M2 MacBook Air throttling is being vastly over-exaggerated

[alexonline]

I have heard the same arguments in other threads on MacRumors. “You should buy 16GB RAM instead of memory swapping.” “You should buy the 512GB model”

In the end of the day, I don’t want to pay those type of prices for a MacBook Air, I’d rather pay a bit more for the 14” then.

So best I can do is just wait for the M3 MBA rather than going into the same arguments again which has no point.

Wait away. The rest of us are doing whatever we want to do, without your permission, whether it is to buy M2, at whatever capacity we choose, while you do whatever you want, which appears to be to wait for M3, or buy an M1 Pro/Max MBP 14.

Or maybe you will wait for the M2 MBP 14/16, or maybe even the M3 MBP, almost like you have free will and choice. Amazing concept!

You’re not Neo in the Matrix, welcome to choices in life on Planet Earth.

Also, at the end of the day, it’s night.

 

[Mike Boreham]

I have only just found this thread, after posting my own Cinebench throttling tests with M1 and M2 here:

My Cinebench Throttling tests M2 vs M1 MacBook Airs

The key chart:

 

[Mike Boreham]

I have only just found this thread, after posting my own Cinebench throttling tests with M1 and M2 here:

My Cinebench Throttling tests M2 vs M1 MacBook Airs

I just repeated this test recording battery usage at the request of some one in the other thread (to compare with M1).

Cinebench score dropped 12.1% today compared to 14.7% yesterday....an indication of degree of repeatability.

FWIW machine used 23% battery for the 32 minutes at 100% CPU

 

[Mike Boreham]

I decided to do a two hour max throttle (100% on all cores) test to see how the machine behaves.

I used the only app I will ever do anything approaching this in real usage HIARCS Chess Explorer running two analysis engines simultaneously with four cores each. This is a CPU test, like Cinebench.

I was mainly interested in what the CPU temps did and what the performance level was. So there are only two graphs below, CPU average temp and Performance core average frequency. (The efficiency core frequency was 2.42GHz throughout the test. All data taken from the Mx Power Gadget log, generated as a spreadsheet.

As seen the temperature dropped to a sustainable c83C after 4 minutes and stayed near there for the rest of the two hours.

Performance core frequency matched this unsurprisingly.

Since the Cinebench score after 30 minutes was c7300 (15% drop) it seems reasonable to assume that Chess analysis speed drop would be similar.

The MBA fully throttled Cinebench R23 score of 7333 is in the ball park of most late cooled Intel Macs (ref).

Main conclusion is that the fanless M2 MBA is capable of sustained high load task at a performance level similar to most recent cooled Intel Macs.


The MBA perception problem is that the existence of the "Professional" models above the Air pushes the Air into a "Grannies and kids light use" category in many peoples minds. I suspect many people buy the Pro model because they don't appreciate how powerful the Air is, and not just for short periods.

 

[Mike Boreham]

The MBA fully throttled Cinebench R23 score of 7333 is in the ball park of most late cooled Intel Macs (ref).

Just to emphasise this point, here is the Mac Geekbench chart marked up:

EDIT Just to clarify, the above chart shows the throttling effect from Cinebench (c16%) applied to the Geekbench chart. Not fully accurate but shows the message broadly.

 

[bcortens]

Just to emphasise this point, here is the Mac Geekbench chart marked up:


View attachment 2149134

Makes me happy to see my 3 year old 8 Core i9 MacBook Pro loosing to the MacBook Air - shows just how far Apple leapt forward.

 

[adrianlondon]

Makes me happy to see my 3 year old 8 Core i9 MacBook Pro loosing to the MacBook Air - shows just how far Apple leapt forward.

My early-2020 Intel MBA doesn't even show on the list, and despite using it for data science and some machine learning, it performs well. As you say, it's good to learn how capable the newer machines are.

 

[Scarrus]

I decided to do a two hour max throttle (100% on all cores) test to see how the machine behaves.

