This is really difficult to speculate about since the implication are unclear. For example, I don’t really know where the neural engine is used. Apple-internal signal processing and possibly image classification, sure, but beyond that? Apple currently has three options for machine learning: the NPU (hidden behind limited API), the CPU AMX units (hidden behind a more general purpose API) and the GPU - they offer intrinsics to do matrix multiplication faster, but don’t accelerate data types commonly used in ML. It’s all quite confusing. Where Nvidia and AMD seem to build ML-relevant hardware into their GPUs, it seems that Apple is pushing the AMX units instead.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Waiting for the M1 16" MacBook Pro Thread
- Thread starter DHagan4755
- Start date
- Sort by reaction score
This is really difficult to speculate about since the implication are unclear. For example, I don’t really know where the neural engine is used. Apple-internal signal processing and possibly image classification, sure, but beyond that? Apple currently has three options for machine learning: the NPU (hidden behind limited API), the CPU AMX units (hidden behind a more general purpose API) and the GPU - they offer intrinsics to do matrix multiplication faster, but don’t accelerate data types commonly used in ML. It’s all quite confusing. Where Nvidia and AMD seem to build ML-relevant hardware into their GPUs, it seems that Apple is pushing the AMX units instead.
I would expect that Apple was working on the SoC for the 16" M1 MacBook Pro simultaneously with the first M1. It is also why I expect it to have the same cores for both the CPU and GPU as the M1 but to have much more memory bandwidth and many more cores. The M1 is pretty conservative and was probably meant to verify the overall design in production. You can assume that any custom SoC is at least a few years in development.
That is my point. The equation now contains more variables than CPU and GPU.
Good point about usage. When is ML used for normal people except for image processing? I believe that Apple showed that an photograph taken on an iPhone cycled between the CPU, GPU, NE, and back the GPU before it was displayed. They argued that is was to use whatever is most efficient for w given sub task . Apart from that, I also have no idea.
I am quite confident that Apple puts something that is not only powerful but also interesting in the next 16 inch.
This all sounds quite plausible, and I agree that a new 16" MBP will probably not make it for WWDC because it will need to be substantially different (faster memory, maybe faster CPU/GPU cores) to the M1, so it won't be just a case of "scaling them up".
In any case, I am more interested in a new MBP14 for portability (& cost) reasons, and it will be very interesting to see what the CPU/GPU improvements are. I would hope for:
8+4 CPU cores (based on existing M1 core-design)
16 core GPU that would match current AMD 5600M performance.
32GB RAM option.
If the MBP14 doesn't reach these specs, then I would probably wait it out until next year's M2 offerings, and continue with my (still quite new) i9 MBP16.
Yes, I agree that Apple will have already planned for larger SoCs with more CPU & GPU cores. However, I would expect these to be based on the existing M1 designs, i.e. same cores, memory, neural engine etc.
This would be good for a new 14" MBP, which might be just a scaled-up version of the current M1 MBP running an "M1X".
I suspect (but don't know) that the 16" MBP will need to be significantly different, possibly with the GPU on a separate silicon die (on the same SoC) and with faster memory (DDR5?). This makes me think that the MBP16 could be using the next generation of Apple Silicon (M2, based on A15 designs), and might arrive towards the end of the year.
A much larger die is possible. The die size on the M1 is only around 120 mm². Nvidia for example has dies for data center GPUs up to 850mm² on 7nm. Obviously that is much too large for a consumer product but it will give an idea for the upper bounds of what is possible.
Yes, there is certainly "room for growth". The XBox X APU, which is similar in concept to the M1, has a die size of c. 360mm2, and obviously has a pretty powerful GPU and 8 CPU cores. Xbox Series X: A Closer Look at the Technology Powering the ...https://news.xbox.com › en-us › 2020/03/16 › xbox-serie...
My point was that this much larger die might use a different or upgraded design that goes further than just scaling up the same M1 building blocks. I don't know of course, but it's a thought.
Apple does have a track record with the A-series chips of building scaled-out chips with X or Z suffixes, before launching a new design (often with smaller transistors). So it wouldn't be suprising to see an M1X chip with more cores, and thenaan M2 at the end of the year. If they did release an MBP16 at WWDC, then I would expect it to be an M1X...but with an M2 coming out within about 6 months.
It's exciting to watch these rapid advances again, but it makes potential buyers nervous about FOMO.
Don't the X & Z SoCs use the same CPU and GPU cores as their iPhone counterparts though? They are clocked slightly higher and have more memory bandwidth and more cores just as I assume a next generation Apple Silicon SoC will have.
Waiting on the 16". I need CPU and portability. I doubt it will take longer than June for the announcement.
My guess is the wait on the 14", for a couple reasons, not unnecessarily competing with the 13.3" models out now, and that it would actually make sense to release a 16" Air and and Pro at the same time. Remember the 16" is not that much bigger than the 15" it replaced.
So my Guess is June (which I hate, because I would like to see it out by March).
16" MBP
CPU 4-8 and 4-12 configurations
RAM 16 - 32
Graphics 16 - 24
Air
CPU 4-4, 4-8
RAM 8-16
Graphics 8- 16
My guess is the wait on the 14", for a couple reasons, not unnecessarily competing with the 13.3" models out now, and that it would actually make sense to release a 16" Air and and Pro at the same time. Remember the 16" is not that much bigger than the 15" it replaced.
So my Guess is June (which I hate, because I would like to see it out by March).
16" MBP
CPU 4-8 and 4-12 configurations
RAM 16 - 32
Graphics 16 - 24
Air
CPU 4-4, 4-8
RAM 8-16
Graphics 8- 16
The 16" Pro is Apple's flagship MacBook (arguably it's flagship Mac given 80% of Mac sales are notebook form factor) and is about 80 days away from going it's longest between refreshes, so surely they've got to do something with it in the first half of this year? The MacBook Pro has tended towards summer refreshes, so if not WWDC then a May or July MacBook Pro digital event, perhaps. I just can't see them leaving it a full 2 years until October or November between refreshes. I guess a March Intel chip bump with an M2X debut at year's end is possible, though probably won't do much for sales in the meantime over just leaving it.
Then, he is in for a rude awakening when they come out and they’re so much better than the one he just bought. Perhaps you should tell him about it
The only thing about upgrading that concerns me is I do enjoy playing the occasional game on my 16" MBP, and the GPU inside of it handles most games well. Perhaps I need to accept that moving forward, I will need a PC to play games or maybe the Apple GPU will run them just as well through Rosetta.
I don't anticipate many game developers will spend time and resources on porting games to the M1.
I don't anticipate many game developers will spend time and resources on porting games to the M1.
The M1’s GPU is faster than the 580X in my 15” 2018 - so from a performance perspective you should be fine.
I personally can’t wait to dump my 2018 15”. The Intel chip runs excessively hot, which in turn has killed my battery (has read service soon since last year). The M1 machines have tempted me, but I need the extra RAM for my work, and I’ve gotten used to the bigger screen.
I have the 5300M in my 16", which I think is much faster than the 580x. Either way, I will wait for benchmarks. I know some games, like Blizzard ones, probably will never be updated for the M1, and I know I won't have the ability to install Windows unless PCs also move towards ARM architecture in the future.
I know that Apple notebooks don’t play games very well so I can come to the conclusion that if I want games I would get the Xbox series X and be done with it. ?
I see your point. Can’t imagine Apple would replace the 5300M with an inferior GPU, however, especially since there isn’t an eGPU solution for AS Macs (yet).
In typical MS fashion they've had an Arm version os Windows forever. It's a Surface, I don't think you can buy it and install it on any machine yet, it runs on a generic Qualcom chip and runs at tablet speeds. Nothing runs on it yet, but the talk is of a new 64 bit only version of Windows and Microsoft designing their own chips. The upside is an Apple Silicon Windows port seems likely at some point, but not before MS do something about lagging behind so severely. Which could mean never. ?
Yes, the X and Z SoCs are scaled-up versions of the base chips, which is why I think an M1X with the same underlying core design (but more of them) will be in the next release. My point was that this may not be sufficient for the MBP16, which could require far more GPU cores, and possibly newer, faster memory. These changes may not be possible by just extending the existing M1 architecture - e.g. a 32-core GPU might be easier to fabricate on a separate silicon die, or they could be cooling benefits to spreading out the components. Bear in mind that the XBox X SoC has a TDP of 180-200W - far more than would be possible in a laptop computer.
I am really curious how Apple will solve this technical design problem. Maybe I'm overthinking it, and Apple can just double the CPU and GPU cores and surpass current MBP16 CPU/dGPU performance?
In any case, if we assume that there will an A15 and M2(?) with architectural improvements some time after October, it may be preferable to wait until then, even if there is an interim M1X released mid-year. Not knowing Apple's roadmap can be frustrating!
Certainly they can beat the 16" CPU performance with a relatively minor increase in the number of CPU cores since they are almost already there with the M1. Doubling the number of performance cores to 8 would certainly beat any Intel notebook. Beating the GPU on the 16" is more problematic. Just doubling the number of GPU cores is only going to just about match the current top of the line 16" with the 5600M (w/ 8GB HBM2). I don't think Apple is going to want to only match the notebook that they released in 2019 in 2021. So I would expect either a new design or 24 to 32 M1 GPU cores. That is going to need much more memory bandwidth and a lot of die space. Should be interesting to see what Apple has planned.
If they achieve the same performance as the high end intel/AMD MBP with twice the battery life, I think many will be satisfied.
Why We Need Big Chips for Deep Learning | Cerebras
Neural network chips are used for AI and deep learning. Cerebras has created a Wafer-Scale Engine chip that will revolutionize the deep learning process.
cerebras.net
From my understanding it’s more difficult to transfer a game between apis than architectures. Hence why most Mac game ports are in OpenGL instead of Metal.
Is that monstrosity shipping? Or is it more of a thought experiment? Anyway, I'm pretty sure a 46225 mm² won't fit in a 16" MacBook Pro. ?View attachment 1725623
Why We Need Big Chips for Deep Learning | Cerebras
Neural network chips are used for AI and deep learning. Cerebras has created a Wafer-Scale Engine chip that will revolutionize the deep learning process.