Will M1x have a TDP of 35-54W?

[senttoschool]

If they want to make the iMac as thick as an Xdr monitor and use those drilled circles as cooling vents then I guess keeping system power down would be an input.

edit - apparently the xdr uses about 100w in full brightness hdr mode. I don’t have a good sense how much power the 5k panel in the current iMac draws (50w?) so could be scope to up the core count on the m1 chip and still fit within the thermal envelope of the xdr chasssis/cooling system

Functionality, I think it makes no difference to make an iMac that is 1inch thick or 1.5 inch thick. But because this is Apple and they want to wow customers inside Apple retail stores, they will aim to make it 1 inch thick.

Still, I think Apple can comfortably fit a 50 - 100w chip inside a 1inch thick chassis. After all, the Macbook Pro 16" with its GPU and 8-core Intel CPU can draw that much power at its peak with a much smaller chassis.

 

[neinjohn]

This is how rumors are created.... asking questions that get spun by marketing people and create a life of its own.

The i3-1000NG4 on the base 2020 Air has a 9W TDP, in the same ball park as the M1 Air, and its performance and thermal profile is from another planet on the same chassis with active ventilation. The i5-8257U on the base 2020 Pro 13'' has the same TDP (15W) as the M1 Pro. Same thing happening.

You'll also not have the same power and thermal profile on ultrabook with a Ryzen 7 5800U or Tiger Lake U with a 15W TDP if the motherboard was tweaked the same way Apple generally runs their chips (allow it to get hot, don't throttle on battery power).

Maybe TDP isn't the best proxy to what interesting here on the further Macs to be announced.

As far as thermal are of interest and I think it's better to ask how far Apple can clock the Firestorm (and maybe the Icestorm) cores without being highly inefficient on single-burst workloads or even if it'll be different from the M1. Can they clock the graphic cores higher, how many cores of each they'll have, will it clock higher on multi-thread workloads (now TDP is important) and how it will work on mixed workload with CPU and GPU.

 

[Homy]

The 2020 Mac Mini Unleashed: Putting Apple Silicon M1 To The Test

www.anandtech.com

 

[cmaier]

So the iPhone gets an A12, the iPad Pro an A12X or A12Z, the MacBook Air an M1 and the top-of-the line iMac... an M1*?
It doesn't make much sense marketing-wise.

*16-core edition

Well to be fair, not every M1 package is currently identical either. They are all called M1.

 

[Krevnik]

Well to be fair, not every M1 package is currently identical either. They are all called M1.

Hmm, I think I see now, and it does make sense.

iPhone and iPad materials don’t put core counts on the spec sheet (or RAM). So the iPad Pro 2020 just lists “A12Z Bionic”. With the Mac materials, Apple puts the core counts on the spec sheets. We’re told the MacBook Pro has an 8C CPU (4+4) and 8C GPU on the tech specs page. That does suggest Apple sees these audiences slightly differently, and expects the Mac buyers want/expect to know core counts and RAM configuration.

So yeah, I think your sources are probably right. The 16” MBP could very well be “M1 with 12C CPU, 16C GPU” or whatever it actually winds up being. It does simplify naming quite a bit, and it makes sense since Apple can quite literally do whatever they need for their product line. Making up categories to put the SoCs in may just be a waste of time.

That said, “not every M1 package is currently identical” feels like a bit of a reach IMO to my engineer brain. They are all the same die. So while it may be evidence, I don’t think it’s terribly compelling evidence to back up the idea. I think folks were more expecting a different name for different die layouts like we see with the A series.

 

[cmaier]

Hmm, I think I see now, and it does make sense.

iPhone and iPad materials don’t put core counts on the spec sheet (or RAM). So the iPad Pro 2020 just lists “A12Z Bionic”. With the Mac materials, Apple puts the core counts on the spec sheets. We’re told the MacBook Pro has an 8C CPU (4+4) and 8C GPU on the tech specs page. That does suggest Apple sees these audiences slightly differently, and expects the Mac buyers want/expect to know core counts and RAM configuration.

So yeah, I think your sources are probably right. The 16” MBP could very well be “M1 with 12C CPU, 16C GPU” or whatever it actually winds up being. It does simplify naming quite a bit, and it makes sense since Apple can quite literally do whatever they need for their product line. Making up categories to put the SoCs in may just be a waste of time.

