be miniscule.

The R&D, and production costs would be hard to justify it when the target audience is tiny.

Also the battery life as others mentioned would be horribly short and battery life is one of the major tent poles Apple markets their laptops. I mean, right now, I can run my 14" MBP on battery and its performance will be that as if it was plugged in. I'm not sure that would be possible with an ulta, and if it was, the battery life would be horribly short.

All speculation to be sure, but I just don't see much of a market for a M2 Ulta MBP.

R&D for the chip interconnect and the resulting chips is already done. The output of all that development is at the moment in just one singular niche product, speculated to be in two products total. That sounds like a bit of a waste to me. Why not a third product?

TDP of the current M1 Ultra at its absolute peak is estimated to be about 220-250W. This is not new territory for high-end laptops in general. Just for Apple. I have to imagine that designing a new case, PCB and thermal management system is a smaller task by comparsion to the original chip R&D.

Battery life would probably not be stellar if put under load, but given ARM and AS in general has great race-to-idle and great idle power draw in general, it would likely be ok as an everyday machine as well.

I'm very sure production people would love it. I'm a post-production person primarily so I don't mind desktops, but I meet and am involved in pre-production planning plenty of times. I know they'd love a crazy powerful laptop instead of lugging around workstations. That I can say for sure.

(Currently there are ~250W windows laptops available, but let's be clear here: those still don't measure up to desktop performance. Apple would do way more with that power budget, as has been clearly shown.)

Not saying it will happen, but I know Apple is more involved with talking with film/tv-industry people now and hearing us out. Researching into the extreme sports space made them make the Apple Watch Ultra, so I wouldn't rule it out that they could do something similar for demanding professionals in other spaces.

 

[Love-hate 🍏 relationship]

The known technical limit would be 240W USB PD charger.

Top-end Mac Studio M1 Ultra 5nm has a max power input of 215W without a display.

A M2 Ultra 3nm with at least a 20% performance per watt improvement would allow an Ultra in a MacBook Pro with 240W charger.

The question would be how big of a market is there for laptop that sells for nearly $7,000 and weighs nearly 2x 4.7 pounds (2.1 kg).

Mac Studio M1 Ultra's heat sink fan was made of a more thermally conductive material that weighs more so it can maintain its heat sink fan's RPM and physical form factor dimension.

As the M1 Ultra chip is 2x the die/surface area of a M1 Max chip then the heat it generates would be 2x as well. So that Macbook Pro M2 Ultra 3nm would run hotter than the top-end Macbook Pro 16 M1 Max.

Considering the latest Geekbench leaks I very much doubt that

It'll be 5nm

 

[fuchsdh]

2019 MBP 16" vs 2021 MBP 16" somewhat reversed this trend

Yeah, but it's still "within reason". A M1 16" MBP is up to 4.8lbs. That's heavier than its direct predecessor, but it's only by 12%— and it's only a 7% increase compared with the earlier retina models. These are still fairly light laptops as a whole—an HP Spectre 16" starts at 4.5lbs, while their Omen gaming laptops crack 5.4lbs. You're not going to find capable laptops with that screen size much lighter than the MBPs, even with their "bulk up".

Either way, they're definitely not going to release a battle station "portable" with an Ultra chip, especially since they can probably get more money out of customers that want that power by having them buy desktops. They've always strove for the lighter and thinner options in the entire modern history of the company, they just aren't apparently dieting products to extreme degrees just for thinness' sake anymore.

 

[maflynn]

R&D for the chip interconnect and the resulting chips is already done. The output of all that development is at the moment in just one singular niche product, speculated to be in two products total. That sounds like a bit of a waste to me. Why not a third product?

Yes, but we're talking about a new case and cooling solution, then there's the cost of retooling, where they need to produce the laptops. That's my point.

 

[deconstruct60]

R&D for the chip interconnect and the resulting chips is already done. The output of all that development is at the moment in just one singular niche product, speculated to be in two products total. That sounds like a bit of a waste to me. Why not a third product?

Mx Ultra : Mac Studio , Mac Pro , and stuffed into current iMac Pro chassis (with screen update ).

Three products and done. Apple already has yet another chassis on the shelf that can handle the Ultra TDP range (don't really need to 'invent' a new one). There would be some limited sales fratricide between those , but the net number of Ultra SoCs sold/deployed would likely go up.

The logic board area of the iMac Pro is much larger even though it also has two fans. ( they are spread much farther apart).

Folks who want very clean, smaller desktop footprint and an all-in-one would have a product. Folks who think modular screens are great, but want literal small desktop would have a product. Folks who think modular should extended to some set of PCI-e add in cards for I/O , storage , and networking would be have a product.

