Here's How Much Faster the A18 Pro Chip is Compared to the A17 Pro

[splifingate]

Having had a recent SE and a 13 Mini, I can say that you lose a LOT of nice things by going to the SE. MagSafe being one of those things, which means no MagSafe battery pack, no MagSafe charging (yeah it has wireless but who knows how well you placed the phone on the charger), way worse camera, smaller screen despite being physically larger, screen doesn't do anywhere near as good black, not as bright, smaller battery capacity, no faceid, etc.


It's a shame apple killed the mini, hopefully eventually that's what the "SE" turns into with a reasonably current chip, because the mini was/is SO much better.

The Mini showed what is actually possible.

I appreciate larger screens (run three 4K's on my M2); but, I don't want to lug all-that-around on the daily

I don't use the Mag thingy, but I do use all the rest, every day.

 

[throAU]

I appreciate larger screens (run three 4K's on my M2); but, I don't want to lug all-that-around on the daily

The mini is physically smaller than the SE with a larger screen because it doesn't have the touchID button.

 

[Confused-User]

Essentially this article should state that the A-series chip took 4 years to beat the M-series chip rather than place the M-series chips in the iPhone Pro. Great achievement, but the iPad has been given the bigger and faster CPU and GPU updates and the iPhone has been incremental at best. Those are facts.

No, that's ridiculously and obviously false. CPU performance in A and M series chips is basically matched.

The CPU performance differences between A and M chips comes down, in the general case, largely to core counts. Turn off two of the P cores on the M3 and see how close you get to A17 performance. The rest of the difference comes from clocks, and (for longer-duration tasks) heat dissipation capacity.

This is completely clear when you compare single-core performance (which is far more important for phones, and for Mac users not doing some specialized tasks like video, simulations, etc.). For example, the M2 MBA SC GB scores ~2600, while the same year's A16 is the same, and the next year's A17 scored ~2900. That was in turn matched against the M3, which at ~3100 was roughly the same when you adjust for the M3's slightly higher clocks.

This year's M4 (so far only in the iPad) scores higher than the A18, again because of clocks.

The situation with the GPU is comparable.

 

[chucker23n1]

Essentially this article should state that the A-series chip took 4 years to beat the M-series chip rather than place the M-series chips in the iPhone Pro. Great achievement, but the iPad has been given the bigger and faster CPU and GPU updates and the iPhone has been incremental at best. Those are facts.

The iPad Pro has higher-end performance than the iPhone Pro, but that’s largely for thermal reasons.

In terms of relative improvement compared to previous generations, they’re on the same trajectory, as they use the same cores, just different counts and at different clocks.

 

[Confused-User]

In terms of relative improvement compared to previous generations, they’re on the same trajectory, as they use the same cores, just different counts and at different clocks.

We agree... so far.

But while as best we can tell that's been true up to this year, it's not clear that it will remain true going forward. One of the big mysteries that will likely take a while to untangle (if it ever is) is whether or not there is a difference between the cores on the iPP's M4, the A18/A18Pro, and the M4 Macs coming in October (or will they be M5s??). Though once the Macs are here, it will at least get somewhat easier.

Obviously it's easiest for them to just design cores once and then use them throughout the product line, but they do a bunch of hard (and expensive) things already. They may decide it's worth doing a few more.

 

[chucker23n1]

We agree... so far.

But while as best we can tell that's been true up to this year, it's not clear that it will remain true going forward. One of the big mysteries that will likely take a while to untangle (if it ever is) is whether or not there is a difference between the cores on the iPP's M4, the A18/A18Pro, and the M4 Macs coming in October (or will they be M5s??). Though once the Macs are here, it will at least get somewhat easier.

Obviously it's easiest for them to just design cores once and then use them throughout the product line, but they do a bunch of hard (and expensive) things already. They may decide it's worth doing a few more.

I cannot imagine them designing new cores for the iPhone non-Pro. I can see them making cores specifically for the higher-end Macs, and of course for low power budgets, such as for AirPods and the Apple Watch (which currently just uses the iPhone's e-cores). I think they enjoy their strategy where mostly the same core designs just scale up and down. But it does come with the catch that, for example, they can't compete with high-end desktops. Maybe an improved scheduler will get them closer.

