Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Made In Machines

macrumors member
Original poster
Jul 30, 2015
75
73
England
Are ALL M1 Pro chips binned versions of the Pro Max (including the 10 core CPU and 16 core GPU M1 Pro)?

Or is the Pro Max a different processor so not all M1 Pros are binned? In that case the:

24 core GPU Pro Max is a binned 32 core GPU Pro Max
8 Core CPU 14 Core GPU is a binned version of the 10 Core CPU 16 core GPU.
10 core CPU and 14 core GPU is also a binned version of the 10 Core CPU 16 core GPU.

I'm not quite sure how it works. Would they cut an M1 Pro Max down to the M1 Pro size?
m1-pro-vs-m1-max-differences-2.jpg
 
I'd say that if they are 'binning', it's unlikely that a Pro would be a Max that failed testing. The die sizes aren't close, and assuming that it's the 'bottom' of the die that failed, slicing it off to make it smaller would risk the entire die failing. I believe that the dies are tested prior to slicing, and the dies that fail are marked, or logged, and disposed of.

It's possible that all the Pro/Max dies are fully capable, and through microcode, parts of the die are disabled. At the moment that the die is cut from the wafer, the cost is minimal. It's a fancy printed disc. After the dies are cut and packaging starts is when the cost of the die adds up. To have multiple runs of different die masks is possible, but it adds to the expense of manufacturing the wafers. They might be doing it, but it would be easier to just disable parts of the die, or use a laser to sever connections than have to reset the fab machines for the different level of chips offered.

But who knows. They might have separate fab machines delivering the different versions of the chips. It would be interesting to see the raw wafers, and cut dies to compare the different versions of say the Pro chips. Are there empty areas where the other GPU cores should be? *shrug* It would be interesting to hack the microcode of a M1 and see what's inside the thing.
 
  • Like
Reactions: T'hain Esh Kelch
The only binning I know of is for the entry level M1 Air. (the model I have) From what I've read, it's just a binned version of the higher specced model that's missing one GPU core.
 
Nope, the Pro is simply a seperate line. If you disable the GPU cores, there are still structural differences (such as more encoders and more bandwidth).

The binned version is the 8-core Pro. The 24-core Max is most likely also a binned version.
 
M1 Pro (binned CPU & GPU) - 8-core CPU (6P/2E) / 14-core GPU
M1 Pro (binned GPU) - 10-core CPU (8P/2E) / 14-core GPU
M1 Pro (full Pro die) - 10-core CPU (8P/2E) / 16-core GPU

M1 Max (binned GPU) - 10-core CPU (8P/2E) / 24-core GPU
M1 Max (full Max die) - 10-core CPU (8P/2E) / 32-core GPU
 
  • Like
Reactions: midkay
I'm not quite sure how it works. Would they cut an M1 Pro Max down to the M1 Pro size?View attachment 1934714


There is no practical means to "cut down" dies with implemented nm-ish sized elements into "smaller chips". There is some rumors guys with some hand waving videos saying this is what Apple is doing but those a just misinformed. The Pro and Max share some die mask elements ( the top half) but there are two different masks used to construct the two dies above.

There is a communication network that goes from 'top' to 'bottom' halve of the Max die and you can't just cleve that off. Where there would have been a transition point to a 'missing' bottom half on the Pro mask there are some limited tweaks to seal/close off paths to areas that don't exist on the Pro version. Relatively speaking it is an incrementally small R&D cost, but it isn't zero.

dies are cut from wafers but the nominal 'scibe width' is far closer to mm than it is to nm . The nominal setting on a common web die yield calcuclator is 0.2mm



This is a close of an Intel Gen 11 (tiger lake ) wafer


6-TheDie2_575px.jpg




The scribe lanes are picking up color here in the context were the photo was taken but zoomed out enough to get the scope of the scribe lane details you loose detail of the die elements.

Wafers are cut to create usable individual dies but the 'scribe' lines used are much, much larger than the size of the intradie connections and features. Cutting is going to cause limited damage and need a buffer were the damage doesn't matter.
 
  • Like
Reactions: Basic75
Are ALL M1 Pro chips binned versions of the Pro Max (including the 10 core CPU and 16 core GPU M1 Pro)?

Or is the Pro Max a different processor so not all M1 Pros are binned? In that case the:

24 core GPU Pro Max is a binned 32 core GPU Pro Max
8 Core CPU 14 Core GPU is a binned version of the 10 Core CPU 16 core GPU.
10 core CPU and 14 core GPU is also a binned version of the 10 Core CPU 16 core GPU.