I used the only app I will ever do anything approaching this in real usage HIARCS Chess Explorer running two analysis engines simultaneously with four cores each. This is a CPU test, like Cinebench.

I was mainly interested in what the CPU temps did and what the performance level was. So there are only two graphs below, CPU average temp and Performance core average frequency. (The efficiency core frequency was 2.42GHz throughout the test. All data taken from the Mx Power Gadget log, generated as a spreadsheet.

As seen the temperature dropped to a sustainable c83C after 4 minutes and stayed near there for the rest of the two hours.

Performance core frequency matched this unsurprisingly.

Since the Cinebench score after 30 minutes was c7300 (15% drop) it seems reasonable to assume that Chess analysis speed drop would be similar.

The MBA fully throttled Cinebench R23 score of 7333 is in the ball park of most late cooled Intel Macs (ref).

Main conclusion is that the fanless M2 MBA is capable of sustained high load task at a performance level similar to most recent cooled Intel Macs.


The MBA perception problem is that the existence of the "Professional" models above the Air pushes the Air into a "Grannies and kids light use" category in many peoples minds. I suspect many people buy the Pro model because they don't appreciate how powerful the Air is, and not just for short periods.



View attachment 2149129

Excellent work. Would be interesting to see if also applying a GPU load would significantly affect the thermals.

 

[Mike Boreham]

Excellent work. Would be interesting to see if also applying a GPU load would significantly affect the thermals.

Thanks. Any suggestion how to do this?

 

[Scarrus]

Hmm, maybe also run a GPU Benchmark/load at the same time? Be sure to leave at least a thread free so the Gpu load doesn’t starve because of the Cpu being busy.

Concercing what to use, I don’t suppose Furmark is available for Mac?

 

[Mike Boreham]

Hmm, maybe also run a GPU Benchmark/load at the same time? Be sure to leave at least a thread free so the Gpu load doesn’t starve because of the Cpu being busy.

I have just tried a Handbrake encode which I am pretty sure loads GPU. I am a novice at video encoding and the range of options is baffling, and prevents any meaningful absolute numbers. However I suggest that the change in framerate during the test might be be a measure of throttling. The test I did was:

Video was 6.3GB, 56 mins
Reduced size to 3.6GB using H264, checking "web optimised"
Job took 38 minutes.
Mx Power Gadget plots show CPU core average pulled back to 86C within 5 minutes as before.

The peak framerate at the start of the test was c60fps, and at the end was c30 fps.
The Average framerate was 37fps

This suggests greater throttling than the CPU tests (Cinebench and Chess analysis) which is probably to be expected, and confirms that if encoding longer video is your thing a Pro would be better....but also confirms that the M2 MBA can get the job the done without fuss when necessary, in a longer time of course...not an issue for occasional use.

Maybe someone expert in encoding could comment.

 

[Mike Boreham]

I have just tried a Handbrake encode which I am pretty sure loads GPU. I am a novice at video encoding and the range of options is baffling, and prevents any meaningful absolute numbers. However I suggest that the change in framerate during the test might be be a measure of throttling. The test I did was:

Video was 6.3GB, 56 mins
Reduced size to 3.6GB using H264, checking "web optimised"
Job took 38 minutes.
Mx Power Gadget plots show CPU core average pulled back to 86C within 5 minutes as before.

The peak framerate at the start of the test was c60fps, and at the end was c30 fps.
The Average framerate was 37fps

This suggests greater throttling than the CPU tests (Cinebench and Chess analysis) which is probably to be expected, and confirms that if encoding longer video is your thing a Pro would be better....but also confirms that the M2 MBA can get the job the done without fuss when necessary, in a longer time of course...not an issue for occasional use.

Maybe someone expert in encoding could comment.

FWIW I just did a repeat of this test with the MBA on a simple aluminium open-back stand which raises the back of the MBA by about 2". No fans.