I also expect the upcoming Macs to be M4s, and for the A18 Pro to be the same cores as the M4, just at a lower clock, with fewer of them, and of course some SoC features such as Thunderbolt removed.

Where I'm less sure is indeed the A18 non-Pro. I'm not sure having one fewer GPU core and less overall cache alone explain the benchmark differences.

 

[Confused-User]

I cannot imagine them designing new cores for the iPhone non-Pro. I can see them making cores specifically for the higher-end Macs, and of course for low power budgets, such as for AirPods and the Apple Watch (which currently just uses the iPhone's e-cores). I think they enjoy their strategy where mostly the same core designs just scale up and down. But it does come with the catch that, for example, they can't compete with high-end desktops. Maybe an improved scheduler will get them closer.

I also expect the upcoming Macs to be M4s, and for the A18 Pro to be the same cores as the M4, just at a lower clock, with fewer of them, and of course some SoC features such as Thunderbolt removed.

Where I'm less sure is indeed the A18 non-Pro. I'm not sure having one fewer GPU core and less overall cache alone explain the benchmark differences.

What makes you say they can't compete with HEDT? There is every indication that Apple's cores are dominant and would rule in any segment Apple chooses to play in. If you mean the OS isn't up to managing that many cores... I haven't seen any evidence on that one way or the other.

As for M4 vs. M5 - I just meant that it's conceivable that Apple will call the October Macs M5s. I don't think they will, but it's not out of the question.

 

[chucker23n1]

What makes you say they can't compete with HEDT? There is every indication that Apple's cores are dominant and would rule in any segment Apple chooses to play in. If you mean the OS isn't up to managing that many cores... I haven't seen any evidence on that one way or the other.

I’m saying Apple doesn’t currently seem interested in making a Threadripper competitor. Which is just as well; it’s not a huge market.

I brought this up because you mentioned different core designs. They could make a core design that scales differently: useless for phones, but higher-end on Macs. Heck, they could make a logic board approach that isn’t a SoC, if they wanted to.

Separately from that, they could also increase the p-core count on the Ultra, but I think that’s getting close to diminishing returns. They could mitigate it with

• a boost approach where some cores get shut down, and others run at a higher clock
• a better scheduler
• generally running the cores at a higher clock than on laptops

As for M4 vs. M5 - I just meant that it's conceivable that Apple will call the October Macs M5s. I don't think they will, but it's not out of the question.

Right.

It’ll be interesting to see how they approach the Air. Do they ship an M4 Air almost a year after the iPad Pro got the M4? Or do they perhaps wait for the M5 so the jump feels more meaningful?

 

[Confused-User]

I’m saying Apple doesn’t currently seem interested in making a Threadripper competitor. Which is just as well; it’s not a huge market.

I brought this up because you mentioned different core designs. They could make a core design that scales differently: useless for phones, but higher-end on Macs. Heck, they could make a logic board approach that isn’t a SoC, if they wanted to.

Separately from that, they could also increase the p-core count on the Ultra, but I think that’s getting close to diminishing returns. They could mitigate it with

• a boost approach where some cores get shut down, and others run at a higher clock
• a better scheduler
• generally running the cores at a higher clock than on laptops

Oh, I see what you're saying. You misread me, but my fault for not being clearer.

I didn't mean that I expected them to build different types of cores in the same gen, aside from the obvious P/E they're already doing... though it's not inconceivable they might have a somewhat different E-ish core type segregated for the kernel.

What I meant was, I don't think we can expect that all new innovations will come out once a year, in a single core, that will then be used until the next gen core comes a year later. Rather, you may see a new core with the A19, then the next month a similar core but with a few miscellaneous improvements in the M5, followed three months later by even more improvements in the M5 Pro/Max. (This is an example timeline, not a prediction!!)

It’ll be interesting to see how they approach the Air. Do they ship an M4 Air almost a year after the iPad Pro got the M4? Or do they perhaps wait for the M5 so the jump feels more meaningful?

Very good question. If the Air were coming next month, then it would be easy to call, but if they wait until next year, they really could have a new design ready... or almost ready, tempting them to wait a few more months. Which would really suck.

 

[chucker23n1]

Oh, I see what you're saying. You misread me, but my fault for not being clearer.

