Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Other A17 Pro. A new step forward.

mprime_17 said:
That's the bad part. Apple has been increasing CPU performance by just increasing the clocks in a similar % two years in a row.
To me this is a clear confirmation that the engineers leaving Apple a few years ago was a huge blow to Apple Silicon, and who knows if they'll ever recover.

Now, if the efficiency went up like with the A15, then it's a huge win IMO. But again, that didn't happen with the A16.
Click to expand...
As of now we don't have information on the architecture of the CPU cores. But if I had to guess, I would say that they are using the same microarchitecture as the A16 with a die shrink accounting for the extra CPU performance. But they built a whole new GPU for this generation. That was supposed to debut last year with the A16, but that chip had thermal and power issues and got delayed. That's why they kept the focus on the GPU. I think it might have been too risky for them to make a new CPU architecture and GPU architecture on the all-new 3nm process. https://www.theinformation.com/articles/inside-apples-war-for-chip-talent
 
zonai said:
As of now we don't have information on the architecture of the CPU cores. But if I had to guess, I would say that they are using the same microarchitecture as the A16 with a die shrink accounting for the extra CPU performance. But they built a whole new GPU for this generation. That was supposed to debut last year with the A16, but that chip had thermal and power issues and got delayed. That's why they kept the focus on the GPU. I think it might have been too risky for them to make a new CPU architecture and GPU architecture on the all-new 3nm process. https://www.theinformation.com/articles/inside-apples-war-for-chip-talent
Click to expand...
Exactly my thoughts, they ditched the new GPU architecture of the A16 and decided to implement it on the A17, making it the selling point of the A17, plus the small CPU overclock thanks to the node shrink.

I think the big redesign we’re all expecting will come with the A18 built on an entire new, more mainstream 3nm node, the N3E. I’m waiting for next year before concluding that Apple’s loss of talent is affecting their R&D.
 
Last edited:
  • Like
Reactions: SpotOnT
What I'm wondering is, what will the M3 look like? because M1 was an extension of the A14, and M2 was an extension of the A15, it would follow that the M3 is an extension of the A16. But the A16 is 5nm, and I feel like I've heard that the M3 chip will be 3nm. Does that mean M3 will be based on the A17 architecture? Or will it be A16 CPUs with the GPU from A17, as was originally intended for the A16?
 
zonai said:
What I'm wondering is, what will the M3 look like? because M1 was an extension of the A14, and M2 was an extension of the A15, it would follow that the M3 is an extension of the A16. But the A16 is 5nm, and I feel like I've heard that the M3 chip will be 3nm. Does that mean M3 will be based on the A17 architecture? Or will it be A16 CPUs with the GPU from A17, as was originally intended for the A16?
Click to expand...

M3 will be based on A17 Pro as it has already been validated for 3nm.

M3 will come out within half a year from today before April 2024.

M3 will be refreshed into these hardware before April 2024.

- MBA 13"/15"
- MBP 13"
- iMac
- iPad Pro

Mac mini will be refreshed with M3 & M3 Pro after April 2024 with the MBP 14"/16" M3 Pro & M3 Max.
 
Why would they skip A16-based M chip though? They've established a pattern of sorts with 2 years doing it that way. And the yields have to be atrocious once they scale up the A17 Pro to 24 cores like on the M2 Ultra.
 
zonai said:
Why would they skip A16-based M chip though? They've established a pattern of sorts with 2 years doing it that way. And the yields have to be atrocious once they scale up the A17 Pro to 24 cores like on the M2 Ultra.
Click to expand...
Economies of scale unless it is a limitation of the leading edge fab.

A chips are refreshed every 12 months.

M chips are refreshed every 19.5 months.

A16 Bionic is a 4nm part while the A17 Pro is 3nm.

M3 is likely to be a 3nm part unless Apple wants to reduce fab cost.

It would be lovely if Apple's M chips gets a refresh every 12 months. For the past 3 years it isn't the case.

Annual refresh cycle would be like this

- A chip: September
- M chip: Q1
- M Pro & Max: Q2
- M Ultra: Q3
 
zonai said:
Why would they skip A16-based M chip though? They've established a pattern of sorts with 2 years doing it that way. And the yields have to be atrocious once they scale up the A17 Pro to 24 cores like on the M2 Ultra.
Click to expand...
Also, there are many reports that the A16 was going to be what’s today the A17, so the A17 Pro is actually a tuned A16 with the GPU that was intended for the A16. The A16 we know it was never meant to exist, but the times of the new node, the delays, and the yield problems of the N3B process, lead Apple to Split the A16 into what today we know as the A16 Bionic and the A17 Pro. That’s according to some rumors.

Based on those rumors, if the A17 Pro is what it was really meant to be the A16, then it makes sense for the M3 to be based on the A17.

Also, what @sack_peak said. Due to the cycle times, there will be some A chips generations that won’t be used for the M family.
 
