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.
Apple's M3 Pro, M3 Max, and M3 Ultra Chips Could Offer Even More CPU and GPU Cores
- Thread starter MacRumors
- Start date
- Sort by reaction score
Absolutely agree.
If more E cores are valuable even for "hardworking machines" (as ways to offload OS work, interrupts, IO, and general background scut work) for a Pro, that's just as true for a Max.
Like TSMC n3 is an Apple exclusive over the long term. x86 will get TSMC N3 also. There is no way Apple can 'nail in the coffin' to x86 just on a fab node that is generally available to anyone who wants to pay.
The N3 supply is also limited. There is no way TSMC can even produce as many x86 chips as Intel does while still servicing other paying customers. Part of the whole point why Intel is farming their iGPU to TSMC and doing the CPU chiplet/tile themselves. They really can't do both on bleeding edge nodes either. There are only so many very top end EUV fab machines out there and the growth rate is not sky high either.
Don't hold your breath. As the FaceID system gets cheap enough , that is probably going in. the notch is just prep.
As for healthy wrists... a flat wrist position is better than hands titled up from the wrist.
Apple's own site:
"...
Support your forearms
Find a neutral positions with your forearms parallel to the surface that you’re working on with your hands and wrists straight. When you can, try to support your forearms with armrests or a nearby surface. To find a comfortable, neutral position, change the height of your chair, work surface, or footrest....."
Ergonomics - Apple Support
Learn to set up your environment and use your devices safely.
Or just go look at modern , higher end ergonomic keyboards. for example
If there is any 'tilt' needed, then it is tenting to put the wrists more so into a natural handshake position.
( again go to the section on Apple's ergonomics page and "check your set". The picture above that section where the arms are positioned over the desk is flat ; not up sloping. ]
To keep your wrists straight/flat you would need to 'shoot at an angle ' your forearm up onto the desktop surface the laptop is sitting on top of . Many folks have 'normalized' that , but it isn't best ergonomic positioning.
[ The other disconnection from ergonomic best practices is where that page says "adjust your monitor height" and by default Apple's monitors/iMacs do no such thing at all. ive and company spun fixed height iMacs as 'good'... that really wasn't grounded in ergonomics. Same thing with the Air. It mostly just looks nifty. ]
The Air was far more positioned to win the thinnest edge pointed at the user prize , than prizes in ultimate ergonomics.
IMO Apple's lowest end will (unsurprisingly) stay lowest end.
Folks who want Apple's lowest end to no longer be Apple's lowest end just because a newer chip iteration comes out need to think more logically.
If the rumors about M3 Pro being stuffed into 13-14" iPad Pro then it would make sense.
First, die area cost for a 'half' E core cluster is much smaller than the full P-core cluster (4 cores plus much larger L2 cache ). N3 wafer costs are going up. Not only more space , but costs more also. The iPad Pro 14" wouldn't be cheap, but also not likely going to free of price elasticity problems either.
Second, adding deployments to iPads would tip the scales toward mobile : 2-3 (MBP 14/16 + iPad Pro . 3 if count 16" as separate) vs 1 ( Mini Pro). So it has a mobile focus. The top end iPad Pro wouldn't have much of a competitive in sight in its class.
Pretty decent chance that a more powerful Pro SoC will mean fewer Max SoCs sold in MBP 14/16 deployments. The unit volume of Pro die will go up and Max probably tread water. ( The Pro SoC is more affordable. it should get more buyers than the Max. When can get more M1 Max performance at M3 Pro prices ... many customers will keep more of their own money. Or shift it over to cost of complements ( more RAM and storage. since those are one-time buys) )
Third, with the Mac Pro now around, the Max goes into just as many desktop products than it does laptop products. (Studio and Mac Pro vs. mostly the same laptop with varying screen size 14"/16" ). Throw on top Apple's problems not having a "bigger than Ultra" option and the desktop products are in a more critical position competitively. [ Even more so if no solution planned for M4 or M5 either. Two die solution is only option. ]
For the 'princess and pea' latency folks in music area , more P cores probably are more highly values than trying to mix and match P/E cores on highly synchronized timings. In the mobile space Apple's P cores very competitive with other x86 'E'/'C' cores. So not really a huge backslide in the laptop space either.
Fourth, the overwhelming die area consumption on a Pro ( and Max) is NOT the CPU cores. The other stuff is the bulk of the 'value add' of the die. Tweaking the CPU core count is really just playing at the edges. The primary reason to keep it exactly the same is to incrementally lower design overhead costs ( on dies and software ). The Pro and Max have likely been each successful enough to stand on their own a bit more than in the early M-series transition when the Pro and Max deployments to Macs were substantively incomplete. ( no Mini Pro or Mac Pro until 2023. )
IMHO, there is a fifth in that the Max really should be heading toward a chlplet first focus rather than monolithic. At least the desktop deployments of the "Max class" SoCs. (split a healthy chunk of I/O and non computational cores off onto a separate die ) . If there is no other Mac to deploy to then desktops may drag the "Max class" away with them over time. If they ever want to cost effectively scale past 2 compute dies ( bigger than Ultra) they'll probably need to go down this path to stay competitive.
This would cost more money. It is 'fishy' in that it isn't the Scrooge McDuck path to fatter profit margins. But technically in the product strategy space. it makes sense. Apple makes different dies for different products. Apple goes overboard sometimes trying to make as few as die varieties as possible.
Running an M2 Max MBP with 96 GB RAM I disagree with this part of your comment. Battery life with the M2 Max chip is IMO very very good. The maximum M2 MBP Max remains solidly a portable system that *can* be an excellent desktop replacement (the way I use it 90% of the time driving internal display + three 4K displays). The only real desktop limitation being the lack of more than 3 TB4 ports.
There is no a Flash/NAND chip glut in 2023-24. Apple could be super greedy and continue to kneecap the entry SSD configurations, but decent chance they could go back to not being quite so miserly with the NAND packages and pair up even the entry configuration. How many decades can they stick to the $400/TB pricing?
Just because the core count is the same doesn't mean the performance is extremely close. It is more than just 'core count' wars. These likely will be different cores with more transistors and features thrown at them. Shouldn't need more to deliver more computational 'horsepower'. Better caches . More specialized compute for common code sequences , etc.
TSMC N3 doesn't really shrink the SRAM/cache size much. So more cores that drag along with them abnormally large L2 caches could actually mean die area growth. Given the N3 wafers are even more expensive , that is just a bigger and more expensive chip. I doubt most folks are jumping out there seat demanding higher cost increases.
[ In the past , small fab process generally lead to same or lower costs which some of that got passed onto the consumer. That is mainly over. ]
No, it will not. And the Studio Max will always better handle high-load heat than the Mini and take much more RAM, which is hella more life-cycle relevant than many realize.
Last edited:
Most of the posters in this thread seem to be missing a lot of relevant background info (aside from d60 and name99).
Here's the key point: The M2 is a (relatively) last-minute rehash of the M1, running at slightly higher clocks. This happened because the original M2 design was for TSMC N3, but it had to be shelved because TSMC couldn't ship N3 in time. The design could not be back-ported to N5 - it was simply too large. (This also affected the A16.) So the cores (GPU, CPU, NPU) are getting roughly two years of design improvements, and all the other pieces (NoC, memory controllers, video pipeline, (de)compressors, etc.) are mostly getting 2-3 years of improvements, though some of those may have had bigger boosts in the M2, leaving less for the M3 to improve upon.
This means that the M3 is going to be an unusually large step up from the M2. *At the same clock speed* we'll probably see CPU single-core performance (IPC) improvements of 20-30%. (The latest shabby rumor puts the number at 22.5% in GB6, though I don't consider it especially reliable.) But will the M3 actually run at the same clock speed? My guess is that the M3 will have only a small or no clock increase, but the Pro, Max, and Ultra are progressively more likely to run at higher clocks. We could easily see a total speed improvement of >50% if they do, but even the baseline M3 will probably take the single-core performance crown from Intel (if it does come out before Intel's next generation, at least).
