I think Apple could go with M3 base for Macbook Air just for battery life, so better in cpu but lack gpu power of the current MBP, thats just my opinion
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.
3nm?
- Thread starter gojilla
- Start date
- Sort by reaction score
Pro and Max use the same exact CPU anyway.
But the main effect of doubling the wattage would be more performance range between the mobile and the desktop. A 14” chassis won’t be able to sustain full 80W for 8 CPU cores. But a larger desktop chassis (Mac Studio etc.) would have no problems with it. So going higher power would address the complaints about the lackluster desktop performance while still delivering desktop-class mobile performance.
I don’t think there is many who decides between the two. Either you want the best and largest display, vastly better GPU, and many ports, and a fan for sustained work - Or you want the most portable machine.
Last edited:
Since power is, to a first approximation, quadratic with frequency, a doubling of power could allow a 40% boost to clock speed, which would be stunning. If the M3 shows a 15% improvement over the M2's ~1950 GB SC score, then 40% on top of that would be 1950*1.15*1.4 = 3100 (!!).
oh yes, of course...now i understand, when you said prosumer i included the laptops too
I would like to see Mac Studio allowed to draw more wattage since half of that thing is cooling
I would hope for this in the Mac Pro, but i am sure the M2 Max that will go into the Mac pro will daw at least double the wattage? maybe even more if thats possible by design?
Yes, but that would not be an laptop thing i guess...but for a Mac Studio or Mac Pro would be ashamed if that will not be near the reality
Laptops would use the same chips, but of course they would be unable to reach the full performance potential. Not unlike modern x86 CPUs by the way where the exact chip can be used for both mobile and desktop.
Design is the key. Of course, the question is whether such design is even possible without sacrificing something else. it can be that by targeting higher clocks Apple could lose the efficiency advantage. But then again they will always have a huge advantage over competition: they can throw much more money at the problem.
There is a lot of evidence that the replacement cycle is supposed to be around 12 months but was disrupted due to various delays and logistic issues.
If I'm right the M3/Pro/Max/Ultra chips will be out by Jan/Apr/Oct 2024. All using 3nm die shrink.
So 3nm chips for 2023 will all be A17 Bionic chip for the iPhone 15 Pro Max.
We’ll have to look at longer timescales to assertion these things. And of course, Apple might also change the plans in the meantime.
The only way to be right or wrong about these things is to get Apple leadership to take about their original plans…
Visualize it...
Will lines stay parallel?
Will lines cross!?
Will lines dead end?
Will lines stay parallel?
Will lines cross!?
Will lines dead end?
Last edited:
I think that has some suspect assumptions built in. In turn ...
1. That Apple is going to use 100% of N3 options just to raise performance to the exclusion of grabbing some power savings.
a. According to some reports the reason the the M2 Pro/Max are squatting on multiple years old N5P is that some new features 'blew' the power and/or space (die area) budget. If going to add new additional power consuming accelerators then pretty good chance going to use some of the N3 tradeoffs to claw back some power usage. Not 100% of the trade-offs spread on chasing single thread , 'top fuel dragster' , single thread performance. They don't have to spend it all on power savings but some far more balanced approach.
b. Caches are going to shrink at all from the N5P they are using now. L2 / System(L3) stuff will be just as big. If try to slap hot rod engines that crank of the number of cache misses ... not really going to be able to counter much with Apple's classic move over the last 6+ years of "throw more cache at it.". If cores get smaller than can add more cache. However, throwing the kitchen sink into the cores so they don't get any smaller (even if the N3 shrink) will likely increase the cache pressure.
( There will be some handwaving that LPDDR5X+++ will help bail out. I wouldn't be on Apple leaning of some super bleeding edge LPDDR5 process for reliable , high volume supply chain in 2023-2024. )
c. The A16 , M2 , M1 Pro/Max have all bloated out to larger die sizes than their predecessors. N3 wafers cost even more. There is very little chance Apple is going to be trying to get to the same M2/A16 die sizes with N3. And again 'smaller die' it going to put pressure on "about the same stuff smaller". I wouldn't look for some large increases in core counts anywhere. ( maybe in some 'new' accelerator logic that doesn't increase the cache pressure much. Or filling in gaps. AV1 en/decode , DisplayPort 2.1 , the HW Ray Trace that some are so desperate to have included . That isn't going to get revolutionary better Geekbench scores though. )
FlexFin could be extremely useful, but it is also new ( in tool support and volume production). It would make sense risk management wise to take the 'shrink wins' and call it a day on the first iteration on N3 family. There will be some performance uplift with the N3 move. Especially, if jumping from older N5P. Some modest clock bumps (on cores and caches) , internal network improvements , taking time to 'fix' things that didn't work as smoothly before but now do with closer distances , etc.