  • Like
Reactions: teh_hunterer and sack_peak
How does 5nm M1 chip compare to 3nm A17 Pro chip in terms of single core & multicore raw performance?

Are they at par or do we need wait for next year for that to occur?
 
sack_peak said:
M3 will be based on A17 Pro as it has already been validated for 3nm.

M3 will come out within half a year from today before April 2024.

M3 will be refreshed into these hardware before April 2024.
Click to expand...
This would be the most intuitive course of action.

But N3B is a very inefficient process, so Apple could exhaust it with A17, and use M3 to skip to N3E effectively predating the A18.

This would have the benefit of lowering M4 development costs, because if M3 came off N3B, it would need a full redesign for N3E next year.

By going straight to N3E, they could just do a M1>M2 thing for M3>M4 and keep it more linear.

But most importantly, let's remember that they really could go either way. Any guess is as good as the next with the information we have, not being insiders.
This situation really is unprecedented, so it cannot be deducted from historical patterns.
 
  • Like
Reactions: sack_peak
gpat said:
This would be the most intuitive course of action.

But N3B is a very inefficient process, so Apple could exhaust it with A17, and use M3 to skip to N3E effectively predating the A18.

This would have the benefit of lowering M4 development costs, because if M3 came off N3B, it would need a full redesign for N3E next year.

By going straight to N3E, they could just do a M1>M2 thing for M3>M4 and keep it more linear.

But most importantly, let's remember that they really could go either way. Any guess is as good as the next with the information we have, not being insiders.
This situation really is unprecedented, so it cannot be deducted from historical patterns.
Click to expand...
Exactly my thoughts. I think there’s a high chance of Apple using the N3E process for the M3 chips, as the N3B is pretty inefficient and troublesome, with very low yield rates. I really hope they use N3E and a new architecture for the M3, otherwise we’ll have to wait for the M4 to see a true jump in performance.

At the same time, there are many leaks pointing that the iphone 16 and 16 pro SoC will be both based on the N3E process node, both the regular A17/A18 and the A18 Pro. So hopefully we will see a much better SoC even on the base iPhone 16. Fingers crossed.

As you can read on my opening post of this thread, I was pretty excited and optimistic about the new 3nm A17 chip, but after the keynote, and the first benchmarks, I’m a bit disappointed. It feels like the A16 all over again.
 
Populus said:
Exactly my thoughts. I think there’s a high chance of Apple using the N3E process for the M3 chips, as the N3B is pretty inefficient and troublesome, with very low yield rates. I really hope they use N3E and a new architecture for the M3, otherwise we’ll have to wait for the M4 to see a true jump in performance.

At the same time, there are many leaks pointing that the iphone 16 and 16 pro SoC will be both based on the N3E process node, both the regular A17/A18 and the A18 Pro. So hopefully we will see a much better SoC even on the base iPhone 16. Fingers crossed.

As you can read on my opening post of this thread, I was pretty excited and optimistic about the new 3nm A17 chip, but after the keynote, and the first benchmarks, I’m a bit disappointed. It feels like the A16 all over again.
Click to expand...

About the iPhones, there is a real chance that we won't ever see a regular A17, and they'll go straight to A18 + A18 Pro all on N3E if they decide to go that way.

If they somehow find a way to milk the N3B process further, in that case we'll see regular A17, plus M3 family made off the N3B process, then M4 around 2025 from the new and improved N3E process.

I think we'll be stuck with 3nm for a looooong time, I believe until 2027 give or take, so maybe it makes more sense to squeeze every evolutionary step for all it's worth.

For that reason, I believe that early adopters of the 3nm node will get most of their money's worth, just like buyers of the first 5nm devices (iPhone 12 and Macbook M1) saw their products have the slowest obsolescence.
So M3 + iPhone 15 Pro for the win.
 
Last edited:
This thread has been super interesting and refreshing so far, so thank you to all who have been contributing. The A17 Pro chip was one of the things I was most looking forward to due to the efficiency gains of moving to a smaller node. I was pretty disappointed after the event. I don’t know that I have many use cases, if any at all, for the better GPU, though I know it’s important as a foundation for the M3 generation of chips. Given that there is usually a trade-off between performance and efficiency, I thought that the 10% increase in CPU would mean that there’d be big efficiency gains.

The new chip combined with:
All would point to the 15 Pro series having noticeably better battery life. Yet, according to Apple, it’s the exact same battery life of the 15 Pro series. I’m hoping that Apple just is being conservative and/or making up for the disappointing 14 Pro battery life compared to its claims. It just doesn’t make sense how battery life wouldn’t be better unless the new chip is actually more of a power suck. It also doesn’t really make sense that the 15 series has the exact same battery life as the 14 series. You’d think if Apple actually tested it, there would be some slight differences. I’m really interested for reviews next week to see if reviewers experience better battery life than expected.
 
  • Like
  • Love
Reactions: Chidoro, redbeard331, JulianL and 1 other person
gpat said:
About the iPhones, there is a real chance that we won't ever see a regular A17, and they'll go straight to A18 + A18 Pro all on N3E if they decide to go that way.