That said, “not every M1 package is currently identical” feels like a bit of a reach IMO to my engineer brain. They are all the same die. So while it may be evidence, I don’t think it’s terribly compelling evidence to back up the idea. I think folks were more expecting a different name for different die layouts like we see with the A series.

Well, the package contains multiple die, and the RAM is different. Some M1s have 7 graphics cores and some have 8. (I assume they are all contain 8 and 1 is disabled, but I don’t know that for a fact).

Anyway, obviously every chip has some different internal part number, but there isn’t necessarily a good reason for apple to market them that way. “Buy this air and get the same great technology as in this iMac” is just as good a marketing scheme as any other.

 

[jeanlain]

? So Apple may just refer to the curent architecture as "M1", and will differentiate SoCs according to GPU and CPU cores in spec sheets.
"Choose your 24" iMac. M1 with:
8 CPU cores
12 CPU cores
16 CPU cores
..."
That makes sense.

 

[jdb8167]

Well, the package contains multiple die, and the RAM is different. Some M1s have 7 graphics cores and some have 8. (I assume they are all contain 8 and 1 is disabled, but I don’t know that for a fact).

Anyway, obviously every chip has some different internal part number, but there isn’t necessarily a good reason for apple to market them that way. “Buy this air and get the same great technology as in this iMac” is just as good a marketing scheme as any other.

I wonder if that means Apple expects too many different SoC configurations to give them all separate names. We have 2 right now. Add in another 2 for the hypothetical 14” and 16” MacBook Pros. Then another 2 for the hypothetical 24” and 30” iMacs. Then another few for the hypothetical Mac Pros. As long as each SoC stays in its own product category, calling everything M1 simplifies the marketing. I suspect that means Apple might call out GPU and CPU core counts but will never discuss clocks.

 

[cmaier]

I wonder if that means Apple expects too many different SoC configurations to give them all separate names. We have 2 right now. Add in another 2 for the hypothetical 14” and 16” MacBook Pros. Then another 2 for the hypothetical 24” and 30” iMacs. Then another few for the hypothetical Mac Pros. As long as each SoC stays in its own product category, calling everything M1 simplifies the marketing. I suspect that means Apple might call out GPU and CPU core counts but will never discuss clocks.

I think it’s more to do with “consumers don’t care.” Probably the only reason the ipads have “x” chips is that people assume “phone cpus” are weak. Tell consumers the clock rate and number of cores, and they have the information they need.

I mean, consider the 16” MBP which will presumably come with a beefier processor than the base model 13s already released. presumably the cores are identical, but there are more of them. What’s the point in saying “hey, buy this base model with an M1, or the higher end model with and M1X?” That doesn’t tell the user anything. “X” doesn’t even necessarily come across as “better.”

 

[jdb8167]

I think it’s more to do with “consumers don’t care.” Probably the only reason the ipads have “x” chips is that people assume “phone cpus” are weak. Tell consumers the clock rate and number of cores, and they have the information they need.

I mean, consider the 16” MBP which will presumably come with a beefier processor than the base model 13s already released. presumably the cores are identical, but there are more of them. What’s the point in saying “hey, buy this base model with an M1, or the higher end model with and M1X?” That doesn’t tell the user anything. “X” doesn’t even necessarily come across as “better.”

I also expect the 16” to be clocked a bit higher but Apple doesn’t seem to ever talk about clock speed. As long as the increased clocks are tied to specific products it doesn’t matter though. In their marketing they can say 2.5x faster etc.

 

[UltimateSyn]

I wonder if that means Apple expects too many different SoC configurations to give them all separate names. We have 2 right now. Add in another 2 for the hypothetical 14” and 16” MacBook Pros. Then another 2 for the hypothetical 24” and 30” iMacs. Then another few for the hypothetical Mac Pros. As long as each SoC stays in its own product category, calling everything M1 simplifies the marketing. I suspect that means Apple might call out GPU and CPU core counts but will never discuss clocks.

I think this is a great reason. You’re right, they’re not going to want 19 different M1{Letter} chips for the multitude of different CPU / GPU core-count combinations there will end up being present across the lineup. With the A-Series chips core counts are hidden so they need a marketing name for each variation, and there’s only 1-3 variants in the family at a time - easy. With M1 I guess it seems likely that they will just specify the number of cores that each product gets.