Key numbers this time:
Job took 20 minutes (cf 38 last time)
Average framerate 47fps (cf 37 last time)
Initial framerate was about the same of course, but at the end it was c43 (cf 30 last time)

Obviously a Mac where all the heat is dissipated through the casing is going to be more sensitive to the case environment than a fan cooled Mac where it it is taken away by airflow, but I am surprised at the amount.

 

[Tagbert]

I have just tried a Handbrake encode which I am pretty sure loads GPU. I am a novice at video encoding and the range of options is baffling, and prevents any meaningful absolute numbers. However I suggest that the change in framerate during the test might be be a measure of throttling. The test I did was:

Video was 6.3GB, 56 mins
Reduced size to 3.6GB using H264, checking "web optimised"
Job took 38 minutes.
Mx Power Gadget plots show CPU core average pulled back to 86C within 5 minutes as before.

The peak framerate at the start of the test was c60fps, and at the end was c30 fps.
The Average framerate was 37fps

This suggests greater throttling than the CPU tests (Cinebench and Chess analysis) which is probably to be expected, and confirms that if encoding longer video is your thing a Pro would be better....but also confirms that the M2 MBA can get the job the done without fuss when necessary, in a longer time of course...not an issue for occasional use.

Maybe someone expert in encoding could comment.

I have heard that Handbrake does not make full use of multiple cores

 

[jdb8167]

I have heard that Handbrake does not make full use of multiple cores

It depends on the codec and the architecture used. On my 6-core 12-thread Mac Pro the x264 codec can saturate all 6 cores and about half of the other 6 threads. If you use the VideoToolbox codecs you probably wont see much CPU activity at all.

 

[Mike Boreham]

I decided to do a two hour max throttle (100% on all cores) test to see how the machine behaves.

I used the only app I will ever do anything approaching this in real usage HIARCS Chess Explorer running two analysis engines simultaneously with four cores each. This is a CPU test, like Cinebench.

I was mainly interested in what the CPU temps did and what the performance level was. So there are only two graphs below, CPU average temp and Performance core average frequency. (The efficiency core frequency was 2.42GHz throughout the test. All data taken from the Mx Power Gadget log, generated as a spreadsheet.

As seen the temperature dropped to a sustainable c83C after 4 minutes and stayed near there for the rest of the two hours.

Performance core frequency matched this unsurprisingly.

Since the Cinebench score after 30 minutes was c7300 (15% drop) it seems reasonable to assume that Chess analysis speed drop would be similar.

The MBA fully throttled Cinebench R23 score of 7333 is in the ball park of most late cooled Intel Macs (ref).

Main conclusion is that the fanless M2 MBA is capable of sustained high load task at a performance level similar to most recent cooled Intel Macs.


The MBA perception problem is that the existence of the "Professional" models above the Air pushes the Air into a "Grannies and kids light use" category in many peoples minds. I suspect many people buy the Pro model because they don't appreciate how powerful the Air is, and not just for short periods.



View attachment 2149129

For reasons I don't need to go into I had use of an M1 Max MBP for a few days and tried the chess analysis engines on it. As with the MBA, CPU usage goes to 100% and the fans ramp up to 5500 RPM within a few minutes. CPU core average was in the 90s. Performance core frequency stayed over 3 as expected.

So as expected it well outperforms the MBA...but having the fans running at 5500 RPM was audible, and detracted from the experience compared with the silent MBA, running more slowly, but fast enough for me. The experience cured my itch about whether chess analysis justified the MBP for me!

 

[GalileoSeven]

Most of the reviewers spouting this FUD have seemingly ignored an entire subset of users - the normal, everyday folks who just use their machines for routine stuff - web browsing, email, music, word processing and the like.

Whether at home (where it's hooked to a Dell 4K monitor) or on the go, my M2 MBA (8/512) has never even come close to throttling.....and that's with 12+ Firefox tabs open & imessage, music, mail, & pages up and/or running in the background. Even when I have Topaz Gigapixel AI running on top of all of that, this thing still doesn't sweat.