If they somehow find a way to milk the N3B process further, in that case we'll see regular A17, plus M3 family made off the N3B process, then M4 around 2025 from the new and improved N3E process.

I think we'll be stuck with 3nm for a looooong time, I believe until 2027 give or take, so maybe it makes more sense to squeeze every evolutionary step for all it's worth.

For that reason, I believe that early adopters of the 3nm node will get most of their money's worth, just like buyers of the first 5nm devices (iPhone 12 and Macbook M1) saw their products have the slowest obsolescence.
So M3 + iPhone 15 Pro for the win.
Click to expand...
Well, while I agree for the most part with your comment, I disagree with the last bit. I think that, if the N3E is much better than the N3B process, the big jump in performance will be seen on the A18/18Pro and the M3 family if the M3 are based on the new N3E process node.

So, at the moment, I don’t feel like the A17 Pro is that compelling, future-proof wise. I’d rather wait for the A18 and the N3E based SoC family.

EDIT: there are already YouTubers talking about the modest gains the 17Pro brings to the table, such as Luke Miani (I know, I know, but he does a good research, and I agree with what he says)


From SnazzyLabs they are a bit suspicious as well.
 
zonai said:
As of now we don't have information on the architecture of the CPU cores. But if I had to guess, I would say that they are using the same microarchitecture as the A16 ...
Click to expand...

No. Time index 1:02:38 in the Wanderlust presentation (
) says - "The new CPU features microarchitectural and design improvements for both our performance cores and efficiency cores". The slide that then appears seconds later on screen specifically mentions "improved branch prediction" and "wider decode & execute engines" for the high performance cores.
 
  • Like
Reactions: Populus
souko said:

iPhone16,1 - Geekbench

Benchmark results for an iPhone16,1 with an ARM processor.
browser.geekbench.com

This looks more like 20% up. And 10% IPC
Click to expand...
Yeah, to be fair that’s more than a decent jump in performance.

What if this generation has more variability regarding benchmarks, because of the poor yields of the A17? Maybe some chips like that one perform at a 20% faster speed while others perform just a 10% better. That’s the only logical reason why Apple would advertise that 10% increase in performance.

I once had an iPhone 13 mini with a strange A15 that had the equivalent in GPU performance to the 5-Core GPU of the 13 Pro, and it was an anomaly because I hadn’t seen that before in any other 13 mini. I guess those things can happen.
 
xxray said:
The new chip combined with:
Click to expand...

Qualcomm X70 continues to use 4nm just like X65. Even if it used 3nm, it would still encounter the same limitations of TSMC N3. The root cause is diminishing gains from new nodes.

Those UWB and LiDAR features aren’t used by most people and definitely not during Apple’s battery tests.

The display driver typically consumes 10% of the overall display system power. About 90% is from the panel backlight. That’s why those DDICs are made using mature nodes like 40nm and 28nm in the first place - because it doesn’t really matter.

The power situation comes down to the display, processor, and cellular/wireless. The other parts aren’t meaningful in terms of power consumption. Apple elected to boost clock by 10% and add 3 billion transistors on A17 Pro so there’s nothing left for power efficiency gains.
 
  • Like
Reactions: JulianL, mprime_17 and xxray
mprime_17 said:
Oh shheeeet

The consumption it's way too high for a phone. That's M1 iPad CPU consumption.
What the heck happened to N3 or apple for that matter?
Click to expand...

Physics happened along with TSMC marketing. The benefits of each node have been getting smaller and smaller.

People who weren't paying attention thought, "5nm to 3nm is a 40% reduction!" We'll see this story replay at M3.
 
JPack said:
Physics happened along with TSMC marketing. The benefits of each node have been getting smaller and smaller.

People who weren't paying attention thought, "5nm to 3nm is a 40% reduction!" We'll see this story replay at M3.
Click to expand...
N5 to N3 was marketed as 30% power reduction, but even accounting for N5P and N4 that brought power down a bit, you would expect at least a 10% reduction.

This just went the other way around, real bad. Especially on CPU, like damn! 40% more power for 10% more performance should be a no no for any company...
 
mprime_17 said:
N5 to N3 was marketed as 30% power reduction, but even accounting for N5P and N4 that brought power down a bit, you would expect at least a 10% reduction.

This just went the other way around, real bad. Especially on CPU, like damn! 40% more power for 10% more performance should be a no no for any company...
Click to expand...

The 10% power reduction is mutually exclusive to frequency increases. Apple chose a 10% clock increase.
 
  • Like
Reactions: ctjack
JPack said:
The 10% power reduction is mutually exclusive to frequency increases. Apple chose a 10% clock increase.
Click to expand...
Yes, it was either 30% power reduction or 10% more performance at same power vs N5.
This doesn't look like neither of them.

I was expecting a balance but damn, this is entirely unexpected tbh.
 
  • Like
Reactions: Populus
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back