M4+ Chip Generation - Speculation Megathread [MERGED]

[Confused-User]

I've been offline since Thursday- pardon the late responses.

So what do you do? You can continue to ship adequate but disappointing products (like an M3 Ultra, say) or you can engage in some hard work to fix all the known problem areas, accepting that that will take as long as it takes.

Or you can do both- which I would argue is what they should have been doing.

But I think this also means that Apple knows they don't get second chances. If they ship a disappointing Ultra or Extreme they will have to live with that for a decade. Better to delay a year of sales than take that hit.

This does not seem convincing to me. If it's true, then they're already finished for this decade. Their track record for high-end desktops is beyond appalling. Failing to ship M3 Studios, and now leaving M$s up in the air, just makes it all worse. And if there are second (3rd, 7th, whatever) chances... well, an M3/M4 2xMax Ultra could only improve on the M2 Ultra, and then they get another chance after that.

I think the real problem here is that they're looking at the market wrong. In general I think Cook is doing an excellent job, but in this area he or whoever is making the decision has completely failed, over and over, probably due to misplaced fiscal prudence.

Yes, *of course* the market for high-end desktop Macs is tiny. Apple's been crapping on those customers for more than a decade. Naturally most are gone. If Apple is to take share in that market they have to execute reliably *and at least somewhat predictably* for a number of years to overwrite the memory of their contempt for their customers. Of course there's always the chance that exceptional/unmatched performance (as you mentioned, not quoted here) may drive rapid uptake in new markets, but they won't damage those chances by building machines that aren't ready for that market yet since in that respect they'd simply be not a candidate that gets considered. All they really have to do is not fail miserably at managing expectations.

The right thing to do is take the short-term fiscal hit. Eat the cost of the masks, the engineering time and opportunity cost, etc. Meanwhile you grow your team's experience, and more rapidly iterate on your advances. You won't (unless you're very lucky) break even for years. So what? It's chicken scratch to Apple. But it's a good long-term investment. Tech they build now for high-end desktops will in many cases flow down to mainstream products over time.

 

[ader42]

I think maybe Cook has been playing it correct from a business perspective despite it being what would be best for my personal work needs. The M1 to M3 series chips have been focused on low-end to mid-end users as that is the bulk of Apple’s custom. Clearly after the M2 Ultra, Apple realised the “interconnect” for an M3 Ultra was the wrong approach so they changed tack.

Now that the M4 series is here there is going to be a lot more oomph available with the next Mac Studio desktops and hopefully they can start to drive more high end customers. They might even launch the M5 series with the Mac Studios in 2025 with the lower end M5 variants coming afterwards.

From a business perspective this approach likely wasn’t a viable financial option before the M4/M5 - as trying to do this would have left a lot of money on the table (much more than the profit from the high end users).

Many times I’ve been frustrated with Apple, but in the end I have always seen with hindsight that their decisions were the correct ones for the business even if they didn’t align with my personal wants.

 

[Confused-User]

I think maybe Cook has been playing it correct from a business perspective despite it being what would be best for my personal work needs. [...]
From a business perspective this approach likely wasn’t a viable financial option before the M4/M5 - as trying to do this would have left a lot of money on the table (much more than the profit from the high end users).

Many times I’ve been frustrated with Apple, but in the end I have always seen with hindsight that their decisions were the correct ones for the business even if they didn’t align with my personal wants.

I think you're illustrating perfectly the thought process of Cook (or, again, whoever is making this particular decision; we don't know for sure it's him). But again, my point is that this may be a good short-term decision while still being the wrong one from a longer-term perspective.

They have decimated their customer base for high-end gear over the last decade. (Let's say that "high-end" encompasses Mac Pros and Mac Studios, while acknowledging that the cheapest Studios might not be a natural fit for that category.) Building an M3 Ultra while working on something better can in no way be seen as worse than continuing to ship the M2 Ultra, which at this point is a bad joke to the customers walking away from it and a source of extreme discontent for most of the few who still buy it - especially the very few Mac Pro buyers.

My suggestion is that while they might be taking a modest loss on high-end sales in the short term by continuing to release new high-end products at a regular cadence, they would be beginning to turn around the loss of whatever relatively few high-end customers they have left, and eventually start building up that customer base again. So that when they finally do release a brilliant product, whether it be an M4 Ultra or an M6 Extreme, they'll have some credibility. They are sorely lacking that right now. In fact, while as a whole the Mac product line has an enviable reputation, Apple probably has a significant negative brand value specifically in the high-end Mac product category.

 

[MayaUser]

No M Extreme, only M Supreme

 

[quarkysg]

They have decimated their customer base for high-end gear over the last decade.

I think it's the cloud that has decimated high-end gear demand. Much cheaper to rent cloud compute short term IMHO.

 

[MRMSFC]

I think maybe Cook has been playing it correct from a business perspective despite it being what would be best for my personal work needs. The M1 to M3 series chips have been focused on low-end to mid-end users as that is the bulk of Apple’s custom. Clearly after the M2 Ultra, Apple realised the “interconnect” for an M3 Ultra was the wrong approach so they changed tack.

Now that the M4 series is here there is going to be a lot more oomph available with the next Mac Studio desktops and hopefully they can start to drive more high end customers. They might even launch the M5 series with the Mac Studios in 2025 with the lower end M5 variants coming afterwards.

From a business perspective this approach likely wasn’t a viable financial option before the M4/M5 - as trying to do this would have left a lot of money on the table (much more than the profit from the high end users).

Many times I’ve been frustrated with Apple, but in the end I have always seen with hindsight that their decisions were the correct ones for the business even if they didn’t align with my personal wants.

I will go to bat for Tim Cook a lot and personally loathe the “Steve would never have done this!” Crowd.

Here’s where I will be a hypocrite though:
Steve (probably) wouldn’t have done this.

Yes, the decisions are sound from a business standpoint, but that’s not what made Apple big. It’s taking risks and making the best products possible that brought them success.

Even if updating the high end products at a slower cadence is a sound business decision, it leaves many vocal customers feeling like second class citizens.

That’s not how Apple built their past brand loyalty, and I do think it’s not the right thing to do.

 

[MacPoulet]

I will go to bat for Tim Cook a lot and personally loathe the “Steve would never have done this!” Crowd.

Here’s where I will be a hypocrite though:
Steve (probably) wouldn’t have done this.

Yes, the decisions are sound from a business standpoint, but that’s not what made Apple big. It’s taking risks and making the best products possible that brought them success.

Even if updating the high end products at a slower cadence is a sound business decision, it leaves many vocal customers feeling like second class citizens.

That’s not how Apple built their past brand loyalty, and I do think it’s not the right thing to do.

Wasn't Steve also a Mac Pro user? That would be the one thing missing now: the boss is customer #1.

 

[novagamer]

I think it's the cloud that has decimated high-end gear demand. Much cheaper to rent cloud compute short term IMHO.

This is only true if you have very short burst-y demands a few times a year.

Continual use it makes far more business sense to own the hardware and depreciate it over 3 years, even at the highest end. There’s a great anecdote about this on 37 signals blog, and despite me really being annoyed by them overall, they are right and it saved something like 70% vs their prior cloud setup.

I’ve worked on projects that used ~$20k of cloud compute over a weekend, but that was necessary due to the scope of the work being done and the timeframe – if we needed to do that more than once or twice we would have bought the servers.

Wasn't Steve also a Mac Pro user? That would be the one thing missing now: the boss is customer #1.

This I agree with, and it’s also a good example of why the iPads are leading both the industrial design and the technology, Cook uses one primarily.

 

[quarkysg]

This is only true if you have very short burst-y demands a few times a year.