I'm not quite sure how it works. Would they cut an M1 Pro Max down to the M1 Pro size?View attachment 1934714
The M1 Pro is a M1 Max with half of the GPU cores and 16 NE cores chopped off. It isn't clear what Apple is doing with the leftover GPU cores. They may end up in a future Apple product. The layout of the silicon is suspicious suggesting the cores could function without the processor and be integrated in another product. We still don't know what Apple's discrete GPU codenamed Lifuka is. It is possible it is some sort of interface for these chopped GPUs. I would guess that Lifuka would be for the Mac Pros for really high end workflows or possibly a return of eGPUs for lower-end Macs to support multiple monitors. There could also be an M1 Ultra with 32 extra GPU cores stacked on its interposer in an upcoming device. Apple may be building some M1 Pros pre-chopped to explain this, but that doesn't appear to be the case since some of the interconnects seem to be severed in chips in the wild. It wouldn't make sense to include the interconnects if it were pre-chopped. Some of these chips might simply be near the edge of the platter and therefore chopped.

Those NE cores that are chopped off on the M1 Pro are the mystery cores that Apple disabled in software and doesn't advertise that they exist. It is possible they will be activated in a future version of macOS and reserved for hardware accelerated ray tracing since they seem to be scaling with number of GPU cores.

The other variations are all using binning on full size silicon.
 
Last edited:
The M1 Pro is a M1 Max with half of the GPU cores and 16 NE cores chopped off.

There are ZERO cores being literally chopped off. The Pro and Max have two different die masks. The Pro dies are just printed smaller from the start. There is nothing extra to 'chop' off.

Apple photoshopped the Max die in the initial pictures. The dies used for the Max and Ultra appear to be the same in Apple's non self doctored photos.

Apple just has three die masks being used in volume. One for the 'plain' M1, one for the Pro , and one for the rest .

It isn't clear what Apple is doing with the leftover GPU cores. They may end up in a future Apple product. The layout of the silicon is suspicious suggesting the cores could function without the processor and be integrated in another product. We still don't know what Apple's discrete GPU codenamed Lifuka is. It is possible it is some sort of interface for these chopped GPUs. I would guess that Lifuka would be for the Mac Pros for really high end workflows or possibly a return of eGPUs for lower-

Lifuka is an island that is part of the Tonga island chain. What Apple was tagging with Lifuka isn't necessarily a dGPU. That code name could have just as easily be for the part of a whole.

Could Apple have another "other half" mask about M1 Pro sized that only has GPU cores , memory controllers , NPU , and A/V en/decode along with the UltraFusion connector? That probably would not be too much harder work than constructing the "Pro" half mask. But if it was required to be connected via a UltraFusion link then a dGPU is really a bit of a seriously flawed description. It isn't really all that discrete if it is pragmatically just a soldered on chiplet. It is it's own die, but really not all that independent in where can put it.


The other variations are all using binning on full size silicon.

'Binning' and physically cutting wafers into dies are two different activities.
 
No, there is no evidence that even one pro is a binned pro max (though there may be some out there). These are two different chips.
Is there a way to find out? Like a CPU ID that can be retrieved or something? Just curious.
 
Is there a way to find out? Like a CPU ID that can be retrieved or something? Just curious.

the interrupt handing scope is different on Pro and Max. That could be fused out differences , but the die pictures basically say so no. There is no 'scribe' cutting line between the two halves. Without that it isn't pragmatically to literally physically chop off the other 'half' fo the die.
 
  • Like
Reactions: bobcomer
Are ALL M1 Pro chips binned versions of the Pro Max (including the 10 core CPU and 16 core GPU M1 Pro)?

I'm not quite sure how it works. Would they cut an M1 Pro Max down to the M1 Pro size?

The M1 Pro SoC is its own product, it is NOT a cut-down M1 Max die...

M1 Pro (binned CPU & GPU) - 8-core CPU (6P/2E) / 14-core GPU
M1 Pro (binned GPU) - 10-core CPU (8P/2E) / 14-core GPU
M1 Pro (full Pro die) - 10-core CPU (8P/2E) / 16-core GPU

M1 Max (binned GPU) - 10-core CPU (8P/2E) / 24-core GPU
M1 Max (full Max die) - 10-core CPU (8P/2E) / 32-core GPU

Apple has three M1 SoC products; M1, M1 Pro, & M1 Max...

These three SoCs have binned variants, as outlined above...

Apple uses two M1 Max SoCs (both GPU binned & full die variants) to make the M1 Ultra, it is not one monolithic die...

The M1 Pro is a M1 Max with half of the GPU cores and 16 NE cores chopped off.

No, it is not...

There are ZERO cores being literally chopped off. The Pro and Max have two different die masks. The Pro dies are just printed smaller from the start. There is nothing extra to 'chop' off.

^^^ This, all of this; the M1 Pro SoC is NOT a cut-down M1 Max die... ^^^
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.