I could just have the right memory/SSD combo, or I could've just plain got lucky. Either way, I find it difficult that Apple screwed the pooch on a much, much larger subset of users/configs/whatever.

People need to seriously stop buying into clickbait like this (and would do well to remember, the MBA is Apple's entry level laptop model)

 

[Droid13]

web browsing, email, music, word processing and the like

Some of these normal, everyday folks may even be working professionals for whom the above represents all the tools that they need.

Don't need to use a device marketed as "Pro" to be a pro.

 

[GalileoSeven]

Some of these normal, everyday folks may even be working professionals for whom the above represents all the tools that they need.

Don't need to use a device marketed as "Pro" to be a pro.

Exactly.

 

[satcomer]

Your definition is far too narrow. The MacBook Air M2 is more than a basic device. It's a modern laptop capable of handling the needs of most people. This is my work computer. I use it for not only emails and browsing, but for a lot of work in Adobe Acrobat Pro (editing and OCR functions), Adobe Photoshop for heavy image editing, all the Office apps, video conferencing (Zoom & Teams), Music, Slack, and more - often with tons of those running in parallel at the same time. I upgraded to 16GB RAM, and 512GB SSD, and I think this computer will last me a very long time.

I don't pay Adobe subscription that milks money from you! I use the shareware Pixelmator Pro to edit my home pictures! It even can color correct videos now too and export them back to Apple's Final Cut!

 

[Yebubbleman]

From the evidence I've seen, I believe firmly that people are vastly overhyping the effects of the throttling in the M2 MacBook Air.

The YouTube review I've seen so far with the most thorough benchmarks is Dave2D's, and he went so far as to say that "if you're someone that wants a very powerful system and you're hoping that something super thin and super light could handle it, unfortunately, I would say that it can't." He also spends a good deal of time throughout the rest of the video highlighting the effects of the throttling as he saw them.

That's a rather extreme conclusion to make, from what I'm seeing. I will use his own graphs to demonstrate.

View attachment 2033243

This is Dave's graph of the effects of throttling on the Cinebench score, comparing the M2 MBA to the M2 MBP (which has a fan). There are some important things to note here.

I calculated that a single run on the M2 would last anywhere from 1.5 to 2 minutes, based on its overall score. So 10 consecutive runs would take about 15-20 minutes. Remember, throttling is related to temperature over time, and the effects of it go up as your time under sustained load increases.

Cinebench pegs the CPU at 99-100% utilization on all cores for the duration of the test. So we're talking 15-20 minutes of that. If you've spent any time doing "Pro" workflows, and monitored your system usage, you would know that this amount of load, for that duration, is extremely rare. Even in a video editor, you rarely have a render that would peg the CPU to that degree, for that amount of time. You do also have the GPU cores to consider in that instance, but many renders for many users also would not take 15-20 minutes to complete. At the highest end, sure, with very complex timelines with hefty codecs. But the majority of users who do video work who would buy this machine? No.

On top of that, I think Dave is over-estimating the impact of the throttling. I calculated that on his x axis, each interval represents 562.5 points. So if you look carefully, you see that after 10 consecutive runs, the M2 MBA has dropped about 1125 points.

View attachment 2033247

A drop of 1125 points puts it at about the same performance as M1's maximum performance, and this occurs after 15-20 minutes straight of 100% CPU load. 8783 to 7658.

So if you somehow have a workload that is pegging the CPU at 100%, or a combination of GPU + CPU that equals the same thermal output, for 15-20 minutes STRAIGHT, your performance STILL only drops to about the equivalent of M1's maximum performance. 8783 vs. 7743.

For the vast majority of users, their workloads will stay well within the zone where they are reaping the benefits of M2's performance increase. Even many sustained loads do not load up as much as Cinebench does. And for users who manage to push to the throttling point, their performance only goes down to about the performance of M1. Pushing past that point will be even rarer still.

So, as is always the case with throttling, the debate is not "Does the machine throttle or not?", it's "How much of an impact does throttling have?"