Continual use it makes far more business sense to own the hardware and depreciate it over 3 years, even at the highest end. There’s a great anecdote about this on 37 signals blog, and despite me really being annoyed by them overall, they are right and it saved something like 70% vs their prior cloud setup.

I’ve worked on projects that used ~$20k of cloud compute over a weekend, but that was necessary due to the scope of the work being done and the timeframe – if we needed to do that more than once or twice we would have bought the servers.

I agree, but is the demand high enough? That’s the million dollar question.

 

[MRMSFC]

Wasn't Steve also a Mac Pro user? That would be the one thing missing now: the boss is customer #1.

IIRC Steve had a custom “mini” Mac Pro on his desk in photos. (I want one too)

I agree, but is the demand high enough? That’s the million dollar question.

Well, from big business customers, probably not. Apple typically can’t provide the sheer volume of products that they demand, that and Macs typically aren’t compatible with Windows centric IT deployment systems.

However, Apple’s strength has typically been smaller enterprises and the “prosumer” demographic. The kind of people who typically don’t follow “upgrade cycles” and buy a new device when they need one.

For those customers having a product line that’s constantly up to date is more beneficial.

 

[MayaUser]

 

[DrWojtek]

M3 was so weird why did the Pro model deviate from the earlier (and later) models that much? What were they hoping for?

 

[Chancha]

M3 was so weird why did the Pro model deviate from the earlier (and later) models that much? What were they hoping for?

To lay the ground work / as a testbed for the 6 E-Cores for the eventual base M4? Also perhaps experimenting with triple memory channel sets?

 

[Antony Newman]

If the Apple H/W teams need over half a year to finalise the Hidra beyond their delivery of the M4 Max - do we really think they have been cooking up a huge low yield monolithic SoC that is twice the size of the M4 Max (>=80 GPU CUs in the Studio)? Do we think a reticle sized M4 fusion connection between two Hidra SoC's for the Mac Pro on one side of the Hidra is practical - or even a good strategy for the future when reticle sizes will halve (due to anamorphic lenses in High NA EUV).

What if Apple are actually dreaming bigger - but using smaller SoCs?

If the (mythical) Hydras had 9 x heads, perhaps Apple have been hinting all along that their top of the line offering for their next Mac Pro will be build from 9 x SoC tiles with flexible (configurable) fusion interconnects between them (Tile-to-Tile or Tile-to-Memory)?

A power hungry Inter-Tile connection might not be suitable for their mobile offerings - but could be the only practical solution where large (low yield) SoC's costs would have to passed onto the consumer.

Thought Experiment:

+) If a MacStudio was a 2x2 configuration of Hidras
- And a single Hidra was an SoC close in the size of the M4 PRO
- With 8P + 4E + 30CU (fewer CPUs, more GPUs than the M4 PRO)
- If the total 32P.16E.120CU machine had a performance that was 65% scalable
- It could behave like a monolithic 20P.10E.80CU machine
-> Roughly twice the GPU performance of the M4 Max

+) If a MacPro was a 3x3 configuration of Hidras
- And a single Hidra was also 8P + 4E + 30CU
- If the total 72P.36E.270CU machine had a performance that was 50% scalable
- It could behave like a monolithic 36P.18E.135CU machine
-> Roughly 70% faster GPU than the (Hidra) MacStudio

 

[tenthousandthings]

FYI, the identifiers are out for the new models (not sure which is which for the two Minis), as follows:

M4

Mac16,1 :: M4 120 (10/10) MacBook Pro 14"

Mac16,2 :: M4 120 (8/8) iMac (Two ports)
Mac16,3 :: M4 120 (10/10) iMac (Four ports)

Mac16,6 :: M4 Pro 273 (12/16, 14/20) MacBook Pro 14"
Mac16,8 :: M4 Max 410 (14/32), 546 (16/40) MacBook Pro 14"

Mac16,7 :: M4 Pro 273 (12/16, 14/20) MacBook Pro 16"
Mac16,5 :: M4 Max 410 (14/32), 546 (16/40) MacBook Pro 16"

Mac16,15 Mac16,10 :: M4 120 (10/10) Mac mini
Mac16,10 Mac16,15 :: M4 Pro 273 (12/16, 14/20) Mac mini

It looks to me like M4 will be the first generation where every Mac gets the current silicon.

