Is this a Mi7chy alt account?
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.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
I picked up my iPhone pro today and ran a simple power and frequency test of my own making. The detailed data and description of methodology is here:
https://techboards.net/threads/thread-iphone-15-apple-watch-9-event.4299/post-146389
What might be most relevant for this thread is the following graph.
My experiments confirm Geekerwan report of A17 Pro using substantially more power for demanding workloads. But I think that people have been focusing to much on this instead of considering what this means for the Mac.
Apples 3N chip is not an efficiency powerhouse like some of us have hoped. At peak performance, its efficiency is comparable to A14/A15. But I think what we see here is Apple moving away from super efficient CPUs to super fast efficient ones. This is the first desktop-oriented design from Apple. A Mac based in this chip will probably consume around 8-10 watts per core at peak clock, but its performance will be state of the art enthusiast desktop. And while it might seem suboptimal for a phone, it can still outperform an A16 at lower power consumption, which is more than sufficient and faster than the closest competition.
And speaking of efficiency, A17 is not that bad either. It uses 30% power than M1 at the same performance level and is around 20% faster at the same power level. I think that’s fairly respectable.
Overall, I think the talk of talent drain at Apple is vastly exaggerated. They simply moved over to making desktop CPUs, it’s just that theirs scale down to phone level.
https://techboards.net/threads/thread-iphone-15-apple-watch-9-event.4299/post-146389
What might be most relevant for this thread is the following graph.
My experiments confirm Geekerwan report of A17 Pro using substantially more power for demanding workloads. But I think that people have been focusing to much on this instead of considering what this means for the Mac.
Apples 3N chip is not an efficiency powerhouse like some of us have hoped. At peak performance, its efficiency is comparable to A14/A15. But I think what we see here is Apple moving away from super efficient CPUs to super fast efficient ones. This is the first desktop-oriented design from Apple. A Mac based in this chip will probably consume around 8-10 watts per core at peak clock, but its performance will be state of the art enthusiast desktop. And while it might seem suboptimal for a phone, it can still outperform an A16 at lower power consumption, which is more than sufficient and faster than the closest competition.
And speaking of efficiency, A17 is not that bad either. It uses 30% power than M1 at the same performance level and is around 20% faster at the same power level. I think that’s fairly respectable.
Overall, I think the talk of talent drain at Apple is vastly exaggerated. They simply moved over to making desktop CPUs, it’s just that theirs scale down to phone level.
Where? They can only dream about the level of performance Apple has with A17.
Do you honestly believe Intel will manage to improve power efficiency by a factor of 6x in a single generation? Based on what?
True. And I am looking forward to the innovation in the personal computing space.
Qualcomm makes fast mobile graphics GPUs. Apple makes desktop GPUs. It’s a very different category. Qualcomm totally sucks at compute or when you need accurate calculations. They are fast because they hyper-focus on the simple usage case.
This just reinforces my belief that the updated CPU architecture was more made to shine on Mac (M3) than on iPhone (A17 Pro).
Well, you can test their testing methodology in english speaking version of their review.
It appears they stay with their numbers from initial review.
If you look at it closely, M1 is a slightly overclocked and horizontally scaled mobile phone chip. M2 doesn’t change much about it. Neither of these CPUs are able to take advantage of the thermal headroom of a modern desktop. I mean, Firestorm tops out at 5 watts. Even for a 14” laptop this is very low. Apple is just leaving performance on the table. It works reasonably well for their laptops thanks to their insane perf/watt, but they have no chance to compete on the desktop.
A17 instead appears to target a different segment in the power curve. I am quite certain it can go above 5 watts (in fact, it wouldn’t make any sense otherwise, because a Mac CPU will surely be faster than a phone, right)? So Apple seems to be going after that unused desktop thermal headroom. They use the new process not to slightly improve the performance at the same segment of the power curve but instead push beyond it, while retaining very good efficiency. It doesn’t make any sense for a phone. It makes all the sense in the world for a desktop computer.
As to performance estimates, well, that one is fairly straightforward I think. Geekerwan shows some SPaeC scores for A17, it is trading blows with Intels 5.8Ghz 12900K (that needs around 25-30 watts to get to this level of performance btw, so 7-8x more). I think M3 will reach around 4.2Ghz, some versions maybe slightly higher.
I get the same numbers for single score stress test using my own code, plus the testing framework produces the same results as powermetrics on the Mac, so I’m inclined to believe the results. Still don’t understand their MC results (will try to do a similar test this weekend).
So let’s assume that their numbers are correct. What conclusion do you draw from that? You already know mine.
I already said, multiple times. Apple is no longer eeking out every last bit of what physical design is giving them, and have to resort to increasing clock frequencies, while also increasing power draw.
It may be because they target desktop performance, but if I understand the consequences of the numbers Geekerwan provided - I would not hope for incredible performance from M3.
The CPU architecture - we know how performs, and at what power, per core. The only hope is that M3 series is on different N3 variant from TSMC and will improve efficiency.
Remember, guys. A17 Pro has 19 billion transistors, which is only 1 BLN less than M2, and 3 bln more than M1.
A17 uses 30% less power for the same frequency as A14 in my tests. What kind of improvement did you expect from 3N? Halving the power?
His benchmarks literally show a smartphone chip performing similar to an enthusiast class desktop CPU at 6x less power. I mean, what kind of expectations do you have? What would you consider “incredible performance“ at this point? Or is this “Apple sucks” unless they can outperform a 120W CPU using a 30W one?
I don’t get it. What’s wrong with just clocking this core higher?
