4 ms·

I wonder what kind of usage metrics Apple has that led it to reduce (8 down to 6) the number of performance cores and increase (4 up to 6) the number of efficie

I wonder what kind of usage metrics Apple has that led it to reduce (8 down to 6) the number of performance cores and increase (4 up to 6) the number of efficiency cores on the M3 Pro? Or the reduction (25%) in memory bandwidth? Maybe they're trying to get people to buy the Max version? The spec differences are interesting: M3 Pro, 6 efficiency, 6 performance, 18 GPU, 16 Neural cores, 150GB/s memory bandwidth M3 Max 1, 4 efficiency, 10 performance, 30 GPU, 16 Neural cores, 300GB/s memory bandwidth M3 Max 2, 4 efficiency, 12 performance, 40 GPU, 16 Neural cores, 400GB/s memory bandwidth They reduced the number of efficiency cores on the more powerful models. I'm not sure that makes sense to me. Edited to add: Why do the number of neural cores (16) stay the same across all 3 CPU levels?

JansjoFromIkea 3y ago
Way outside of my field of knowledge but would it be around prioritizing battery life above all else on the pro? Would also nudge people who _need_ the higher power features to not stick with Pro too, obviously.
- 3c6bYDXLMj 3y ago
  [dead]
- stetrain 3y ago
  That would make sense, but they also didn't break out battery life ratings between the Pro and Max chips, both are listed as 18 hours. I suspect the Pro will have longer battery life in real world testing but we haven't seen that tested yet.
- s3p 3y ago
  That confuses me because shouldn't those people just get an M3? and then those who get the M3 Pro want better performance. I don't understand the 'pro' naming scheme means if they're not serious about making it pro.
  - dagmx 3y ago
    What is Pro to you? I ask, because it can be a huge range of professional tasks and many won’t need the higher performance of the Max, but will benefit from the efficiency of the Pro. The Pro is still a significant jump over the base M3, both in performance and capability across every metric. It just happens to be a mid range SKU and for some reason people seem to think in binary high/low terms. I’d bet it’s the most common of the SKUs sold.
  - JansjoFromIkea 3y ago
    Isn't it basically the same reason loads of things will have sizing options along the lines of medium/large/massive instead of small/medium/large. At this stage I'd say "Pro" in terms of Macbooks is is more likely to be for people who want something more powerful or a bit fancier than an Air than people who need something more powerful, so there needs to be another level above that
  - rsynnott 3y ago
    "Why don't people get either a Celeron or an i9?" Every CPU lineup has an intermediate; Apple actually has fewer than most, and I'd expect they'll probably add one or two more SKUs as time passes to deal with binning more efficiently.
api 3y ago
Reducing memory bandwidth is particularly bad given that most AI work loads are memory bound. With the memory bandwidth Apple's fully integrated GPU with shared memory is actually pretty good for smaller AI stuff like running and fine tuning llama2-sized models. It gives you a decently fast GPU option with a lot of RAM. NVIDIA cards with similar RAM are a decent fraction of the price of a whole Pro or Max M chip laptop.
- bloopernova 3y ago
  > most AI work loads are memory bound. Is that the reason why the number of neural cores stays the same, do you think?
  - api 3y ago
    The neural engine in Apple chips is really for small models for doing things like Face ID, local inference for Photos.app, etc. It's there to run those small models with very little power consumption. It's not that great for huge models. The GPU is better for those, and they have been adding cores and performance there. But yes, most large models are pretty heavily memory bound.
    - Tagbert 3y ago
      Memory throughput or memory capacity? Is that true for discrete GPUs like Nvidia? Is it the same for unified memory like the M-series?
      - api 3y ago
        Both really, since models are pretty much 100% "hot" RAM. The bigger the model the more bandwidth it demands too since evaluating it means streaming the entire thing through the processor.
alexpenny 3y ago
I think differentiating the pro and max is a big part of it, but it also seems like they're trying to spread out the benefits of the switch to 3nm over time. I would expect performance cores and bandwidth to go back up next year to maintain steady performance increases year over year.

JansjoFromIkea 3y ago

Way outside of my field of knowledge but would it be around prioritizing battery life above all else on the pro? Would also nudge people who _need_ the higher power features to not stick with Pro too, obviously.

3c6bYDXLMj 3y ago
[dead]
stetrain 3y ago
That would make sense, but they also didn't break out battery life ratings between the Pro and Max chips, both are listed as 18 hours. I suspect the Pro will have longer battery life in real world testing but we haven't seen that tested yet.
s3p 3y ago
That confuses me because shouldn't those people just get an M3? and then those who get the M3 Pro want better performance. I don't understand the 'pro' naming scheme means if they're not serious about making it pro.
- dagmx 3y ago
  What is Pro to you? I ask, because it can be a huge range of professional tasks and many won’t need the higher performance of the Max, but will benefit from the efficiency of the Pro. The Pro is still a significant jump over the base M3, both in performance and capability across every metric. It just happens to be a mid range SKU and for some reason people seem to think in binary high/low terms. I’d bet it’s the most common of the SKUs sold.
- JansjoFromIkea 3y ago
  Isn't it basically the same reason loads of things will have sizing options along the lines of medium/large/massive instead of small/medium/large. At this stage I'd say "Pro" in terms of Macbooks is is more likely to be for people who want something more powerful or a bit fancier than an Air than people who need something more powerful, so there needs to be another level above that
- rsynnott 3y ago
  "Why don't people get either a Celeron or an i9?" Every CPU lineup has an intermediate; Apple actually has fewer than most, and I'd expect they'll probably add one or two more SKUs as time passes to deal with binning more efficiently.

api 3y ago

Reducing memory bandwidth is particularly bad given that most AI work loads are memory bound. With the memory bandwidth Apple's fully integrated GPU with shared memory is actually pretty good for smaller AI stuff like running and fine tuning llama2-sized models. It gives you a decently fast GPU option with a lot of RAM. NVIDIA cards with similar RAM are a decent fraction of the price of a whole Pro or Max M chip laptop.

bloopernova 3y ago
> most AI work loads are memory bound. Is that the reason why the number of neural cores stays the same, do you think?
- api 3y ago
  The neural engine in Apple chips is really for small models for doing things like Face ID, local inference for Photos.app, etc. It's there to run those small models with very little power consumption. It's not that great for huge models. The GPU is better for those, and they have been adding cores and performance there. But yes, most large models are pretty heavily memory bound.
  - Tagbert 3y ago
    Memory throughput or memory capacity? Is that true for discrete GPUs like Nvidia? Is it the same for unified memory like the M-series?
    - api 3y ago
      Both really, since models are pretty much 100% "hot" RAM. The bigger the model the more bandwidth it demands too since evaluating it means streaming the entire thing through the processor.

alexpenny 3y ago

I think differentiating the pro and max is a big part of it, but it also seems like they're trying to spread out the benefits of the switch to 3nm over time. I would expect performance cores and bandwidth to go back up next year to maintain steady performance increases year over year.