 

[Krevnik]

Well, the package contains multiple die, and the RAM is different. Some M1s have 7 graphics cores and some have 8. (I assume they are all contain 8 and 1 is disabled, but I don’t know that for a fact).

Anyway, obviously every chip has some different internal part number, but there isn’t necessarily a good reason for apple to market them that way. “Buy this air and get the same great technology as in this iMac” is just as good a marketing scheme as any other.

Is there any case where the 7-core being a separate die layout actually buys Apple anything? I would think that disabling a core would be the easier option from a logistics perspective, for multiple reasons.

Also, what are the dies you refer to here? My understanding is that the SoC is a single die and the on-package RAM is itself using a BGA package.

I’m not necessarily trying to attack your statement here, but since we are getting into the weeds a little bit and seems like splitting hairs, I am kinda curious how you define “multiple dies”, since I tend to think more along the lines of Ryzen where the dies are directly on the package, rather than a package with BGA packages soldered onto it.

 

[cmaier]

Is there any case where the 7-core being a separate die layout actually buys Apple anything? I would think that disabling a core would be the easier option from a logistics perspective, for multiple reasons.

Also, what are the dies you refer to here? My understanding is that the SoC is a single die and the on-package RAM is itself using a BGA package.

I’m not necessarily trying to attack your statement here, but since we are getting into the weeds a little bit and seems like splitting hairs, I am kinda curious how you define “multiple dies”, since I tend to think more along the lines of Ryzen where the dies are directly on the package, rather than a package with BGA packages soldered onto it.

My understanding is that the RAM die is in the same package as the SoC and is not in a separate BGA? I could be wrong. [Update: i found a photo that appears to show the package with the lid off, and the RAM in its own packages within, so it looks like you are correct. I find that a very weird way to do it, fwiw.] In any event, my point was simply that these are different ”chips” (in the sense that the overall package has different capabilities) and yet apple doesn’t feel the need to name them different things. I mean, isn’t it the case that the two most recent ipad pros differ only in GPU core count, and one of them gets an x suffix and the other a z? Apple didn’t do anything like that this time.

As for the 7-core situation, I suspect that these are all 8-core with 1 core disabled, but I just don’t know. I’ve worked at companies where some die for a particular product had disabled functions and others were missing the circuits, so it’s not unheard of (you would only do it if you could make the die smaller, or make it function better (e.g. by improving thermals by spreading other circuits into that space) by removing circuits.

 

[Gerdi]

As for the 7-core situation, I suspect that these are all 8-core with 1 core disabled, but I just don’t know. I’ve worked at companies where some die for a particular product had disabled functions and others were missing the circuits, so it’s not unheard of (you would only do it if you could make the die smaller, or make it function better (e.g. by improving thermals by spreading other circuits into that space) by removing circuits.

Thats very common, because the masks costing millions - not to mention the additional verification effort for what is essentially a different chip from DV perspective.

 

[cmaier]

Thats very common, because the masks costing millions - not to mention the additional verification effort for what is essentially a different chip from DV perspective.

The masks and DV aren’t as expensive as you think, particularly for this sort of change. We even had a separate flow just for these sort of things (engineering change orders), where you can freeze various mask layers and allow changes on others. DV for such a thing is also really easy, because you already passed DV, and now you are removing stuff (a lot easier than verifying additions). You’d also be surprised at how ... um... informal the process can be. One time, for an x86-64 chip, on the day of tapeout, I implemented a bug fix (for somebody else’s block) by loading their .def file into vi and manually removing a metal polygon on one layer and adding a separate metal polygon on another layer (due to historical reasons, I knew how to read the file format and generate the variation acceptable to our downstream tools). I thus bypassed the entire front-end of our flow to manually hack the file . After that, I ran the block through LVS, DRC, and skipped our electrical checks (IR drop, parasitics extraction, electromigration, etc.) because I eyeballed and it and guessed it would work. From the time I fired up vi, jotted down on a piece of scrap paper the relevant coordinates, edited the file, etc. until LVS completed was about 35 minutes, which was good because about a half hour later we needed to send the gdsii file to the fab.

 

[iPadified]

As long as it will be higher end versions of M1, the name really does not matter. On the forum we call it M1X as an easy placeholder for any higher end version - no problem with that. If Apple can avoid Intels naming h*ll, I am fine with a generic "M1" plus a definition of GPU/CPU cores when ordering. That will scale to the Mac Pro as well. M2 will be next architecture change.

