Apple Silicon M1: benchmarks and review for Mac Mini, MacBook Air and MBP 13 M1 and comparison with other brands

[EugW]

I knew full well what the difference was going to be when I canceled my initial pro order and went with the air instead. I prioritized the fact that the air is lighter and passively cooled over sustained performance in addition to being able to get more storage while still being cheaper then the base pro. Most of my workflow consists of burst operations anyways instead of sustained load so the air and pro should be pretty comparable for me. That’s not even considering the fact that the air is still faster then most of the current Intel MBP 13” configs.

It was heavily hinted throughout the keynote that the M1 targets 15watts as its max TDP.

The best part of my 12" MacBook is that it is fanless. Enjoy that in your new Air!

The worst part of my 12" MacBook is that it has just one USB-C port. However, the 13" Air does not have that problem. I'm jealous.

 

[Henk van Ess]

Another live one , in Chinese EDIT Japanese

 

[EugW]

Worst first review ever

Haha. Serial number C02DN20N0KPF.

 

[thingstoponder]

The 16" 2019 is faster than 13" 2019 in both single core and multi core scenarios. Not only that, it has better battery life and better screen/keyboard as well. It was better in every way possible except for maybe size and weight.

And the 13" M1 MacBook is faster than A14 in iPhone 12 Pro in every way possible. The iPhone 12 Pro is only... portable, but that's about it. There's no way you can make an iPhone 12 Pro last 20 hours with the screen on all that time, but supposedly, that should be possible with the 13" Pro.

So there was always a natural progression in terms of performance scaling and value proposition.

I'm just not sure we'll see that with the next 16" compared to the 13" M1 if Apple isn't willing to sacrifice on efficiency. So there will be drawbacks and tradeoffs.

That’s not true though. The newest Intel Air and higher end 13” Pro are both faster than the 16” in single core as they have a newer architecture and node.

The M1 is marginally faster in single core than the iPhone 12, it’s negligible really.

The 16” has a huge battery, I see no reason it wouldn’t get better battery life than the 13”. What I think they’ll do is have the same chip in the higher end 13” and the 16”. If they do it will get way more battery life. Even if it had a higher end chip than the higher end 13” Pro I wouldn’t rule it out. The new chips are just so much more efficient than the old Intel chips and discrete GPUs.

 

[bill-p]

If they don't understand how Geekbench works, that's their problem. Geekbench is a good bench for what it is. It is a good general purpose measure, but it's lousy at measuring sustained workloads, because it doesn't last very long. OTOH, for bursty workloads, it's very good.

But that's OK, because in general, people who use Airs work in bursts. For example, rendering a complex multi-media webpage still only takes seconds, and then the computer just sits there doing nothing, or almost nothing. It's not as if it has to render that webpage continuously at full tilt for 20 minutes.

To oversimplify things, Geekbench tells you how fast the computer will feel when you're sitting at it and actively using it, whereas Cinebench tells how long that long render is going to take and how loud your fans are going to get.

Well, some websites have animations that keep playing. Especially if those are ads. Non-advanced users won't block those ads, and they'll remain in motion as long as you have that tab somewhere. It's not like a random website will be smart enough to pause ad while you're navigating away.

And if you scroll the site, depending on the scrolling algorithm, things may be rendered in and out all the time, so there is always load on the CPU, even if that load may not be too high. A mobile website like the kind presented on the iPhone and iPad may not have all those dynamic elements, but... a full desktop website? Yes, those exist everywhere. Case in point of a site that is arguably one of the worst offenders, and yet the average Joes will visit a lot: Facebook.com.

If the internet is not full of these websites now, the 2020 MacBook Air would not be accused of thermal throttling.

It's kind of ironic, but Flash was killed because its dynamic animations were causing computers to run hot and thermal throttle back in 2010. And now in 2020, we have JavaScript, CSS3 and HTML5, and they aren't going anywhere.