Would those collective sell as many as the aggregate MBP 14"/16" sales? Probably not. However, it does 'have to' to recoup the UltraFusion R&D costs. The Mac Pro likely will have an even more expensive SoC that drives much higher margins ( but not higher volume). Just need the money to cover the costs. Not necessarily totally in raw volume.


Actually would help effective wafer usability for the MBP 14"/16" Max SoCs also if they did not have a pragmatically useless UltraFusion connector on them. That will make even less sense as TSMC N3 , N2 , etc wafer costs climb even higher to have a contextually , completely functionally useless connector on the die.

For the M1 generation (with probably no Mac Pro) it made some sense to coupled the Max-with-connector die to the MPB 14"/16" Max sales to generate some volume to amortize costs over. Apple really have no hard numbers on how well the Mac Studio would sell. If it has sold poorly and had to solely pay for UltraFusion work in short term, that would be risky.

At this point though Apple has a much more grounded idea of how well the Mac Studio Ultra/Max will sell in what ratio. It would also be far more easy to see have a Mac Pro with Ultra would map into the aggregate numbers also.
They probably don't need the MBP crutch. Six months in the Studio Ultra variant is still backorder. It wasn't a sales volume flop.

Apple is likely going to try to stick with completely monolithic dies for the laptops as long as they can because it is an incrementally higher performance/watt approach. At some point they may have to fork off the Max sized die class into a disaggregated chiplet package set up. However, that is probably a couple of iterations down the road.

UltraFusion for a line up of desktop only SoCs makes far, far , far more sense. The M1 Max is a really far too chunky to be a good chiplet. Scaling past two "chiplets" really wouldn't work well at all for its size , shape, and RAM connectivity layout. It is really a not so good chiplet. It was a "handy" chiplet for a generation 1 limited line up that Apple had. But going forward Apple needs a better chiplet decomposition that scales better.

 

[MayaUser]

M2 Ultra probably will be just for the Mac Studio and maybe for an bigger imac/mac pro base
So take your dreams away from an laptop with M2 ultra under 5nm, 3nm etc

 

[deconstruct60]

Then why don’t they carry a Mac Studio and a portable battery?

Carrying all that plus a mouse, keyboard, and display is significantly more cumbersome than even a 7#, 18" laptop. You own a MBA instead of a MacMini for the same convenience reasons.

If going from fixed desktop location to fixed desktop location then really don't need much of a battery.
The Mac Studio plugs into the Studio Display docking station on either side (and just 'sneakernet the powered off Mac Studio between locations). The mouse , keyboard , and display are just duplicated at each end. ( just incrementally more if the Mac Studio is in the $4,000-6,000 range. )


If there is a 'remote location' production site ( 'rustic in the woods' site) then decent chance there are portable power generators for Lights , battery chargers , displays , etc. The Studio can just plug into one those with everything else.

It won't be maximum portable, but much better shape than trying to plug in some 1.5kw desktop monster tower.


If the M2/M3/M4 Max SoC improves to cover more of the performance space that the M1 Ultra covered then will get the "more power in laptop format" in 1-2 generations anyway. It is not like the M2 Max is going 'feeble' when given access to the same underlying fab tech that is being used to cram the Ultra into an oversized laptop.

There's probably a small, niche market for such a MBP-Ultra. If we weren't in a component shortage Apple might choose to serve it, but right now they won't.

Generationally that niche will get smaller and smaller. Why create a new product that is going to immediately go into decline? Apple has stated repetitively they are not out to try to make everything for everybody.

the general Windows PC market is around x8-x9 times bigger than the Mac market. So those narrow niches "in between" the major product catagories have x8-x9 times more folks in them. There are also dozens of other system vendors there looking for some zone to get a bigger trickle down coverage of users than other vendors. So more vendors chasing those individual niches. Mac space there is just one vendor. Apple isn't suffering from lack of system volume Mac sales.

Apple dropped the MBP 17" a long while back and they aren't likely coming back. MBP 16" is pretty close to the old MBP 15" size (more 'war' on bezels than affinity for larger laptop sizes. there is a notch on that 16" screen. If was about maximum pixels that wouldn't be there. ) . Apple hasn't returned to that space. That space makes substantially less sense in an era of one cable plugin docking station to a 24-32" screen docking station with ease. Or Chromecast/Airplay cast of a screen wireless to a presentation screen.

 

[theorist9]

Exactly, plus they don't realize how expensive is N3 silicon is from TSMC.
It's not gonna be used for something larger than an iPhone chip for a long time (and mostly likely only the iPhone Pro margins can offset the insane cost of N3 silicon for the first 1-2 years).