 

[leman]

Is there any case where the 7-core being a separate die layout actually buys Apple anything? I would think that disabling a core would be the easier option from a logistics perspective, for multiple reasons.

It is almost certainly the same chip but with one GPU core disabled. And it's very likely to be power binning — rather than classical "disable the defective core", Apple might be doing "disable a GPU core in chips with subpar thermal behavior". This would allow them to still sell "lower quality" chips in a MBA.

There is no economical sense in manufacturing a separate variant for the entry-level MBA...

 

[throAU]

This is how rumors are created.... asking questions that get spun by marketing people and create a life of its own.

I suspect there will be an Apple SOC with a TDP around that range, and maybe even higher for the 16" machine.

Processing needs power, and the thermal/power budget for a 16" class chassis is generally in the 50-100 watt range (with a fan and good cooling for the upper end). Apple have a choice of similar processing with less energy, way better processing power in the same power/heat - or a mix of both.

It would not surprise me at all to see a 35-65w SOC with better than Comet Lake performance on Mac applications inside a 16" machine.

It may be called M2, M3 or M1X or something else.

I'm not concerned with what it is called as that's not what is important imho. What IS important is that the above part will exist and will likely ship this year.

 

[leman]

I suspect there will be an Apple SOC with a TDP around that range, and maybe even higher for the 16" machine.

Processing needs power, and the thermal/power budget for a 16" class chassis is generally in the 50-100 watt range (with a fan and good cooling for the upper end). Apple have a choice of similar processing with less energy, way better processing power in the same power/heat - or a mix of both.

It would not surprise me at all to see a 35-65w SOC with better than Comet Lake performance on Mac applications inside a 16" machine.

Knowing Apple, they design their machines around predetermined power brackets. Those power brackets remained fairly constant over the years:

MacBook Air: 10-15 watts
MacBook (Pro) 13": 25-30W (exception is the two-port model which inherits stuff from Air)
MacBook Pro 15": 60-70W (bumped to around 80W for the 16" model)

Similar power brackets also apply to their desktop machines, but they are less clear. For laptops, it's all about the chassis size and the relation of battery to idle power consumption.

I do not think that they will use the Apple Silicon opportunity to tune these brackets down. They could have easily provided competitive performance with say, Tiger Lake, by capping M1 at 10watts total power. But there is little reason to do this. They know that their chassis can dissipate X watts, so that's how they are going to configure the machine. Anything else is very "un-Apple".

So yeah, my prediction is that the TDP of their new machines will roughly match the TDP of the old ones (of course, there is the entire issue of TDP being a problematic concept and that Intel TDP ≠ Apple Silicon TDP, but that's all besides the point). And that again gives us a good idea what performance to expect.

The ~30W 13/14" model will have multi-core performance rivaling that of enthusiast-level desktop Zen3/Rocket Lake, and GPU performance at least on par with current 5500M, while the 16" model will have CPU performance of a large desktop workstation and GPU comparable to current higher-mid-range.

You are welcome to bookmark this post and revisit it in one year or so when the lineup is out

I'm not concerned with what it is called as that's not what is important imho. What IS important is that the above part will exist and will likely ship this year.

Exactly!

 

[iPadified]

Knowing Apple, they design their machines around predetermined power brackets. Those power brackets remained fairly constant over the years:

MacBook Air: 10-15 watts
MacBook (Pro) 13": 25-30W (exception is the two-port model which inherits stuff from Air)
MacBook Pro 15": 60-70W (bumped to around 80W for the 16" model)

Similar power brackets also apply to their desktop machines, but they are less clear. For laptops, it's all about the chassis size and the relation of battery to idle power consumption.

I do not think that they will use the Apple Silicon opportunity to tune these brackets down. They could have easily provided competitive performance with say, Tiger Lake, by capping M1 at 10watts total power. But there is little reason to do this. They know that their chassis can dissipate X watts, so that's how they are going to configure the machine. Anything else is very "un-Apple".

So yeah, my prediction is that the TDP of their new machines will roughly match the TDP of the old ones (of course, there is the entire issue of TDP being a problematic concept and that Intel TDP ≠ Apple Silicon TDP, but that's all besides the point). And that again gives us a good idea what performance to expect.