The same story applies to the GPUs. The entire original A16/M2 GPU core design, reportedly including ray tracing, was shelved. Bringing that back, adding another year of design work, should produce real improvements before any clock increases.
I too am highly skeptical of the Pro having 6 E cores. E cores are tiny; if they were going to have more than 4 cores, they'd probably just design a 6-core cluster, and use that in the base M3 too, or else just put two 4-core clusters in the Pro/Max. But my bet is 4 E cores in the M3 across all three chips. (Ultra, being 2x Max, would have 8 total.)
Lastly, Apple continues to file patents involving more than two chips connected together (the "Extreme"). But I don't think we have any reliable information on whether we'll ever see that. The current Mac Pro is depressing but not necessarily an indicator.
Here's the key point: The M2 is a (relatively) last-minute rehash of the M1, running at slightly higher clocks. This happened because the original M2 design was for TSMC N3, but it had to be shelved because TSMC couldn't ship N3 in time. The design could not be back-ported to N5 - it was simply too large. (This also affected the A16.) So the cores (GPU, CPU, NPU) are getting roughly two years of design improvements, and all the other pieces (NoC, memory controllers, video pipeline, (de)compressors, etc.) are mostly getting 2-3 years of improvements, though some of those may have had bigger boosts in the M2, leaving less for the M3 to improve upon.
This means that the M3 is going to be an unusually large step up from the M2. *At the same clock speed* we'll probably see CPU single-core performance (IPC) improvements of 20-30%. (The latest shabby rumor puts the number at 22.5% in GB6, though I don't consider it especially reliable.) But will the M3 actually run at the same clock speed? My guess is that the M3 will have only a small or no clock increase, but the Pro, Max, and Ultra are progressively more likely to run at higher clocks. We could easily see a total speed improvement of >50% if they do, but even the baseline M3 will probably take the single-core performance crown from Intel (if it does come out before Intel's next generation, at least).
The same story applies to the GPUs. The entire original A16/M2 GPU core design, reportedly including ray tracing, was shelved. Bringing that back, adding another year of design work, should produce real improvements before any clock increases.
I too am highly skeptical of the Pro having 6 E cores. E cores are tiny; if they were going to have more than 4 cores, they'd probably just design a 6-core cluster, and use that in the base M3 too, or else just put two 4-core clusters in the Pro/Max. But my bet is 4 E cores in the M3 across all three chips. (Ultra, being 2x Max, would have 8 total.)
Lastly, Apple continues to file patents involving more than two chips connected together (the "Extreme"). But I don't think we have any reliable information on whether we'll ever see that. The current Mac Pro is depressing but not necessarily an indicator.
My thinking is that they’ll all be around the same clock speed as they’ve always been. There’s most likely to be a 15-25% improvement JUST because that’s all Apple has to do and still ship the fastest Macs ever.
Regarding single-core performance, Intel will simply goose one processor in their line with some obscene amount of juice just in time to juussst barely eke out single-core performance, officially. For that reason alone, no Apple Silicon processor will ever be able to claim that crown, and that’s totally fine. A “number higher than Intel” is likely to account for far fewer sales than the effort to create it would be worth.
The Face ID components are twice as thick as the entire MBP lid.
There's this constant thread of whining in these fora about how Apple does the minimum possible, usually associated with other whining about "cash grabs". Historically, that's quite inaccurate, at least with respect to their silicon team.
Certainly at least from the A7 onwards, when they went 64-bit years ahead of a shellshocked and disbelieving competition, Apple went all-out on performance and efficiency, generally prioritizing the latter (but not always) while still running wildly ahead of everyone else - so much so that the competition has been deeply embarrassed from that day to this one. So much so that there are legions of deniers out there unwilling to accept plain-as-day evidence they can easily see for themselves. It might be the biggest competitive blowout in the history of CPUs - when else did one company have cores (or CPUs, back when it was all uniprocessors) so superior the the rest of their industry for so long? The Intel/AMD wars looked a bit like that for a while, in both directions at different times, but for most of the time there were other CPU makers out there with competitive performance, just different architectures (all of the time, if you are willing to lump POWER in there). In mobile chips, that hasn't been true for roughly a decade.