So far, I am not seeing any evidence that suggests that M2 MacBook Air users will have a poor experience. I think the effects of this throttling are far less than how it's being described. I also believe Apple would have weighed all of this themselves when testing this machine and determining how it would perform for its expected user base. For most users, it will still exceed their expectations, even while editing video and doing other "pro" workflows.

There are two schools of thought on this and I'm frankly beyond tired of having these arguments so I won't engage in one here directly.

One school of thought is: This isn't a pro machine, it's geared toward low-end computer users and for them an M2 MacBook Air will be perfectly fine.

The second school of thought (that I find myself arguing ad infinitum) is: This machine has worse thermals than its immediate predecessor, gets hotter and sooner than its predecessor did and therefore is a downgrade when it comes to being able to do higher-end tasks that its immediate predecessor was actually able to do somewhat passably, if not decently.

These two schools of thought are not mutually exclusive, yet people will debate the second one with the first one like the first one doesn't matter.

But yeah, the M2 is a hotter running chip, and the M2 MacBook Air has a smaller cooling system than the M1 MacBook Air, while the cooling system in the M2 13-inch MacBook Pro is virtually unchanged from that of the M1 13-inch MacBook Pro and it was largely considered unnecessary in the M1 13-inch MacBook Pro (presumably still being more than adequate on the M2 13-inch MacBook Pro). Which is to say, active cooling will always result in cooler operation and a hotter throttle point, especially on a hotter running chip.

So, yeah, it'll be more than fine for people who owned any Intel MacBook Air AND it isn't as good as its immediate predecessor when it comes to thermal management. Your mileage may very on whether or not this actually presents a real-world difference that matters, despite it not being a good look for Apple to release a product that is a downgrade in any respect from its predecessor in any case.

 

[Mike Boreham]

There are two schools of thought on this and I'm frankly beyond tired of having these arguments so I won't engage in one here directly.

One school of thought is: This isn't a pro machine, it's geared toward low-end computer users and for them an M2 MacBook Air will be perfectly fine.

The second school of thought (that I find myself arguing ad infinitum) is: This machine has worse thermals than its immediate predecessor, gets hotter and sooner than its predecessor did and therefore is a downgrade when it comes to being able to do higher-end tasks that its immediate predecessor was actually able to do somewhat passably, if not decently.

These two schools of thought are not mutually exclusive, yet people will debate the second one with the first one like the first one doesn't matter.

But yeah, the M2 is a hotter running chip, and the M2 MacBook Air has a smaller cooling system than the M1 MacBook Air, while the cooling system in the M2 13-inch MacBook Pro is virtually unchanged from that of the M1 13-inch MacBook Pro and it was largely considered unnecessary in the M1 13-inch MacBook Pro (presumably still being more than adequate on the M2 13-inch MacBook Pro). Which is to say, active cooling will always result in cooler operation and a hotter throttle point, especially on a hotter running chip.

So, yeah, it'll be more than fine for people who owned any Intel MacBook Air AND it isn't as good as its immediate predecessor when it comes to thermal management. Your mileage may very on whether or not this actually presents a real-world difference that matters, despite it not being a good look for Apple to release a product that is a downgrade in any respect from its predecessor in any case.

There is an underlying assumption in your post that more throttling equals worse design, backwards step etc. This is an understandable point of view because for the last thirty years the general wisdom has been that computers should be designed with cooling systems that prevent throttling. Throttling equals bad design.

The capability of silicon means the whole paradigm has changed. Throttling is now a valid design option to avoid the negatives that go with fans. This is very hard for many people to accept.

More throttling in the M2 Air is just a different design choice, not a bad design choice. Even in its fully throttled state it is faster than the M1 (here). The M2 can run safely and silently for hours doing high load tasks in the throttled state. (here).

So yes you are quite correct that the M2 Air throttles more but it is still a faster computer than the M1, and it is still faster than many recent cooled intel macs. (here).