So I'd argue neither Cinebench nor Geekbench represent real world workloads... but Cinebench would be closer, because at least it doesn't assume the user will be a robot that clicks a button and does absolutely nothing for the next 15 minutes.

 

[M1 Processor]

Valuable to whom? We know that Zen 2 single core performance is not very impressive and we know that they are no 10W x86 CPU. What's the point in discussing that? We can compare it to a server CPU if we find it interesting for some reason. I think a lot of people are taking this thread way too seriously. We are not the official judges of the "worlds best CPU" competition, so we can make unfair comparisons. And I think that's interesting for people that want to have a good overview of the whole CPU market and I think that's in line with the topic.

Erm... Incorrect, the 8 core 4800u can run at 10w, the performance will be lower, but the performance per watt will go up

 

[EugW]

Well, some websites have animations that keep playing. Especially if those are ads. Non-advanced users won't block those ads, and they'll remain in motion as long as you have that tab somewhere. It's not like a random website will be smart enough to pause ad while you're navigating away.

And if you scroll the site, depending on the scrolling algorithm, things may be rendered in and out all the time, so there is always load on the CPU, even if that load may not be too high. A mobile website like the kind presented on the iPhone and iPad may not have all those dynamic elements, but... a full desktop website? Yes, those exist everywhere. Case in point of a site that is arguably one of the worst offenders, and yet the average Joes will visit a lot: Facebook.com.

If the internet is not full of these websites now, the 2020 MacBook Air would not be accused of thermal throttling.

It's kind of ironic, but Flash was killed because its dynamic animations were causing computers to run hot and thermal throttle back in 2010. And now in 2020, we have JavaScript, CSS3 and HTML5, and they aren't going anywhere.

So I'd argue neither Cinebench nor Geekbench represent real world workloads... but Cinebench would be closer, because at least it doesn't assume the user will be a robot that clicks a button and does absolutely nothing for the next 15 minutes.

Those websites will only slow down old low performance computers. In contrast, just the efficiency cores on the MacBook Air M1 would be able to handle those just fine.

Anyhow, what you've described is still bursty workload. Scrolling up and down going in and out of rendered elements is the poster child of this. You scroll and then it renders. And then stops rendering, or else does stuff at relatively low CPU usage. Honestly, I don't think you quite understand how this works. Burst and idle doesn't mean working for 10 minutes and then doing nothing for 20 minutes. It means working for 2 seconds and then doing nothing for 1 second and then working for 3 sections and doing nothing for 5 seconds, etc. I suggest you open up Activity Monitor, and increase the update frequency, and then start surfing while keeping an eye on Activity Monitor. Then I suggest you stop doing that and run Cinebench, again keeping an eye on Cinebench.

With surfing, you'll see it jump up and then drop to like 5%, and jump up again and drop back down to 15%, etc. With Cinebench, it will run at 100% all the time. Thus, Cinebench is a terrible measure in that regard.

To put it another way, in regular everyday usage, the MacBook Air M1 and the MacBook Pro M1 should feel very similar. However, for rendering long clips of video using CPU software encode, the MacBook Pro will win every time.

BTW, Flash was was not killed because of performance. Flash was killed because it was a buggy mess, and because it was a security hazard. Yes it was also full of bloat, but the other factors were even more important.

 

[M1 Processor]

And you are making me wonder how many people bought the MacBook Air without knowing this...

Then again, can't blame them when all we had for most of last week were Geekbench scores that basically showed zero difference in performance between the Air and Pro.

Geekbench pulled the wool over their eyes, but no worries if you are not stressing the system, you won’t see a difference.

 

[bill-p]

That’s not true though. The newest Intel Air and higher end 13” Pro are both faster than the 16” in single core as they have a newer architecture and node.

The M1 is marginally faster in single core than the iPhone 12, it’s negligible really.