The ~30W 13/14" model will have multi-core performance rivaling that of enthusiast-level desktop Zen3/Rocket Lake, and GPU performance at least on par with current 5500M, while the 16" model will have CPU performance of a large desktop workstation and GPU comparable to current higher-mid-range.

You are welcome to bookmark this post and revisit it in one year or so when the lineup is out



Exactly!

For the laptops yes, I agree. For the desktop no. Their form factors change very often. We have no clue how much a redesigned iMac will dissipate.

What is it in the current iMac 21.5 inch? About 100W for CPU and GPU? A 100W M1 chip?(!) Yes please, but I doubt it.

35-50W is more reasonable.

 

[Argon_]

The CPU in the 16" is actually configured for around 60 watt sustained. But the combine dTDP of CPU+GPU is just around 80 watts.

Depends on what you mean by the "M1X". If that's the chip that will go into the next 16",it will probably be closer to 60-80 watts total. If you mean the chip that goes into the higher-end 13"/14", it will be 30W at most.

Alternatively, tuning the chip for 40-60W continuous would make for either a silent computer, or a cool, not entirely silent one.

 

[neinjohn]

Alternatively, tuning the chip for 40-60W continuous would make for either a silent computer, or a cool, not entirely silent one.

A i9-9880H on a Pro 16'' has a 45W TDP but can go for 85W average on a Prime95 stress test according to notebookcheck. Intel, and Apple implementation of their CPUs, tried to allow it to Turbo for as long as possible to extract maximum performance. So one has to judge how much impact on thermal build-up inside the laptop those giant wattage peaks have compared to constant but controlled 45W peak consumption on a M1 variant.

 

[Krevnik]

My understanding is that the RAM die is in the same package as the SoC and is not in a separate BGA? I could be wrong. [Update: i found a photo that appears to show the package with the lid off, and the RAM in its own packages within, so it looks like you are correct. I find that a very weird way to do it, fwiw.] In any event, my point was simply that these are different ”chips” (in the sense that the overall package has different capabilities) and yet apple doesn’t feel the need to name them different things.

Yeah, no worries, I just wanted to understand a bit more what you were thinking rather than just attack what you said.

And the A12X and A12Z both do the same thing here. BGA LPDDR soldered onto the package's PCB. It is a bit weird, since I haven't heard of anyone else doing it this way, but it worked for the iPad Pro. The M1 currently in the wild definitely comes across as a customized iPad Pro SoC. Not that it is a bad thing for the entry-level Mac models, honestly.

As for the 7-core situation, I suspect that these are all 8-core with 1 core disabled, but I just don’t know. I’ve worked at companies where some die for a particular product had disabled functions and others were missing the circuits, so it’s not unheard of (you would only do it if you could make the die smaller, or make it function better (e.g. by improving thermals by spreading other circuits into that space) by removing circuits.

Yeah, my thinking here was that even if they had to disable perfectly good GPU cores, it'd still be easier to manage the supply chain if it was a single die layout and adjust to shifting demand between models using the same base die.

But that said, I don't know if that actually buys anything here for sure. I'm not a hardware engineer by trade, and I rarely break out my oscilloscope these days for debugging personal EE projects, maybe once a year. So I was more curious if there was nuance here I was maybe missing on the production side that made using a separate layout beneficial.

It is almost certainly the same chip but with one GPU core disabled. And it's very likely to be power binning — rather than classical "disable the defective core", Apple might be doing "disable a GPU core in chips with subpar thermal behavior". This would allow them to still sell "lower quality" chips in a MBA.

There is no economical sense in manufacturing a separate variant for the entry-level MBA...

I agree, but keep in mind that I was more looking to see if there was more nuance there, knowing cmaier has worked in this industry. And also knowing the details you provided here, already.

 

[ohbrilliance]

A very interesting discussion! I haven't seen it mentioned explicitly, but unlike iOS devices, some of the Macs will come with CPU options for users to choose between. Your only choice of Apple CPUs to-date has been by buying entirely different product lines. Naming like A12 and A12X work well for that, but don't help in a purchasing decision between CPUs for the same computer. Specs (CPU and GPU cores) do and a CPU/SOC name such as M1 with X CPU cores and Y GPU cores is clean and informative.

 

[cmaier]

See? iMac uses "M1," not "M1x" (They haven't said anything about cores or clock, yet, looks like)

Update: 8 cores/8 cores. So may be literally the same M1.

Update 2: And in iPad Pro! "M1" everywhere!