M3 Chip Generation - Discussion Megathread

[MRMSFC]

Considering that iOS is practically a black box in between the SoC and the benchmark software, could the inconsistency be in there?

If, for example, iOS isn’t using the AV1 decoding, and just running the relevant tests on the CPU, that would detrimentally effect the efficiency.

This is all really confusing.

 

[Andropov]

Agree with all of your points, however as far as I know they don't turn off the P-cores in Low Power Mode, but rather limit them to a very low clock speed (~1500MHz). If they turned off the P-cores it wouldn't be a very meaningful data point in this chart since you're not taking all of the cores and restricting their total power draw to a specific wattage (you're not taking advantage of all of the hardware). Though, we don't know if this behavior is the same on the 15 Pros as it has always been on previous iPhones, i.e if the efficiency cores exceed a certain ideal performance threshold Apple might just choose to turn off the P-cores and use only the E-cores in the future for further power savings in that mode. Geekerwan didn't say anything about the low power mode in that regard so I'm assuming it's the same (he did say that he experienced worse performance in Genshin in low power mode on the 15 Pro Max than on the 14 Pro Max towards the end of the video).

Maybe you’re right. Notebookcheck mentioned this behavior in an article, and this Reddit post said the same, I should have verified it further. It seemed plausible to me because 4 E cores should have (roughly) a multicore performance score in that ballpark.

 

[falainber]

According to wccftech, Snapdragon 8 gen 3 reaches 7400 in GB (MC).

 

[leman]

According to wccftech, Snapdragon 8 gen 3 reaches 7400 in GB (MC).

I'm not surprised. They have eight cores in that thing, five of which are throughput specialists (medium performance, medium power consumption). Never really understood that design philosophy. It's literally built to win multicore benchmarks. User utility? Zero.

 

[APCX]

According to wccftech, Snapdragon 8 gen 3 reaches 7400 in GB (MC).

Lol wccftech. Pretty sure they reported the A17 would hit 3900 gb single core. Also imagine boasting about your 8 core cpu matching a 6 core cpu. Oof

 

[koyoot]

It appears another reviewer reporting that 15 Pro and 15 Pro max have lower battery life than 14 pro and 14 Pro max.

And he bought those 14 pro and 14 Pro Max last week to test the battery life, and longevity of battery.

So it may be, that the tests done by the chinese reviewers were not wrong, especially when there is thermal throttling in sustained workloads over longer period of time.

 

[koyoot]

A15: 2P+4E and 5 GPU cores
M2: 4P+4E and 10 GPU cores
M2 Pro: 8P+4E and 19 GPU cores
M2 Max: 8P+4E and 38 GPU cores
M2 Ultra: 16P+8E and 76 GPU cores

A17 Pro: 2P+4E and 6 GPU cores
M3: 4P+4E and 10 GPU cores
M3 Pro: 8P+6E and 20 GPU cores
M3 Max: 12P+4E and 40 GPU cores
M3 Ultra: 24P+8E and 80 GPU cores

I really do wonder what Apple is doing with the M3 lineup. M3 doesn’t have double the GPU cores as A17 Pro.

M3 Pro has 2 extra efficiency cores but M3 Max has 4 extra performance cores but doesn’t have the 2 extra efficiency cores. I’m so confused.

Have you considered that what Gurman found was cut down SKUs?

And the actual lineup is: 6>12>24>48 GPU cores?

 

[APCX]

It appears another reviewer reporting that 15 Pro and 15 Pro max have lower battery life than 14 pro and 14 Pro max.

And he bought those 14 pro and 14 Pro Max last week to test the battery life, and longevity of battery.

So it may be, that the tests done by the chinese reviewers were not wrong, especially when there is thermal throttling in sustained workloads over longer period of time.

I mean it might also be that they are both wrong given Tech chap is not a particularly “tech” and geekerwan admitted they rushed it.

I want to make clear however, that does not mean either are wrong. It’s just too early to say. We need to wait for a substantial amount of evidence before we make judgment. It seems some here are desperate to proclaim the A17 as “bad” for whatever mysterious reasons.