The 16” has a huge battery, I see no reason it wouldn’t get better battery life than the 13”. What I think they’ll do is have the same chip in the higher end 13” and the 16”. If they do it will get way more battery life. Even if it had a higher end chip than the higher end 13” Pro I wouldn’t rule it out. The new chips are just so much more efficient than the old Intel chips and discrete GPUs.

I don't see how you can claim the 2020 Intel Air and Pro 13" are both "faster" than the 16" 2019 but cannot concede that the M1 is still faster than the iPhone 12. It's the exact same kind of difference. It's not "negligible" or "marginal":

MacBook Pro (13-inch Late 2020) - Geekbench

Benchmark results for a MacBook Pro (13-inch Late 2020) with an Apple M1 processor.

browser.geekbench.com

iPhone 12 Pro Benchmarks - Geekbench

browser.geekbench.com

The 16" has a huge battery but also a huge screen, and it'll need a GPU that's anywhere between 3-4x faster than whatever is in the M1 just to "break even" with the 5600M that is in the current 16". That and let's say it doesn't need faster single cores, but it'll still need more cores. More cores will consume more power.

Realistically, we're looking at a 28 - 35W chip for the MacBook Pro 16". More if Apple needs to squeeze out more GPU power to claim anything like a meaningful 1.5x improvement, because even as it is, the M1 cannot claim more than a 30% improvement over the Core i9, except for maybe in single-core performance. At 35W for whatever that next chip is, the 16" worst-case battery life is down to about 3 - 4 hours. If it's 40W, we're down to 2.5 hours. Will such a powerful chip last even as long as the MacBook Air M1? I sincerely doubt it.

In contrast, the current 13" Pro with the same 10W as the Air can easily last 5 - 6 hours even when it's stressed at full load. And it'll most likely float around the 12 - 14 hours mark.

So the problem now is... you're asking Apple to pull off a chip that has 4x the GPU performance of the M1... at just about 2x the power consumption increase, but without a new architecture or process node. And that's not to mention bundling in more ports, more bandwidth, more RAM...

I don't know. That sounds unrealistic to me.

 

[M1 Processor]

Cinebench is a good measure of sustained Cinebench-type CPU workload, but it isn't really a good gauge of other stuff.

If you want to see if your fanless computer throttles though, it's a good bench. Or if you want to see how loud your fans get, it's a good bench.


If they don't understand how Geekbench works, that's their problem. Geekbench is a good bench for what it is. It is a good general purpose measure, but it's lousy at measuring sustained workloads, because it doesn't last very long. OTOH, for bursty workloads, it's very good.

But that's OK, because in general, people who use Airs work in bursts. For example, rendering a complex multi-media webpage still only takes seconds, and then the computer just sits there doing nothing, or almost nothing. It's not as if it has to render that webpage continuously at full tilt for 20 minutes.

To oversimplify things, Geekbench tells you how fast the computer will feel when you're sitting at it and actively using it, whereas Cinebench tells how long that long render is going to take and how loud your fans are going to get.

I have to disagree with you there because geekbench scores don’t always line up with real world performance. Exynos is a prime example in which you can “cheat” on Geekbench but do poorly in the real world. There were others in x86 too, but I don’t have time on hand at the moment.

 

[sultanq]

So I'd argue neither Cinebench nor Geekbench represent real world workloads... but Cinebench would be closer, because at least it doesn't assume the user will be a robot that clicks a button and does absolutely nothing for the next 15 minutes.

I'm not sure what you mean by that. Cinebench is a realistic representation of long renders, but not a realistic representation of typical photo editing workloads, and definitely very far from a typical web browsing workload, where most of the time is spent reading or scrolling or typing rather than rendering new pages. For the first phase of the test, Geekbench keeps one core fully loaded with a roughly 60% duty cyle, and in the second phase, it keeps all the cores fully loaded with a similar duty cycle. The Geekbench workload is much more intensive than typical web browsing or photo editing workloads, but with its burstiness and not-100% duty cycle, it's a better representation of such bursty user-interaction workloads than the 100% duty cycle of Cinebench.