The point your missing, is they are comparing Apple Silicon to the marketing information of future chips from Intel/AMD/Qualcomm with no critical judgment.
Have you ACTUALLY checked how much power per core recent AMD and Intel designs draw?
PER CORE. And have you payed attention to power per core in SPEC tests done by Geekerwan?
Do you understand the consequences of it?
Few links for you:
Welcome to AMD
AMD delivers leadership high-performance and adaptive computing solutions to advance data center AI, AI PCs, intelligent edge devices, gaming, & beyond.
28W design for 8C/16T CPU. Do math for yourself, and look at where AMD and Intel will be with next gen designs: Arrow Lake and Zen 5, which both will deliver 20% IPC increases.
Is this genuine?Have you ACTUALLY checked how much power per core recent AMD and Intel designs draw?
PER CORE. And have you payed attention to power per core in SPEC tests done by Geekerwan?
Do you understand the consequences of it?
Few links for you:
Welcome to AMD
AMD delivers leadership high-performance and adaptive computing solutions to advance data center AI, AI PCs, intelligent edge devices, gaming, & beyond.www.amd.com
28W design for 8C/16T CPU. Do math for yourself, and look at where AMD and Intel will be with next gen designs: Arrow Lake and Zen 5, which both will deliver 20% IPC increases.
Why is everything comparing next gen chips with current gen Apple Silicon chips? Why not compare actual chips that exist?
In any case, we all know that the M4 will destroy Arrow Lake and Zen 5.
Because A17 and M3 are next gen Apple chips.
And M3 at the earliest will be released in 2024, just like Zen 5, and Arrow Lake.
You are confusing multi core and single core. A 7840u draws about 10 watt per core at peak performance. It’s also 20-30% slower than A17 at 5 watts. If AMD delivers 25% improvement in IPC than the AMD’s core will be as fast as Apples smartphone. While still using two times more power.
They can advertise 28W TDP for that CPU because they drop the core clocks down by 40%. AMD cores have a wide power range and a flexible power curve. So their strategy to achieving good throughput per watt is to use many downclocked cores.
P.S. And yes, current AMD designs can reach similar efficiency in multicore compared to M2. Because AMD has more cores and can clock them lower, improving efficiency (having higher TDP doesn’t hurt either). It’s certainly an advantage AMD has, and I don’t see it changing any time soon. If you want MC performance for low price, AMD is the way to go.
Last edited:
TIL the A17, the chip people are CURRENTLY using is next gen. I believe Gurman is still speculating that the first M3 may come out in October.
Ok buddy.
While we debate endlessly about the future chips that may or may not bring three digit IPC increases, I took the liberty to interpolate the actual data from actual chips Geekerwan measured. Mind you: the interpolated curve goes through all the experimental points Geekerwan measured, it's not a loose fit. So while obviously the values in between are not accurate to the mW the qualitative shape of the actual performance curves can't look too different from this because this is how all performance curves look.
So for A14 Bionic, A15 Bionic, A16 Bionic and A17 Pro:
And if we add the Snapdragon 8 Gen 2, the only other chip on Geekerwan's measurements:
So the multicore performance curve of the Snapdragon 8 Gen 2 (2022, N4P) fits squarely between the A14 (2020, N5) and the A15 (2021, N5P).
So for A14 Bionic, A15 Bionic, A16 Bionic and A17 Pro:
And if we add the Snapdragon 8 Gen 2, the only other chip on Geekerwan's measurements:
So the multicore performance curve of the Snapdragon 8 Gen 2 (2022, N4P) fits squarely between the A14 (2020, N5) and the A15 (2021, N5P).
A17/M3-series SoCs suck, they will never be able to compete with the competitions chips from a decade in the future; thanks Tim Apple...
this would suggest that the achived win is quite close to the specs published by TSMC.
I really wonder to what frequenzy they can bring the CPU up on the M3.
Your estimation looks quite optimistic to me. Hope you are right
What i disagree with you is your 20% win over the M1, in that sense, that -imho- the M1 is nolonger relevant, for this type of discussion. 20% vs. the M1 also looks not exactly like much in my book.
(Though, i see, you took what you had at hand
)Take the M2max and its frequenzy. I think numbers would look then quite low. So, i see there sort of a mismatch, -optimism vs. pessimism-, somehow.
Your A17 graph stalls at some point. You seem to expect that this won´t be the case within a desktop M3 ?
Thanks everybody for insightful thoughts and posts !
Can´t wait to get a M3pro or max, haha
i´m probably one of the few ones who is entirely dependend on the raw single Core CPU speed. ( Not vs. a multicore application. its really just running on a single core / and i suck out all the SC juice of my M2pro, daily. Realtime-play sounddesign that is)
Why would Apple all of a sudden decided to prioritize Macs higher than iPhones - their milk cow? It's hard to believe that they would do it deliberately. Something must have gone wrong.
I don’t see it as prioritizing the Mac over the iPhone, but rather as making Apple Silicon better suitable to the needs of the Apple ecosystem.
Making a more performant CPU - even if it draws more power - does not harm the iPhone in the slightest. One just doesn’t run a phone at peak performance all the time, and there are many ways to compensate the higher peak power usage. Again, the A17 uses less energy for the same performance as its predecessors. But having more performance will definitely help in addressing some of the performance criticisms of Apple Silicon on the desktop.
How do you come to this conclusion? For me it just means that the A17 consumes more energy to achieve the yearly performance delta (not ratio) that Apple has been targeting since the A11 or so. The curves shown by @Andropov make it very clear.
This doesn't bode well for the Mac. There is less headroom to increase clock speed.
Last edited: