Thread: iPhone 15 / Apple Watch 9 Event

[Citysnaps]

You’re a braver man than I

It is entertaining! Especially the discussion about Apple selling new iPhone cables.

 

[Jimmyjames]

You’re a braver man than I

Lol. Also helps if you haven’t been banned.

 

[Jimmyjames]

Hopefully we’ll also see what perf/W looks like which is kinda more important as it can tell us what the headroom is and how consistently it can hit its performance

Good poin.

 

[dada_dave]

It is entertaining! Especially the discussion about Apple selling new iPhone cables.

Eoof … no … not going … there’s a cap on how much aggravation I can endure

Lol. Also helps if you haven’t been banned.

For myself, more of a self-ban. While I wouldn’t describe these forums as an echo chamber, the quietude in comparison to MR is appreciated by my blood pressure. Keeping track of the Ukraine war on Twitter and wading through the trolls there really puts people being querulous over technology and gadgets in perspective and does enough damage to my calm by itself.

 

[Jimmyjames]

I think it’s important to remember that those disappointed are that way because they are concerned there has been little improvements in the architecture for a few generations now, rather than the improvements in scores. It’s very possible they are wrong, or missing some nuance, but that does seem to be the general view: process shrink/clock increases.

 

[dada_dave]

I think it’s important to remember that those disappointed are that way because they are concerned there has been little improvements in the architecture for a few generations now, rather than the improvements in scores. It’s very possible they are wrong, or missing some nuance, but that does seem to be the general view: process shrink/clock increases.

Absolutely and I don’t think those concerns are unfounded at all. But I remember even when Apple introduced the M1 a lot of people wondered, where do you go from here in terms of single core performance? It’s already so wide … I also remember an AMD engineer discussing “performance” with Anandtech or another such publication mentioning that while IPC is the sexy headline, in the end it’s not everything that matters. Which is why I’m more interested in perf/W. IPC gives you a good idea of where that might be, but isn’t the whole story.

 

[Jimmyjames]

Ok I give up! 15 Pro Max, much slower single core (running at a lower frequency) but 8 cores and therefore higher multi core! None of this makes sense to me.

iPhone16,2 - Geekbench

Benchmark results for an iPhone16,2 with an ARM processor.

browser.geekbench.com

 

[exoticspice1]

T

Ok I give up! 15 Pro Max, much slower single core (running at a lower frequency) but 8 cores and therefore higher multi core! None of this makes sense to me.

iPhone16,2 - Geekbench

Benchmark results for an iPhone16,2 with an ARM processor.

browser.geekbench.com

This is running on an emulator

 

[Andropov]

I think it’s important to remember that those disappointed are that way because they are concerned there has been little improvements in the architecture for a few generations now, rather than the improvements in scores. It’s very possible they are wrong, or missing some nuance, but that does seem to be the general view: process shrink/clock increases.

I understand, but so far Apple has steadily managed to increase single core performance year over year, without a single failed year. Ultimately it does not matter if it's because of node shrinks or architectural redesigns. Maybe Apple sets a target performance / efficiency goal every year and, once it's met, improvement efforts go elsewhere (next year's chip, or the GPU, for example). IF next year they don't achieve the usual performance improvement, we can start to worrying. But this same discussion, almost word for word, happens every year!

Still, I do think we're still missing something about the A17 Pro. Apple mentioned wider architecture and decoding, and yet the improvements seem to be mainly due to clock increases. Isn't that weird?

 

[dada_dave]

The other side of the coin is let’s wait to get official scores first …

 

[Jimmyjames]

T

This is running on an emulator

Ok so my question is, how do you know this and how can we be sure the other scores are genuine as a result of that?

 

[exoticspice1]

Ok so my question is, how do you know this and how can we be sure the other scores are genuine as a result of that?

Scroll down till you see metal information

iPhone16,2 - Geekbench

Benchmark results for an iPhone16,2 with an ARM processor.

browser.geekbench.com

 

[dada_dave]

I understand, but so far Apple has steadily managed to increase single core performance year over year, without a single failed year. Ultimately it does not matter if it's because of node shrinks or architectural redesigns. Maybe Apple sets a target performance / efficiency goal every year and, once it's met, improvement efforts go elsewhere (next year's chip, or the GPU, for example). IF next year they don't achieve the usual performance improvement, we can start to worrying. But this same discussion, almost word for word, happens every year!

Still, I do think we're still missing something about the A17 Pro. Apple mentioned wider architecture and decoding, and yet the improvements seem to be mainly due to clock increases. Isn't that weird?

