M3 Chip Generation - Discussion Megathread

[leman]

Well gaming is a sustained workload, its an use case where you want the performance

Yes, and fortunately A17 arrears to have good sustained performance at reasonably low power too.

and I would say people charging the phone and using it at the same time can get thermal throtling and battery will stop charging. I would say in summer recording videos it will happen too. A guy that is constantly loading tweets, instagram posts and Tik toks, just by constantly hitting more than 4.2W the thermal throttle will end happening

Hot weather operation is needed where I anticipate some issues. Continuous internet use probably not - while there will be some performance fluctuation you probably won’t notice it at all.

 

[leman]

A passively cooled M2 laptop only ties the desktop CPU in a single core usage, in multi core It smokes It, and on sustained loads even more

Well of course.

 

[PgR7]

Yes, and fortunately A17 arrears to have good sustained performance at reasonably low power too.



Hot weather operation is needed where I anticipate some issues. Continuous internet use probably not - while there will be some performance fluctuation you probably won’t notice it at all.

If you limit the power of the A17 Pro to the same of the A16, in Geekbench 5 looking at the Geekerwan's numbers you would get these figures:

A16: ~5600 points
A17 Pro: ~5850 points

4% improvement, so they knowing that Geekbench doesnt throttle they decided to increase power to 14W to squeeze 400 extra points to get a 11-12% extra performance. In Genshin Impact its 9% more efficient, the problem is that those 14W spikes makes the energy savings go down the drain

 

[jeanlain]

Passmark is a famously bad microbenchmark that doesn’t predict real-world performance very well. A17 Pro probably is particularly favored thanks its wider int backend. Doesn’t change the fact that A17 is very impressive of course.

The results are consistent with Geekbench's, except that AMD CPUs perform a bit less well.

 

[Gudi]

Well, I don't know enough to really understand what it means. But back when the M1 was released, an 8-wide Instruction Decoder (up from 4 on Intel/AMD) was touted as a major achievement and reasons for its amazing parallel performance at incredibly low power consumption. Apple shipping a 9-wide decoder is awesome!

 

[quarkysg]

Well, I don't know enough to really understand what it means. But back when the M1 was released, an 8-wide Instruction Decoder (up from 4 on Intel/AMD) was touted as a major achievement and reasons for its amazing parallel performance at incredibly low power consumption. Apple shipping a 9-wide decoder is awesome!

Apple most probably profiled tons of App Store codes and likely found that making the decoder wider benefits the majority of iOS and macOS apps. Maybe not so much for Geekbench. Heh heh.

 

[dmccloud]

That has been AMD’s and Intel’s excuse for years and we gave them crap for it. Now, Apple does it and we can’t give them a pass.

If you're gonna quote me, at least take the full statement rather than one sentence fragment.

 

[Xiao_Xi]

Yes, and fortunately A17 arrears to have good sustained performance at reasonably low power too.

It's a shame that Apple couldn't improve the iPhone's thermal dissipation to maintain peak performance for longer.

 

[altaic]

It's a shame that Apple couldn't improve the iPhone's thermal dissipation to maintain peak performance for longer.

View attachment 2279620

A graph with no time scale, nice.

 

[PgR7]

It's a shame that Apple couldn't improve the iPhone's thermal dissipation to maintain peak performance for longer.

View attachment 2279620

The thing is that when the performance is around 44 fps the phone is consuming 10W power, with simple math, in the 15 Pro Max the battery would last less than 2 hours, with 4w you play at about 30 fps and you have 4 hours battery life in an entire battery cycle of the 15 Pro Max and 3 hours in the 15 Pro

 

[APCX]

It's a shame that Apple couldn't improve the iPhone's thermal dissipation to maintain peak performance for longer.

View attachment 2279620

I’m confused. REV isn’t going to be available for the 14 Pro Max. How was this tested?

 

[leman]

It's a shame that Apple couldn't improve the iPhone's thermal dissipation to maintain peak performance for longer.

View attachment 2279620

I don’t see much point in running 10 fps higher if it’s going to halve your gaming battery. That said, I’m sure that we will soon see “gaming stations” for the iPhone that offer some additional cooling.

 

[APCX]

I don’t see much point in running 10 fps higher if it’s going to halve your gaming battery. That said, I’m sure that we will soon see “gaming stations” for the iPhone that offer some additional cooling.

I also wonder if using MetalFX would enable more consistency? I don’t know if that was enabled on this test.

 

[Confused-User]

Well gaming is a sustained workload, its an use case where you want the performance, and I would say people charging the phone and using it at the same time can get thermal throtling and battery will stop charging.

This is definitely true in a limited set of cases (mostly gaming), and an interesting area where Apple could make some improvements. It would be nice to be able to say "limit charging to 5W" or "limit charging to maintain current battery %age" or "charge if battery falls below X%". It would also be nice to be able to say "limit performance to allow thermal headroom for [5w charging | maintain battery | fast charging]". I don't know that Apple will ever do this, though, as the potential for confusing the crap out of most people is pretty high. Exposing APIs to control this so a third party could write a sophisticated power/energy tool would be a decent compromise but quite uncharacteristic for Apple.

I would say in summer recording videos it will happen too. A guy that is constantly loading tweets, instagram posts and Tik toks, just by constantly hitting more than 4.2W the thermal throttle will end happening

Lol. Perhaps, but unlikely, and even if it does and the phone throttles down by 50% the user still won't notice.

I say it's unlikely because social media use, except for video/audio (which is handled in large part by hw decoders with minimal energy use), is a classic race-to-idle situation.