They could bump performance by 5% and have the fastest Macs ever. Or 1%. I will bet my life against a wooden nickel that that's not what they're doing.
They design cores with A series chips top of mind, also pushing them as hard as they can for their M series without screwing efficiency. There is no doubt at all that they're doing that for the A17/M3 as well. Within some very broad constraints, they're going to build the best chips they can.
You don't know that, and I think there's an even chance you're going to be wrong in the next three months.
It's possible that starting with the Intel 4 process, you might be right. For a while. "Ever" is a long time.
The dual edge sword there is probably creates more bandwidth hungry CPU, GPU , and NPU cores. Probably the display controllers also ( especially if Apple picks up DisplayPortv2.1 support or just minimally goes after supporting more very high refresh displays. ).
There is probably also pressure to increase the SSD bandwidth. ( as PCI-e v5 SSD start to perculate onto the market).
The M2 did bring internal network improvements over the M1. In that subarea I don't think going to see huge multiple generational kind of improvements. Something better, but already got something better in M2. Probably not as big of a 'step function'. ( All the more so if the 'doom and gloom' about being stuck on LPDDR5 is true. But if really sitting on a 2022 design arriving in 2023 then perhaps it is . )
TSMC N3 never was going to be on time for the A16. Even back before the pandemic. Kind of skeptiical that the M2 really was suppose to be N3 because it couldn't have arrived any eariler than Q4 2022 even if TSMC schedule had not slipped. Apple was going to run the Mac on two year update cycles? 18 months or so seems more likely.
If the internal network bandwidth is saturated it may have come down to a choice between adding a P-core cluster or an E-core cluster. If the additional 2 Ecores are primarily after better , longer battery life. then going to 6 can be be does with two E core clusters. The additional one is chopped down (or binned down on purpose). For very low loadings can turn off not only all the P clusters but the other E cluster also. If can handle with just two E-cores ( asleep and screen doing nothing) can drop the other four E cores. [ Intel's meteor lake reportedly has two cores on the SoC chiplet/tile that just handle sleep/standby loads so they can turn the whole CPU chiplet off completely. Can do something very similar with a 'chopped down' 2-core E cluster. Go back and look at M1 Pro/Max (they have already done that). ]
Putting in a half size E core cluster is not a bigger 'rocket to the moon science' project than adding a third P core cluster. They are both primarily designed to be used as building blocks to make a incrementally bigger die. I think you are inventing complexity ( a custom 6 core cluster) where it is not even needed in the slightest.
Likely not because probably don't have the die area allocation. And also cannot propagate that down to the A17 which has even less of a die area allocation ( or need ) for that. P or E core cluster addition would bring more cache ( SRAM). SRAM doesn't really shrink with TSMC N3 ( N3B a little and N3E and others not at all. ). The M2 generation already install die bloat on the M-series dies. The last thing Apple need with the shift to far more expensive N3 wafers to do more die bloat. Apple is probably looking for a shrink back to 'normal' sizes that M1 was allocated.
Again goes back to whether they internal bandwidth to do that and also add the 'extra' P core cluster on the Max. The GPUs ( and NPUs) are probably sucking up much if not all of the improvements. So that's why I suspect it is an 'either or' and not a 'just grow everything ' choice. The cluster count probably isn't going up on NPU or GPU either ( if Gurman's core counts are right).
Apple reportedly built the Extreme and got it working in prototype Mac Pro systems. It is more so that it was too expensive. Patents that get you some Rube Goldberg complex ,too expensive to make SoC really don't do much.
If Apple is willing to let the Mac Pro and Studio pull the Max die into a more P core count than the others configurations that is likely a good sign long term for the Mac Pro. That is the system that extra P core cluster is the biggest 'win' for (presuming that back those four CPU clusters up with sufficient bandwidth without draining the GPU/PNUs any. ) . And crank up the baseline I/O on the Max unit. ( Mac ( M3 Pro/Max ) -on-add-in card with full x16 PCI-e v4 bandwidth to the host would be significant. )
Yes all the rumors point to that. So the gain will be in battery life not so much performance. Which is why they want to differentiate the Pro machines with much much more power.