EDIT: To clarify, the information is from Apple’s official “Identify your iMac/MacBook Pro/Mac mini” support pages. My observation about it is based on the fact there are at least five unaccounted-for identifiers in between Mac16,1 and Mac16,15. So seems there is room for all remaining Macs in the current lineup.

ADDITIONAL NOTE: It looks like “Mac16,15” is an error, though it hasn’t yet been corrected. See comments below.

 

[name99]

IIRC Steve had a custom “mini” Mac Pro on his desk in photos. (I want one too)

You can get one!

https://twitter.com/i/web/status/1854565512610525195

 

[Gnattu]

M3 was so weird why did the Pro model deviate from the earlier (and later) models that much? What were they hoping for?

My speculation is that it was a compromise to the terrible yield and huge manufacturing cost of the TSMC N3B. When the cost reduced with the new N3E process they bring the P cores back in M4.

 

[scottrichardson]

My speculation is that it was a compromise to the terrible yield and huge manufacturing cost of the TSMC N3B. When the cost reduced with the new N3E process they bring the P cores back in M4.

6 P cores being exactly half what the max got for M3 lends credence to this. I wonder if their P core clusters had to be in sets of 3/6?

 

[Chancha]

FYI, the identifiers are out for the new models (not sure which is which for the two Minis), as follows:

M4

Mac16,1 :: M4 120 (10/10) MacBook Pro 14"

Mac16,2 :: M4 120 (8/8) iMac (Two ports)
Mac16,3 :: M4 120 (10/10) iMac (Four ports)

Mac16,6 :: M4 Pro 273 (12/16, 14/20) MacBook Pro 14"
Mac16,8 :: M4 Max 410 (14/32), 546 (16/40) MacBook Pro 14"

Mac16,7 :: M4 Pro 273 (12/16, 14/20) MacBook Pro 16"
Mac16,5 :: M4 Max 410 (14/32), 546 (16/40) MacBook Pro 16"

Mac16,15 :: M4 120 (10/10) Mac mini
Mac16,10 :: M4 Pro 273 (12/16, 14/20) Mac mini

It looks to me like M4 will be the first generation where every Mac gets the current silicon.

EDIT: Not sure why @komuh is angry with regard to this news? To clarify, the information is from Apple’s official “Identify your iMac/MacBook Pro/Mac mini” help pages. My observation about it is based on the fact there are at least five unaccounted-for identifiers in between Mac16,1 and Mac16,15. So seems there is room for all remaining Macs in the current lineup. That’s a good thing, no?

Mac16,10 is M4 mini
Mac16,11 is M4 Pro mini (full / binned)

Don't think 16,15 has been used yet.

 

[Gnattu]

6 P cores being exactly half what the max got for M3 lends credence to this. I wonder if their P core clusters had to be in sets of 3/6?

No, the binned M3 Max and M4 Pro has 5 core P cluster

 

[Confused-User]

My speculation is that it was a compromise to the terrible yield and huge manufacturing cost of the TSMC N3B. When the cost reduced with the new N3E process they bring the P cores back in M4.

There was no "terrible yield". That was an internet rumor that somehow took hold, but it was never true. There is zero evidence for it, and tons against it, including public statements by TSMC both before and after N3B shipped, and their financials.

The most likely explanation is that Apple was experimenting with product positioning. In the M4 generation, they're doing the opposite (building a Pro with extra-beefy CPU). No doubt what they do with the M5 and future generations will be informed by which type of pro maximized total profit (not just sales of Pros, but the whole sales mix).