 

[M1 Processor]

I'm not sure what you mean by that. Cinebench is a realistic representation of long renders, but not a realistic representation of typical photo editing workloads, and definitely very far from a typical web browsing workload, where most of the time is spent reading or scrolling or typing rather than rendering new pages. For the first phase of the test, Geekbench keeps one core fully loaded with a roughly 60% duty cyle, and in the second phase, it keeps all the cores fully loaded with a similar duty cycle. The Geekbench workload is much more intensive than typical web browsing or photo editing workloads, but with its burstiness and not-100% duty cycle, it's a better representation of such bursty user-interaction workloads than the 100% duty cycle of Cinebench.

Well you see that the single core portion of it will give an idea of web browsing and the M1 will perform similar to Tiger Lake in web browsing. I would agree with you on the photo editing part, but often that’s performed by the GPU

 

[Henk van Ess]

A tender test, 15 minutes old

 

[Henk van Ess]

Cinebench r23

 

[Sanpete]

The AMD CPUs were released early this year so they are a lot of models with them. They are difficult to find because they are out of stock almost everywhere, but one example would be the Yoga Slim 7 4800U. But for which comparison are you talking about? They are very vague except the ones that compare to the previous generations

We were originally talking about the claim that the M1 is twice as fast as the latest and greatest PC chip. @M1 Processor thought that claim showed they were comparing to Intel, not AMD. I'm not so sure.

The claim was actually quite vague, as you say: "At just 10 watts (the thermal envelope of a MacBook Air), M1 delivers up to 2x the CPU performance of the PC chip." I think that claim is relative to the power, so up to 2x at a particular power draw, probably 10W. I don't see any reason that couldn't be about the 4800U or a similar chip.

It's the claim that followed that one that I think would be a tougher comparison with a chip like the 4800U: "And M1 can match the peak performance of the PC chip while using just a quarter of the power." That suggests a TDP of maybe 40W at peak power. The footnote explains that multiple benchmarks were used, so I'd guess some kind of composite was used.

Supposing they were comparing to the 4800U operating at 40W, the M1 would likely beat the 4800U by a little in Geekbench. It would lose by a larger margin in Cinebench r23. Don't think either of those tests was available in native form for the M1 in October, so who knows what they used and how. Supposing they used other benchmarks with potentially similarly mixed results, it still seems possible, pending further results, they were comparing to a chip like the 4800U.

I am doubtful they are comparing AMD for two reasons.

1. Everything that they said up into this point compared Intel, and can’t directly compared with AMD since AMD silicon is not available for macOS. Geekbench scores are a bit higher on macOS than windows btw.

2. Their benchmarks don’t line up with what we know about chips such as the 4800u. As a point of reference the 4800u scores about 10000 in R23 compared to 7600 in the M1. Of course the 4800u is probably its 25w tpd config, vs 15 for the MBP.

Well, they might have said Intel if they meant Intel. Not qualifying their claim should make it inclusive of the best PC chip available.

I think the numbers may line up fine, as explained above. But we'll have a better idea as other CPU benchmark results become available.

Anyway, this is just a matter of curiosity, what Apple was really comparing to.

 

[bill-p]

Those websites will only slow down old low performance computers. In contrast, just the efficiency cores on the MacBook Air M1 would be able to handle those just fine.

Anyhow, what you've described is still bursty workload. Scrolling up and down going in and out of rendered elements is the poster child of this. You scroll and then it renders. And then stops rendering, or else does stuff at relatively low CPU usage. Honestly, I don't think you quite understand how this works. Burst and idle doesn't mean working for 10 minutes and then doing nothing for 20 minutes. It means working for 2 seconds and then doing nothing for 1 second and then working for 3 sections and doing nothing for 5 seconds, etc. I suggest you open up Activity Monitor, and increase the update frequency, and then start surfing while keeping an eye on Activity Monitor. Then I suggest you stop doing that and run Cinebench, again keeping an eye on Cinebench.