 

[dmccloud]

I mean it might also be that they are both wrong given Tech chap is not a particularly “tech” and geekerwan admitted they rushed it.

I want to make clear however, that does not mean either are wrong. It’s just too early to say. We need to wait for a substantial amount of evidence before we make judgment. It seems some here are desperate to proclaim the A17 as “bad” for whatever mysterious reasons.

Reviews of people I've never even heard of aside, I think these battery claims are a bit overblown for a couple of reasons. First, the move to 3nm with A17 should mean the SoC itself draws less power than A16. Second, it has been confirmed by MKBHD and others that the Pro and Pro Max have slightly larger batteries than the 14 Pro/Pro Max. The other factor that could be affecting their battery life evaluations are system settings such as screen brightness, time to autolock, even network connectivity. When a phone is having issues connecting to either a WiFi or cellular network, it can actually contribute to quicker depletion of battery (this is something we see all too often here in Alaska, especially when not in one of the larger cities). I've even seen identical iPhone 14 Pros have wildly different battery life simply because they were using different cellular providers and one of the two had poor coverage at the location.

 

[Boil]

• M3 Extreme
• 4200 SC
• 69000 MC
• Nice
• ;^p

 

[JPack]

It appears another reviewer reporting that 15 Pro and 15 Pro max have lower battery life than 14 pro and 14 Pro max.

And he bought those 14 pro and 14 Pro Max last week to test the battery life, and longevity of battery.

So it may be, that the tests done by the chinese reviewers were not wrong, especially when there is thermal throttling in sustained workloads over longer period of time.

The converging body the evidence suggests Geekerwan's data is accurate. Apple increased battery capacity but did not mention anything about increased battery life.

Many people still seem to believe the myth that moving to 3nm must mean lower power even though we already saw the opposite with 4nm A16. Old beliefs die hard.

 

[koyoot]

Reviews of people I've never even heard of aside, I think these battery claims are a bit overblown for a couple of reasons. First, the move to 3nm with A17 should mean the SoC itself draws less power than A16. Second, it has been confirmed by MKBHD and others that the Pro and Pro Max have slightly larger batteries than the 14 Pro/Pro Max. The other factor that could be affecting their battery life evaluations are system settings such as screen brightness, time to autolock, even network connectivity. When a phone is having issues connecting to either a WiFi or cellular network, it can actually contribute to quicker depletion of battery (this is something we see all too often here in Alaska, especially when not in one of the larger cities). I've even seen identical iPhone 14 Pros have wildly different battery life simply because they were using different cellular providers and one of the two had poor coverage at the location.

How much less power draws 8W's from 5W's of power?


Smaller node means that you can LOWER THE VOLTAGE at the same frequency. Not that magically 5W's of power on 3 nm process will draw less power than 5W's on 5 nm process. You people clearly do not understand how energy works.

5W of power is 5W's of power regardless of process node.

Apple has increased the maximum frequency, while also increasing the maximum power draw for the SOC, which results in thermal throttling over longer period of time.

Lower voltage at the same frequency means: lower power draw while doing the same thing that is framerate limited to previous generation, which results in snappier feel of gaming on this phone over previous gen, but that is caused by higher UNDERUTILIZATION of components, rather than higher power efficiency.

But over longer period of time - the power draw, and battery drain, will be higher than previous generation.


And lastly. If A16 CPU was limited to 5W's of power, and A17 Pro is limited to 8W's of power - that is 60% power draw increase over previous generation. For 10% higher performance.

 

[leman]

It appears another reviewer reporting that 15 Pro and 15 Pro max have lower battery life than 14 pro and 14 Pro max.

And he bought those 14 pro and 14 Pro Max last week to test the battery life, and longevity of battery.

Do I understand correctly that his method of testing a battery involves dng various activities and then checking the remaining battery percentage? Its hardly what I’d call a scientific test.