(Edited to remove some repetition)

 

[jdb8167]

This machine has worse thermals than its immediate predecessor, gets hotter and sooner than its predecessor did and therefore is a downgrade when it comes to being able to do higher-end tasks that its immediate predecessor was actually able to do somewhat passably, if not decently.

Your definition of what is a downgrade is curious. Ultimately the M2 is faster than its predecessor even when throttled. Why do I care if the internals of my computer gets hotter faster than the previous model? It’s still faster and the extra heat isn’t particularly noticeable.

 

[Yebubbleman]

There is an underlying assumption in your post that more throttling equals worse design, backwards step etc.

It lowers performance at a lower performance threshold. You can say that it doesn't matter. But you can't say that in and of itself isn't a step backwards. Incidentally, given that Apple uses and plans for body styles to last over the course of multiple different processor generations, I'm not going to say that the body style of the M2 MacBook Air is a flawed one; though I will totally assert that it was planned with a later SoC in mind more than it was for the M2.

This is an understandable point of view because for the last thirty years the general wisdom has been that computers should be designed with cooling systems that prevent throttling. Throttling equals bad design.

There's nothing wrong with throttling. However, if I have an M1 MacBook Air that doesn't throttle until I start loading complex plug-in #3 and an M2 MacBook Air that throttles upon loading complex plug-in #2, that still is a downgrade in real-world performance from the standpoint of that workflow especially if M2 throttled is not as fast as M1 pre-throttling (which some have pointed out as being the case).

The capability of silicon means the whole paradigm has changed. Throttling is now a valid design option to avoid the negatives that go with fans. This is very hard for many people to accept.

You don't need to talk to me like I don't understand what Thermal throttling is, how it works, and why it happens when it happens. Most people that act like it's no big deal are comparing M2 MacBook Airs to Intel MacBook Airs and not to M1 MacBook Airs. Those same people are often engaging in those arguments to defend their own purchase of the M2 MacBook Air because they feel as though it is a wholesale attack on their own Apple products.

More throttling in the M2 Air is just a different design choice, not a bad design choice. Even in its fully throttled state it is faster than the M1.

Debatable and highly dependent on the workflow in question.

M1 and M2 can run safely for hours doing high load tasks in the throttled state. (eg here and here)

The whole point of throttling is to be safe for the chip whose frequencies are being throttled down. If either MacBook Air didn't throttle to a safe point upon load, it would overheat and then people would get mad and never buy it again.

So yes you are quite correct that the M2 Air throttles more but it is still a faster computer.

Furthermore even in its fully throttled state M2 air can carry out high load tasks for extended periods, silently and safely. It does them more slowly of course, and it is still faster than many recent cooled intel macs. (here).

Again, if the goalpost is Intel Macs, then an M2 MacBook Air is still unquestionably better. Most people justify the sooner throttling under lesser load with that goalpost and with the notion that it is still more powerful than most Intel Airs and Intel 13-inch Pros. That's not saying a lot! To your credit, at least you're giving me links to actual benchmarks to back up claims of real-world performance rather than giving me the same nonsense "this machine is designed for casual users" argument.

 

[jdb8167]

Debatable and highly dependent on the workflow in question.

Not really. The M2 is going to be faster in all real world situations than the M1. Trust me. I have both.

 

[Yebubbleman]

Your definition of what is a downgrade is curious. Ultimately the M2 is faster than its predecessor even when throttled.

Because that part in bold isn't universally true for all workloads. M2 isn't so much faster than M1 that M2 throttled is unarguably and universally faster than M1 un-throttled. M2 is not that big of an upgrade over M1.

Why do I care if the internals of my computer gets hotter faster than the previous model? It’s still faster and the extra heat isn’t particularly noticeable.

Personally, I don't care about a hotter running laptop in general. If I'm buying a laptop that's hotter than its predecessor, then I had better be getting substantially more performance to justify that extra heat. But that's as far as I care about it, personally. But if you read how others on here talk about how hot Intel Mac laptops were on their laps, you'd think that every fraction of a Fahrenheit degree mattered.