For me, I'm not too concerned with more cores, CPU or GPU (for M3, M4, etc....)
I'd be interested in seeing how the RAM ceiling has been addressed, overall.
richmlow
I'd be interested in seeing how the RAM ceiling has been addressed, overall.
richmlow
Sure, but that's just the cost of doing business. Bandwidth needs go up over time. The answer to that is relatively obvious, and it's not challenging for Apple to know when they need to deal with that. They've repeatedly shown (in the various M chips) that they're willing to spend resources to get this right.
That seems highly unlikely. Like, 100% not likely. Area is a much more compelling argument.
Your argument (not quoted) about power efficiency is more interesting, and it's why I said I don't believe the (still plausible) 6E cluster idea. But I also doubt the 2E cluster idea. They moved phones from 2P2E to 2P4E, which I think says it all.
Your (also unquoted) argument that another 2E cores (going from 6 to 8) is too expensive in die area seems unlikely to me. I mean, I can easily see the engineers deciding "we just don't need more E cores", but saying "we don't have room for 8, but we have room for 6" seems very unlikely. The E cores are really small. And unless you redesign the shared cache and the rest of the cluster support, you're not saving any space on that.
Current pricing suggests they see a lot of elasticity in Mac Pro pricing. If they had it working reasonably well, I think they'd have built it.
Gee, the M3 will have more than the M2, which had more than the M1.
The question is, when will there be a meaningful gap for those not chasing every last bit of performance. My guess is the M4 or M5 over the M1.
At this point I could care less about additional battery life or power efficiency. Give me more performance per current watts (or an option for more watts and more performance - especially the GPU, which is still anemic compared to Nvidia).
The question is, when will there be a meaningful gap for those not chasing every last bit of performance. My guess is the M4 or M5 over the M1.
At this point I could care less about additional battery life or power efficiency. Give me more performance per current watts (or an option for more watts and more performance - especially the GPU, which is still anemic compared to Nvidia).
I'd argue that the GPU situation below is the largest limitation.
This is what I was referring to - I just built a gaming PC with a 4070 and my M1 Max mac isn't even in the same zip code let alone ballpark. No way does an M3 even come close given Apple's over-rotation on performance per watt.
Honestly, even if it were, Apple really needs to add eGPU support for things like ML and heavy graphics loads. Because most of those projects/applications are built for Nvidia, or maybe AMD, and they aren't going to port to a platform that uses a completely different set of ML APIs (and gives middling performance).
Great data point: The new PC uses ~475 watts under load. That's 3x what my m1 mac does (which includes the monitor). There's simply no way an Mx chip can compete - even with it's rather amazing architecture.
Last edited:
You seem to be the only one, except for me, that understands what binning is. Even this MacRumors' article refers to binned and un-binned chips. The only chips that are unbind are those that have not been tested yet. All tested chips are binned. Some of them with fewer cores enabled, and the ones that pass all test are binned as fully functioning. There are no unbind chips in any products.
No whining here, just reality.
Historically, one processor generation is roughly 20% or so better in single threaded performance than the last generation for the chips Apple produces. That held true for the M2 over the M1 as well, so there’s a bit of a precedence here. When the M3 comes out with roughly 20% better single threaded performance, there will still be those that think M4 will have 40-50% performance. So, too with the M5 and on and on.Right, and the “best chips they can” that meet the thermal and power envelopes they’re looking for, will be roughly 20% more performant than the M2.
Happily willing to be wrong there, but Intel has shown a very strong desire to NOT lose that crown, by any means necessary.
They are JUST desperate enough to put out numbers that require the type of system cooling that, separately, costs as much as the rest of the system! As long as they can claim to be shipping it to OEM’s, it becomes a data point on the list… and they’ll be sure to announce it right before the next Apple Silicon generation ships.