The converging body the evidence suggests Geekerwan's data is accurate. Apple increased battery capacity but did not mention anything about increased battery life.

I am not convinced about claims of significantly lower battery life. I’d expect it to remain the same as the previous generation. The tests I‘ve seen so far did little to convince me otherwise.

Many people still seem to believe the myth that moving to 3nm must mean lower power even though we already saw the opposite with 4nm A16. Old beliefs die hard.

Well, that was a quite silly thing to believe to begin with. But I think we moved past that discussion a while ago. What is interesting now is whether A17 indeed uses more power than A16, what the power/performance curves are, and what this means for the Mac.

BTW, Geekerwan data suggests that A17 consumes 20% less power than A16 for the same GB5 MC score. I’d say that is a respectable improvement in efficiency.

 

[falainber]

I'm not surprised. They have eight cores in that thing, five of which are throughput specialists (medium performance, medium power consumption). Never really understood that design philosophy. It's literally built to win multicore benchmarks. User utility? Zero.

Multicore vs single core is an obvious tradeoff. Both have value in different scenarios. For example, one can argue that, on the mobile devices, single core tasks (UI etc.) are not bottlenecked anymore and the only tasks that need higher performance are multithreaded tasks (ML, image processing etc.) There is no way to say which combination of SC/MC performance is more optimal. To make such decision one would have to assign some priority ratings for both specs and this is a very subjective process. There is also an OS factor. In general, Android allows for more multitasking than iOS thus, one can argue, MC performance is more important for Android.

 

[falainber]

Do I understand correctly that his method of testing a battery involves dng various activities and then checking the remaining battery percentage? Its hardly what I’d call a scientific test.

Is there such a thing as scientific test for battery life testing? Most tech publications do exactly what this guy does.

 

[APCX]

The converging body the evidence suggests Geekerwan's data is accurate. Apple increased battery capacity but did not mention anything about increased battery life.

Many people still seem to believe the myth that moving to 3nm must mean lower power even though we already saw the opposite with 4nm A16. Old beliefs die hard.

Which converging body of evidence? lol

 

[falainber]

I mean it might also be that they are both wrong given Tech chap is not a particularly “tech” and geekerwan admitted they rushed it.

I want to make clear however, that does not mean either are wrong. It’s just too early to say. We need to wait for a substantial amount of evidence before we make judgment. It seems some here are desperate to proclaim the A17 as “bad” for whatever mysterious reasons.

You are not going to get test results from Apple. As far as "not quite tech" and rushed results are concerned... while they may be skewed for one reason or another, they are as likely to be skewed in favor of A17 as in favor of A16. Once we get more results, if all (or most) of them show A16 battery life is better the verdict will be out. We probably should have enough test results to make a judgement by the next week.

 

[APCX]

You are not going to get test results from Apple. As far as "not quite tech" and rushed results are concerned... while they may be skewed for one reason or another, they are as likely to be skewed in favor of A17 as in favor of A16. Once we get more results, if all (or most) of them show A16 battery life is better the verdict will be out. We probably should have enough test results to make a judgement by the next week.

Im not implying bias for these reviewers, but given the rush to be first, or meet a deadline for video reviews, the chances that they are judging/benchmarking while setup/background work takes place is very high. As you say, more evidence is required. It doesn't seem that what evidence exists, shows a drop in battery life, but we will see.

 

[leman]

And lastly. If A16 CPU was limited to 5W's of power, and A17 Pro is limited to 8W's of power - that is 60% power draw increase over previous generation. For 10% higher performance.

Where do you get these numbers from? Geekerwan measured 5.79W for A17 and 4.57W for A16. That’s 25% difference for 13% higher score. Not stellar of course, but far from what you are saying.

Besides, that’s peak, for short bursts only. As you said yourself, for longer work the phone will throttle and the power consumption will fall below A16 while delivering the same performance as A16 at its peak.