View attachment 2124300

What is this chart showing? Is it manufacturing cost/die at constant die size (and thus increased transitior count as the process gets smaller)? Or is it manufacturing cost/die at constant transistor count?

 

[leman]

What is this chart showing? Is it manufacturing cost/die at constant die size (and thus increased transitior count as the process gets smaller)? Or is it manufacturing cost/die at constant transistor count?

It's manufacturing cost of the silicon wafer, which is fixed size.

 

[theorist9]

A Macbook Ultra.. I like it! Very niché, but a lot of production people would probably say it changed their lives.

17", or even 18". A bit chonkier. Whatever, the right tool for the job etc.

I was going to say that you could get a sense for whether production people would buy an Ultra laptop by seeing what PC-using production people are buying today, since equivalent large workstation laptops are available in the PC world.

But then, in checking, I found they're really not. Yes, you can get Ultra-level RAM (128 GB), but you can't get Ultra-level multi-core performance; all the PC workstations I checked were limited to standard mobile core counts.

 

[theorist9]

It's manufacturing cost of the silicon wafer, which is fixed size.

Is it showing the manufacturing cost of the blank wafer alone, where the cost goes up with decreasing process size because of increased purity requirements, or is it the total per-wafer cost of making the finished chips?

 

[jdb8167]

Is it showing the manufacturing cost of the blank wafer alone, where the cost goes up with decreasing process size because of increased purity requirements, or is it the total per-wafer cost of making the finished chips?

You don't buy individual chips from a fab, you buy complete wafers. The number of defects isn't part of the equation. You calculate costs per chip based on die size and yield. If your die size is smaller, generally your yield goes up because you have more chips without defects.

 

[theorist9]

You don't buy individual chips from a fab, you buy complete wafers. The number of defects isn't part of the equation. You calculate costs per chip based on die size and yield. If your die size is smaller, generally your yield goes up because you have more chips without defects.

That's all good and well, but it doesn't answer my question--is the cost per wafer shown in the chart the cost per blank wafer, or the cost for fabricating all the chips that can be made from a wafer?

 

[jdb8167]

That's all good and well, but it doesn't answer my question--is the cost per wafer shown in the chart the cost per blank wafer, or the cost for fabricating all the chips that can be made from a wafer?

It can't be the second because as I explained, the cost of the chips is completely dependent on die size and yield. The first one makes no sense. Who cares what it costs for a blank piece of silicon and what does that have to do with node size? Obviously, given that you buy completed wafers from the fab, it is the cost for a completed wafer at a particular node—in this case TSMC N3.

 

[deconstruct60]

That's all good and well, but it doesn't answer my question--is the cost per wafer shown in the chart the cost per blank wafer, or the cost for fabricating all the chips that can be made from a wafer?

It is cost for a 'printed' wafer. The wafer buying customer has to 'eat' the defects. The fabrication seller isn't offering individual dies on warrantee.

Now if shift to counting costs of the wafer buyer. Defects pragmatically impact how much each working effectively die costs. The working dies have to collectively pay for the defective ones. Also have to pay to have it sliced , tested , and sorted. And then packaged. TSMC may or may not slice, test , and/or package the dies.

 

[theorist9]

It can't be the second because as I explained, the cost of the chips is completely dependent on die size and yield. The first one makes no sense. Who cares what it costs for a blank piece of silicon and what does that have to do with node size? Obviously, given that you buy completed wafers from the fab, it is the cost for a completed wafer at a particular node—in this case TSMC N3.

Yeah, you seriously don't understand my question.

As to your question: The smaller the node size, the higher the purity and defect requirements for the the doped silicon wafers used for chip fabrication.

This doped material is a semiconductor, which means it is switchable from a conducting to a non-conducting state (the doping improves the switching characteristics over those of pure silicion, for chip applications). All wafers have impurities and defects. If the size scale of those impurities/defects is small relative to the feature size, then the conducting/non-conducting pathways won't be significantly impeded (imagine a 20' diameter rock dropped into a river that's several hundred feet wide). But if the impurity or defect is large relative to the feature size (imagine that same rock dropped into a stream that's 10' wide), it could entirely disrupt a pathway, leading to a bad chip. Thus as feature sizes decrease, purity and defect requirements become more stringent, and the cost of blank wafers goes up.

 

[theorist9]

It is cost for a 'printed' wafer. The wafer buying customer has to 'eat' the defects. The fabrication seller isn't offering individual dies on warrantee.

Now if shift to counting costs of the wafer buyer. Defects pragmatically impact how much each working effectively die costs. The working dies have to collectively pay for the defective ones. Also have to pay to have it sliced , tested , and sorted. And then packaged. TSMC may or may not slice, test , and/or package the dies.