 

[Confused-User]

BTW for those wondering at the low IPC gains: It's plausible (and I think likely, based on everything discovered in the last few days) that Apple gained meaningful IPC from their architecture changes, and gave most of those gains up in order to buy more clock headroom. The obvious possibility would be to lengthen the pipeline, though I'm not suggesting that that's what they actually did - they could have done anything, and it would have depended on where in the core the critical paths are. (Paths, because those can change as clocks go up.)

 

[Xiao_Xi]

I’m confused. REV isn’t going to be available for the 14 Pro Max.

It wiil be available on October 30th.

Resident Evil Village for iPhone, iPad launches October 30

The iPhone and iPad version of Resident Evil Village will launch on October 30, Capcom announced.

www.gematsu.com

How was this tested?

It's the testflight version.

I don’t see much point in running 10 fps higher if it’s going to halve your gaming battery.

You are not the only one. In fact, the game is locked to 30 fps, at which the iPhone consumes about 4W. But the reviewer hacked the game and unlocked the fps to test the A17's peak performance.

BTW for those wondering at the low IPC gains:

Dougall made a comment about the IPC increase on A17.

With the clock speed increase, the IPC gain looks bizarrely slim: 33% more integer units turns into ~3.5% more IPC.

I suspect there are two issues at play. The first is that increasing clock speed decreases IPC. Cache-misses take more cycles, and there's less per-cycle memory bandwidth. The other is that A14-A16 were already very wide, and a lot of code was already latency-bound, so the extra integer units sometimes just don't help.

Dougall (@dougall@mastodon.social)

With the clock speed increase, the IPC gain looks bizarrely slim: 33% more integer units turns into ~3.5% more IPC. I suspect there are two issues at play. The first is that increasing clock speed decreases IPC. Cache-misses take more cycles, and there's less per-cycle memory bandwidth. The...

mastodon.social

 

[leman]

BTW for those wondering at the low IPC gains: It's plausible (and I think likely, based on everything discovered in the last few days) that Apple gained meaningful IPC from their architecture changes, and gave most of those gains up in order to buy more clock headroom. The obvious possibility would be to lengthen the pipeline, though I'm not suggesting that that's what they actually did - they could have done anything, and it would have depended on where in the core the critical paths are. (Paths, because those can change as clocks go up.)

It’s interesting to note that some SPEC subtests do show a substantial improvement. It’s a shame SPEC is not publicly available, I’d like to have a look at the results. The reviewers only show some point estimates but it’s the distribution that’s really interesting.

 

[APCX]

It wiil be available on October 30th.

Resident Evil Village for iPhone, iPad launches October 30

The iPhone and iPad version of Resident Evil Village will launch on October 30, Capcom announced.

www.gematsu.com

It's the testflight version.


You are not the only one. In fact, the game is locked to 30 fps, at which the iPhone consumes about 4W. But the reviewer hacked the game and unlocked the fps to test the A17's peak performance.


Dougall made a comment about the IPC increase on A17.

Dougall (@dougall@mastodon.social)

With the clock speed increase, the IPC gain looks bizarrely slim: 33% more integer units turns into ~3.5% more IPC. I suspect there are two issues at play. The first is that increasing clock speed decreases IPC. Cache-misses take more cycles, and there's less per-cycle memory bandwidth. The...

mastodon.social

It isn’t available for the 14 Pro Max, just the 15 Pro/Max and M1. That’s my point. How can they have figures for that?

 

[Xiao_Xi]

How can they have figures for that?

The game works on older iPhones, but has some crashes on older phones because they don't have enough RAM for the game.

 

[APCX]

The game works on older iPhones, but has some crashes on older phones because they don't have enough RAM for the game.
View attachment 2279754

Interesting. Weird given they don’t support it. Wonder if that will change once it’s released.

 

[name99]

It’s interesting to note that some SPEC subtests do show a substantial improvement. It’s a shame SPEC is not publicly available, I’d like to have a look at the results. The reviewers only show some point estimates but it’s the distribution that’s really interesting.

Are the spec results PRINTED somewhere? Or only some YouTube video?

 

[leman]

Are the spec results PRINTED somewhere? Or only some YouTube video?

Unfortunately only in the video AFAIK. What’s sad is that the test wrapper saves the sampled CLPC counter data as a csv, but they don’t make it available. If someone here happens to be in contact with the Geekerwan team, maybe they can ask them to upload the data?

 

[scottrichardson]

BTW for those wondering at the low IPC gains: It's plausible (and I think likely, based on everything discovered in the last few days) that Apple gained meaningful IPC from their architecture changes, and gave most of those gains up in order to buy more clock headroom. The obvious possibility would be to lengthen the pipeline, though I'm not suggesting that that's what they actually did - they could have done anything, and it would have depended on where in the core the critical paths are. (Paths, because those can change as clocks go up.)

Another member commented a couple pages ago breaking down the changes for the p cores and e cores and essentially explained that the ipc boost was likely hamstrung by the clock boost. Something about cache hit misses increase as clocks go up.

 

[caribbeanblue]

Lorem ipsum dolor sit amet, consectetur adipiscing elit.

 

[picpicmac]

What is limiting the amount of RAM with SoC chips?

Real estate. One of the reasons to go to ever small devices on silicon is to squeeze in more transistors and hence more storage on the chip. That's why some are speculating (note: speculating, not reporting, for all those rumor monger types out there) that the basic M3 will come with 12GB of RAM (and hence the base iMac will have 12GB RAM.)

Personally I'm quite fine with that. To me RAM is more important, as evidently virtual machines (VM) are going to be more likely to be encountered going forward for both home and business uses.