 

[leman]

Multicore vs single core is an obvious tradeoff. Both have value in different scenarios. For example, one can argue that, on the mobile devices, single core tasks (UI etc.) are not bottlenecked anymore and the only tasks that need higher performance are multithreaded tasks (ML, image processing etc.) There is no way to say which combination of SC/MC performance is more optimal. To make such decision one would have to assign some priority ratings for both specs and this is a very subjective process. There is also an OS factor. In general, Android allows for more multitasking than iOS thus, one can argue, MC performance is more important for Android.

Burst performance is important for a phone because it determines the responsiveness. MC is important for running low–priority tasks with low energy and distraction impact. But Qualcomm builds these things as if it were a server part. What do you need throughtout for in a smartphone? You are not going to run scientific workloads or use it as a render/build workstation.

Is there such a thing as scientific test for battery life testing? Most tech publications do exactly what this guy does.

If you want accurate results you’ll have to use an external power meter and measure the amount of energy put into the battery. Relying on the battery indicator probably has at least 5% margin of error.

 

[Xiao_Xi]

Chinese reviewers are on fire!

YouTuber Plays Resident Evil Village On iPhone 15 Pro Using External Monitor & DualSense Controller, Results Usher A New Age Of Mobile Gaming

A user connected his iPhone 15 Pro to an external monitor via USB-C and paired his DualSense to play Resident Evil Village on a bigger screen.

wccftech.com

 

[falainber]

Burst performance is important for a phone because it determines the responsiveness. MC is important for running low–priority tasks with low energy and distraction impact. But Qualcomm builds these things as if it were a server part. What do you need throughtout for in a smartphone? You are not going to run scientific workloads or use it as a render/build workstation.



If you want accurate results you’ll have to use an external power meter and measure the amount of energy put into the battery. Relying on the battery indicator probably has at least 5% margin of error.

Burst performance is not the same as single core performance. It could use multiple cores as well. Some tasks are inherently single cored. Primarily it's UI. But it's only true for the main UI thread. UI still can use worker threads for some tasks. "Responsiveness" is too vague of a term. For example, in games, the responsiveness can definitely benefit from MC performance.

 

[koyoot]

Burst performance is important for a phone because it determines the responsiveness. MC is important for running low–priority tasks with low energy and distraction impact. But Qualcomm builds these things as if it were a server part. What do you need throughtout for in a smartphone? You are not going to run scientific workloads or use it as a render/build workstation.



If you want accurate results you’ll have to use an external power meter and measure the amount of energy put into the battery. Relying on the battery indicator probably has at least 5% margin of error.

Responsivenes of a system is deterrmined by how much resources you have available for when the burst performance is needed.

The lower the utilization of GPU, at highest possible framerate, the lower the input lag - the higher the feeling of responsiveness.

The reason why A17 Pro is achieving higher efficiency and responsiveness in Genshin Impact tests is because... it has more resources available - lower GPU utilization under full load, at the same framerate. All of iOS games are either locked to 30, 60 or 120 FPS. Genshin is locked to 60 FPS, no matter the device you run it on.

 

[leman]

Burst performance is not the same as single core performance. It could use multiple cores as well. Some tasks are inherently single cored. Primarily it's UI. But it's only true for the main UI thread. UI still can use worker threads for some tasks. "Responsiveness" is too vague of a term. For example, in games, the responsiveness can definitely benefit from MC performance.

I am talking about latency sensitive work, where you want to minimize the time between an input and an output. That’s what “responsiveness“ means in this context. Layout computation, animation, that kind of stuff. Apple design is optimized for this as they give you two very fast cores. Qualcomm instead gives you one fast core and a bunch of progressively slower cores at different perf/watt sweet spots. I don’t see any point in this design except winning MC benchmarks. Not in a phone at least.

 

[Jamie I]

Have you considered that what Gurman found was cut down SKUs?

And the actual lineup is: 6>12>24>48 GPU cores?

He says

8, 18, 32, 64 for binned GPU cores
10, 20, 40, 80 for max GPU cores

He might be guessing some of these but he definitely did find M3 Pro with 18 GPU cores