Agreed

Absolutely and I don’t think those concerns are unfounded at all. But I remember even when Apple introduced the M1 a lot of people wondered, where do you go from here in terms of single core performance? It’s already so wide … I also remember an AMD engineer discussing “performance” with Anandtech or another such publication mentioning that while IPC is the sexy headline, in the end it’s not everything that matters. Which is why I’m more interested in perf/W. IPC gives you a good idea of where that might be, but isn’t the whole story.

We do know that with wider architectures you eventually hit diminishing returns. I don’t know if this is a symptom of that or if there’s nuance left to be found.

 

[exoticspice1]

The other scores with 3.7ghz are legit

 

[Jimmyjames]

Scroll down till you see metal information

iPhone16,2 - Geekbench

Benchmark results for an iPhone16,2 with an ARM processor.

browser.geekbench.com

Many thanks.

 

[Alli]

- The f number is the same, let's say f/2.8 => Then, the two lenses have different aperture diameters. The lens for the large sensor has an aperture diameter of D = f / N = 35 / 2.8 = 12.5 mm, while the lens for the smaller sensor has an aperture diameter of D' = 17 / 2.8 = 6.07mm. The lens for the smaller sensor has a smaller aperture diameter, because it needs to collect less light for the same field of view, as the image circle is smaller, and the f number keeps the amount of light per unit area the same, in a smaller image circle.

How dare you use so much…MATH! I’ve a mind to give you a short ban for that. Maybe fx-32+1/4.2 hours. Or something. I dunno. I don’t do math.

There was a time when it was midnight, and it didn’t actually work sometimes until later.

I can remember setting my alarm for 1:58 a.m. Guaranteed to really piss off my Android-loving husband.

 

[Jimmyjames]

Nice work! Btw, this clearly shows what M. Handley wrote on MR earlier - the performance improvements are additive, not multiplicative. The gain from A16 to A17 is the same as from A13 to A14. So cancel that doom&gloom, people

Edit: may I post this chart over on MR?

…but the gains from the A12 to the A13, or A11 to A12. Wow. That’s the difference. After A13 everything slowed down comparatively. I suppose that’s to be expected.

 

[leman]

IF next year they don't achieve the usual performance improvement, we can start to worrying. But this same discussion, almost word for word, happens every year!

Next time I hope we will remember that CPU performance improvements grow additively instead of multiplicatively (which makes much more sense anyway). We are just so used to exponential growth in the GPU land because one constantly increases the number of cores (the performance of individual cores doesn't change much, because it can't anyway)

Still, I do think we're still missing something about the A17 Pro. Apple mentioned wider architecture and decoding, and yet the improvements seem to be mainly due to clock increases. Isn't that weird?

Some of the GB subtests show quite substantial improvements, especially more branchy ones or some that rely on a lot of computation. Once can certainly see an increase in IPC, it will just vary across workloads.

…but the gains from the A12 to the A13, or A11 to A12. Wow. That’s the difference. After A13 everything slowed down comparatively. I suppose that’s to be expected.

Did it though? A12 to A13 is ~400 points GB6, A16 to A17 is 380 points. Doesn't look like stuff has changed much.

In fact, A16 to A17 seems to be a bigger improvement than i9 11900K (Rocket Lake) to i9 12900KS (Alder Lake) with 360 points — and that was considered the best generation-to-generation jump in single-core performance since the Core 2 Duo. I think Apple is doing just fine.

 

[Eric]

How dare you use so much…MATH! I’ve a mind to give you a short ban for that. Maybe fx-32+1/4.2 hours. Or something. I dunno. I don’t do math.

Me, whenever these guys start talking specs and math.

 

[Jimmyjames]

Next time I hope we will remember that CPU performance improvements grow additively instead of multiplicatively (which makes much more sense anyway). We are just so used to exponential growth in the GPU land because one constantly increases the number of cores (the performance of individual cores doesn't change much, because it can't anyway)



Some of the GB subtests show quite substantial improvements, especially more branchy ones or some that rely on a lot of computation. Once can certainly see an increase in IPC, it will just vary across workloads.



Did it though? A12 to A13 is ~400 points GB6, A16 to A17 is 380 points. Doesn't look like stuff has changed much.

In fact, A16 to A17 seems to be a bigger improvement than i9 11900K (Rocket Lake) to i9 12900KS (Alder Lake) with 360 points — and that was considered the best generation-to-generation jump in single-core performance since the Core 2 Duo. I think Apple is doing just fine.

Ahh I probably miscounted!