I didn't mean that I expected them to build different types of cores in the same gen, aside from the obvious P/E they're already doing... though it's not inconceivable they might have a somewhat different E-ish core type segregated for the kernel.

What I meant was, I don't think we can expect that all new innovations will come out once a year, in a single core, that will then be used until the next gen core comes a year later. Rather, you may see a new core with the A19, then the next month a similar core but with a few miscellaneous improvements in the M5, followed three months later by even more improvements in the M5 Pro/Max. (This is an example timeline, not a prediction!!)

Gotcha. So you’re saying they’d do micro-generations and accelerate the timeline that way, but eventually, the changes would still trickle down?

We’ve kind of seen that with the SoC already (but not its CPU cores). For example, the M1 Pro added more advanced video encoding; the M2 then inherited that.

For the cores, I wonder if that’s worth the investment. Maybe to add a new instruction to higher-end SoCs first?

Very good question. If the Air were coming next month, then it would be easy to call, but if they wait until next year, they really could have a new design ready... or almost ready, tempting them to wait a few more months. Which would really suck.

Yep. The Air shipped in March, so another update this year would be unusual. (Although not unprecedented. There was the March 2020 Ice Lake Air, then the November 2020 M1 Air.)

 

[throAU]

Gotcha. So you’re saying they’d do micro-generations and accelerate the timeline that way, but eventually, the changes would still trickle down?

We’ve kind of seen that with the SoC already (but not its CPU cores). For example, the M1 Pro added more advanced video encoding; the M2 then inherited that.

For the cores, I wonder if that’s worth the investment. Maybe to add a new instruction to higher-end SoCs first?

I don't think they'll do any of that.

I think what apple want are well tested building blocks and an architecture that scales by linking said blocks together. Which they have, with the P and E cores and their GPU cores.

The blocks go into mobile devices first where the resulting dies are smaller and less mission critical, before they scale up and put into devices used for more complex tasks, running less trusted/controlled code.

IMHO effort spent tweaking cores within a generation is wasted effort that could be better spent on the next P/E core design for the next generation of parts. All you're doing there is complicating the drivers/software stack for what? Minor intra-generational improvement that you're taking development resources away from the next generation of development to achieve.

 

[Confused-User]

Gotcha. So you’re saying they’d do micro-generations and accelerate the timeline that way, but eventually, the changes would still trickle down?

We’ve kind of seen that with the SoC already (but not its CPU cores). For example, the M1 Pro added more advanced video encoding; the M2 then inherited that.

For the cores, I wonder if that’s worth the investment. Maybe to add a new instruction to higher-end SoCs first?

Yes, see below.

It's not so much about accelerating the timeline, as much as it is rolling out new features when they're ready.

I don't think they'll do any of that.

I think what apple want are well tested building blocks and an architecture that scales by linking said blocks together. Which they have, with the P and E cores and their GPU cores.

The blocks go into mobile devices first where the resulting dies are smaller and less mission critical, before they scale up and put into devices used for more complex tasks, running less trusted/controlled code.

IMHO effort spent tweaking cores within a generation is wasted effort that could be better spent on the next P/E core design for the next generation of parts. All you're doing there is complicating the drivers/software stack for what? Minor intra-generational improvement that you're taking development resources away from the next generation of development to achieve.

I'm not predicting, I'm reporting. They *do* do that. The recent example I have to hand is the E core in the M3 Max, vs. the E core in the Pro and base M3. It's not the same core.

Your argument is a sensible one, but clearly Apple has a different viewpoint, at least sometimes.

Most changes to cores should not require changes to the OS or drivers, though that won't always be true. (For example, some change might make altering DVFS useful.)

 

[splifingate]

The mini is physically smaller than the SE with a larger screen because it doesn't have the touchID button.

I gave AAPL USD1400 for the privileged of owning the 12-mini, which--in my honest opinion--directly represents the most-profound evolution in AAPL History.

The 13-Mini (I now own/use) is a logarithmic variation on this evolution; one with which I am 97% satisfied.

I neither suffer the OLED issues of my SO; nor am I wedded to the antiquated bio-metrics of yester-yore

IMO, the SE and the Mini are absolutely, and entirely, different creatures.