2. The power gains that Apple is tap dancing on with M2 Pro/Max are largely on doing almost nothing on the CPU/GPU cores. The video decode get more power efficient. That is nice by how long can keep that misdirection going? AMD is getting better at "performance on battery". They haven't gotten to parity but they are closing the gap. Same thing with dGPUs from Nvidia in laptop space. That humongous vast gulf between 3070 level performance and Apple's stuff is substantively smaller now. Still somewhat large, but it has a bit less 'wow' factor to it.
End of the day this M3 family CPU core , GPU core , etc are all going into a phone SoC. Taking the N3 trade-offs on power/performance to 100% leaning on the performance side isn't going to help that phone SoC much.
N3 wafers cost more money. So the notion of 'completely fork' the m-series off from A-series is actually likely harder because margins are going to get squeeze more. Apple isn't going to the 'poor house' , but probably not likely to arbitrary take the attitude of "throw money at all with zero concern about profit" on how much 'shared across product lines" the SoCs leverage.
A smaller chip that consumes less power is probably on the priority list. Not to the point there is zero performance uplift, but high enough priority that "more performance" is not the only design parameter on the table.
M3 might threatened the M1 Pro but that should be relatively OK. The M2 Pro has twice the GPU cores of the M2. N3 isn't going to bring some 'doubling' of GPU core counts and the individual cores won't jump that high either. Probably zero increase in CPU/GPU core count (smaller die with same size cache). There is zero rational reason to skimp on 'hiding'/'removing' the accelerators from M3 .. Especially if there was a 'huge miss' that forced them onto N5P. If there was no space ... now there is on the node progression. N3 allows them to put in a proportionally sized accelerator without the die bloat. That why you pay more for the new fab process. To pay more and skip it is somewhat looney toons. There is enough shrink with N3 that can pull back the die size and also put in another feature or two as long as pick something this is highly aligned with the "shrinks better with N3" properties. (e.g. can't throw N3 at bigger cache size and get smaller. More compute with relatively little additional cache is a far better 'fit' to pay more. )
There is none! I guess it implies "class of chip", but I'm sure I could do # of transistors or something similar...
Bear in mind that part of the cost of an A78 is AArch32 support. Apple took AArch32 out of its cores from A11 on, more than 5 years go. Granted, it does not really cost that much, but every little bit helps.
Last month's prediction from DigiTimes: We won't be seeing M2 Pro/Max MBP's in early 2023:
Apple's M2 Pro and M2 Max MacBook Pros Reportedly Delayed Again
Apple's M2 Pro/M2 Max MacBook Pro refresh was initially expected late last year.
I expect M3 to show a much bigger improvement over M2 than 15% because it should be using A17 and 3nm. Remember that the M2 is using A15, which would be 2 generations and 1 node behind M3.
- M2 GB5 score is 1950.
- A16 improvement over A15 in ST is 11%. 1950 * 1.11 = 2165.
- Let's assume that A17 will have a 20% improvement over A16 because it will be using 3nm and we're hearing reports that A16 ran into engineering issues. 2165 * 1.2 = 2597. A score of 2600 GB5 seems reasonable for M3.
- If we go by your assumption that doubling the power can yield 40% more performance, then 2597 * 1.4 = 3635.
Anyways, I doubt Apple would double the power for M3 Ultra. Their own marketing is centered around efficiency.
Last edited:
That was just meant as a rough approximation, to illustrate how much faster future M-series processors could be, if those processors were scalable to those higher frequencies.
The idea was that Apple might choose to take advantage of this for their desktop devices, specifically the Studio the Mac Pro, where Apple could shift their marketing emphasis from efficiency to power (as would be appropriate for desktop devices) (and they could choose frequencies that still leave them more efficient than comparable x86 chips).
Last edited:
Given that the M1 was built off A14 and M2 off the A15, wouldn't it be more logical to build M3 off A16 instead of a yet to be announced A17 variant?
A16 had a little to no IPC improvement and GPU is the same as it was on A15. If M3 is based on A16 it will more of a failure than M2.
That means we don't get new GPUs cores nor Ray Tracing cores and a 10% faster CPU.
I have this belief that M3 or M4 will use P cores and H cores: H core design will support efficiency mode and also be able to bank in extra renames and EUs so that it can run near P levels. The M series will only have a couple H cores per SoC while the A series will use a couple of H cores where it might normally have a couple P cores.