6 P cores being exactly half what the max got for M3 lends credence to this. I wonder if their P core clusters had to be in sets of 3/6?

I can't imagine why you think core counts "lends credence" to that argument. It does not. Core clusters can be in whatever arrangement Apple thinks makes the most sense. They've used cluster sizes of both 4 and 6 for P cores, and 2, 4, and 6 for E cores, up through the M3 generation. M4 is likely using 5-core clusters in the Pro, though there's a small chance they're 6-core clusters with one core fused off (and possibly cut at the edge of the chip) per cluster. I haven't seen any word on this today but now that they're shipping we should know in the next couple of days.

No, the binned M3 Max and M4 Pro has 5 core P cluster

M4 Pro, most likely. M3 Max, no. It's a 6-core cluster, with one core fused off.

 

[DrWojtek]

There was no "terrible yield". That was an internet rumor that somehow took hold, but it was never true. There is zero evidence for it, and tons against it, including public statements by TSMC both before and after N3B shipped, and their financials.

The most likely explanation is that Apple was experimenting with product positioning. In the M4 generation, they're doing the opposite (building a Pro with extra-beefy CPU). No doubt what they do with the M5 and future generations will be informed by which type of pro maximized total profit (not just sales of Pros, but the whole sales mix).


I can't imagine why you think core counts "lends credence" to that argument. It does not. Core clusters can be in whatever arrangement Apple thinks makes the most sense. They've used cluster sizes of both 4 and 6 for P cores, and 2, 4, and 6 for E cores, up through the M3 generation. M4 is likely using 5-core clusters in the Pro, though there's a small chance they're 6-core clusters with one core fused off (and possibly cut at the edge of the chip) per cluster. I haven't seen any word on this today but now that they're shipping we should know in the next couple of days.


M4 Pro, most likely. M3 Max, no. It's a 6-core cluster, with one core fused off.

As in, for the M3, Apple wanted to see how badly the average Mx Pro-buyer needed the CPU power. If they needed it a lot, lowering the CPU performance on Pro could upsell them to taking the Max instead.

But it turned out, these customers were already upselled to the Pro from the base, and when CPU weren’t that big of an upgrade anymore, people were more likely to just get the base version and save the cash.

Meaning those who get the Max mostly do so because of GPU, not CPU.

That’s my take on it.

 

[Gnattu]

There was no "terrible yield". That was an internet rumor that somehow took hold, but it was never true. There is zero evidence for it, and tons against it, including public statements by TSMC both before and after N3B shipped, and their financials.

Even if it has no "terrible yield", it can still be much more expensive and if you have evidence against this claim please elaborate.

The most likely explanation is that Apple was experimenting with product positioning. In the M4 generation, they're doing the opposite (building a Pro with extra-beefy CPU). No doubt what they do with the M5 and future generations will be informed by which type of pro maximized total profit (not just sales of Pros, but the whole sales mix).

I highly doubt this because silicon developments has to take some time and it is very unlikely to make such decisions based on the actual sales data because the time frame will be way too tight. Or you can say that Apple designed two plans for the Pro and they can make a decision to pick one to mass production in the last minute, this is more realistic but why wasting efforts for a silicon that will never bing released at all? I'm confused and it looks like waste of R&D money to me.

M4 Pro, most likely. M3 Max, no. It's a 6-core cluster, with one core fused off.

It does not make much difference with just fuse off a core and physically remove a core, that is why I'm calling both a 5-core cluster, because it is very unlikely to redesign the cache and the interconnect just for this 5-core cluster.

 

[MayaUser]

he doesnt have evidence but for sure M3 could be more expensive to make for no good reasons, fact that Apple moved for M4 more quickly than they did with the M1-2-3 with still good gains, he is an confused user, so dont take it too much personally

 

[MayaUser]

There was no "terrible yield"

He said that hes speculating while you are saying with confidence that it wasnt, please provide with your proof otherwise we let this just an attempt of your love to argue with almost everybody