OK, then if we know the relative defect rates and densities of chips fabricated with the difference processes, then we can translate these costs into cost/transistor, and resolve the discussion we were having about whether the cost/transistor increases or decreases as you go to smaller processes. Do we have those figures?

 

[deconstruct60]

Exactly, plus they don't realize how expensive is N3 silicon is from TSMC.
It's not gonna be used for something larger than an iPhone chip for a long time (and mostly likely only the iPhone Pro margins can offset the insane cost of N3 silicon for the first 1-2 years).

It isn't going to be completely constrained just to iphone dies. If someone can get $1,500 per die and 40 working dies out of a $20,000 wafer than they can easily pay for the wafer ( $60,000 > $20,000 ).

What not going to get soon is mainstream , affordable larger PC dies that sell for $150-600 out of N3.

Apple isn't selling the Ultra for super discount prices ( ~ $2K) . An Extreme would probably have an even higher mark up ( probably around ~ $4K ). The Max is not super affordable either, but may not make the cut.

There are two major inputs to the cost equation. Number of working dies that can earn money out of a wafer. How much charging for those working dies. Can go small to generate more working dies. Or can charge lots of money on a smaller number of dies. Not solely limited to just small die.

That said don't want to make the 'larger' die gratuitously large. If there are subsections of the die that scale really badly on N3 then probably wasting money including them onto a N3 die. If packaging up heterogeneous chiplets costs less than the space 'wasting' on N3, then can make the 'bigger than an iphone die' just large enough and keep stuff that doesn't shrink on cheaper fab process. But will have to deliver some kind of value add to justify the higher prices.

If Apple has P core complexes , E core complexes , and NPU/GPU core complexes in a N3 then not much problem to composing a bigger clusters of those complexes. Done most of the work anyway.

Making a smaller , more effective chiplet for an Ultra , Extreme (and 'desktop' Max class package) could make financial sense. The Ultra and extreme don't have to be single monolithic dies in package. Neither does a "desktop Max class" if not constrained to laptop price , power targets, and package size.

 

[sunny5]

The known technical limit would be 240W USB PD charger.

Top-end Mac Studio M1 Ultra 5nm has a max power input of 215W without a display.

A M2 Ultra 3nm with at least a 20% performance per watt improvement would allow an Ultra in a MacBook Pro with 240W charger.

The question would be how big of a market is there for laptop that sells for nearly $7,000 and weighs nearly 2x 4.7 pounds (2.1 kg).

Mac Studio M1 Ultra's heat sink fan was made of a more thermally conductive material that weighs more so it can maintain its heat sink fan's RPM and physical form factor dimension.

As the M1 Ultra chip is 2x the die/surface area of a M1 Max chip then the heat it generates would be 2x as well. So that Macbook Pro M2 Ultra 3nm would run hotter than the top-end Macbook Pro 16 M1 Max.

1. The charger is not a problem. There are already 240W cables and making 240W power is not hard. The only problem is which device need that such power and therefore, there aren't any for now. If Apple needs it, then it's easy to make.

2. M1 Ultra's max power consumption for CPU and GPU are 60W and 120W each which isn't really high compared to high end laptops. A mobile RTX 4090 itself consumes 150W or more.

3. A high end laptop is already common in PC for several decades so it's not something new. First of all, laptops are the major revenue source for all computer companies including Apple. Several years ago, Apple had an interview that more than 80% of total sales were laptops. Apple is the only one who is not making high end laptops.

4. The battery performance will be reduced but who cares when people are using laptops while plugging the power cable to use? Even now, it's totally common even for MacBook series. Nobody works with laptops without a power cable especially if they are working seriously.

5. The size and weight will be massive but it's also common in PC. Nothing new.

6. M1,2 Ultra chip itself might be big but that's because memory chips are taking a lot of space. Compared to high end laptops, it's nothing. Just make it bigger.

7. The only concern is how about other Macs with M1,2 Ultra? Even Mac Pro might gonna get M2 Ultra to start with so it might hurt other Macs and Apple doesn't wanna cannibalize their products.

 

[Tyler O'Bannon]

I don’t think we will see an Ultra in a MBP. I think as process nm’s shrink, apple will continue to make about the same physically sized chips, that draw about the same amount of power, and require the same amount of cooling, pushing lots more power as the result. I think the class of chips we’ve seen so far will remain the same for the foreseeable future. Maybe they’ll introduce more variants, but I don’t think we will see and Ultra in a laptop. And hopefully we will see some insanely killer silicon in the Mac
Pro.