With surfing, you'll see it jump up and then drop to like 5%, and jump up again and drop back down to 15%, etc. With Cinebench, it will run at 100% all the time. Thus, Cinebench is a terrible measure in that regard.

To put it another way, in regular everyday usage, the MacBook Air M1 and the MacBook Pro M1 should feel very similar. However, for rendering long clips of video using CPU software encode, the MacBook Pro will win every time.

BTW, Flash was was not killed because of performance. Flash was killed because it was a buggy mess, and because it was a security hazard. Yes it was also full of bloat, but the other factors were even more important.

Well, let's just agree to disagree. I don't think my definition of "bursty" workloads will match up with yours then. I've had all of the fanless 12" MacBooks, as I don't doubt you have as well, and both of us are very well aware of the performance level of those machines. They were supposed to be "bursty" as well, but they didn't perform very well.

Also the 2020 13" Air didn't perform very well. If most of its users do just the "bursty" kind of workload as you've said, the CPU should not have heated up so much, but here we are.

I don't doubt the M1 Air will perform well, courtesy of the M1 chip, but I still think you and many others are putting too much faith into a fanless machine. We'll see how it is when we get our computers.

Last but not least, however Flash was killed, let's not forget that we went without it and vector animations on the web for half a decade before HTML5 and CSS3 stabilized. And now those are coming back as mobile hardwares are picking up in performance and can cope with animations in ads again. It's a vicious cycle.

 

[Henk van Ess]

 

[defn]

Thanks for continuously posting these real results. Wish there was a way to mute all these other hypothetical discussions....

 

[thingstoponder]

I don't see how you can claim the 2020 Intel Air and Pro 13" are both "faster" than the 16" 2019 but cannot concede that the M1 is still faster than the iPhone 12. It's the exact same kind of difference. It's not "negligible" or "marginal":

MacBook Pro (13-inch Late 2020) - Geekbench

Benchmark results for a MacBook Pro (13-inch Late 2020) with an Apple M1 processor.

browser.geekbench.com

iPhone 12 Pro Benchmarks - Geekbench

browser.geekbench.com

I never said the M1 wasn't faster than the A14, only that it was a relatively small difference. You were saying every higher end device needs to be better in every way than the ones under it, all I said was that isn't even true now. The difference in the Intel 13" and 16" is a bigger percentage than the A14 and M1. it's not the same kind of difference.

The 16" has a huge battery but also a huge screen, and it'll need a GPU that's anywhere between 3-4x faster than whatever is in the M1 just to "break even" with the 5600M that is in the current 16". That and let's say it doesn't need faster single cores, but it'll still need more cores. More cores will consume more power.

Yes but not that much more power assuming they don't crank the clock speed and ruin their efficiency.

Realistically, we're looking at a 28 - 35W chip for the MacBook Pro 16". More if Apple needs to squeeze out more GPU power to claim anything like a meaningful 1.5x improvement, because even as it is, the M1 cannot claim more than a 30% improvement over the Core i9, except for maybe in single-core performance. At 35W for whatever that next chip is, the 16" worst-case battery life is down to about 3 - 4 hours. If it's 40W, we're down to 2.5 hours. Will such a powerful chip last even as long as the MacBook Air M1? I sincerely doubt it.

28-35W sounds great to me, the GPU alone is 50W in the current 16" Pro. The single core will easily be 30% faster than the i9 in single core.

When I said it will have better battery life I meant in the ways Apple advertises, web browsing and video playback. If 2.5-4 hours is the worst case I will be disappointed, but keep in mind people report that they can drain the current one in about an hour or less. Thats a big improvement.

In contrast, the current 13" Pro with the same 10W as the Air can easily last 5 - 6 hours even when it's stressed at full load. And it'll most likely float around the 12 - 14 hours mark.

So the problem now is... you're asking Apple to pull off a chip that has 4x the GPU performance of the M1... at just about 2x the power consumption increase, but without a new architecture or process node. And that's not to mention bundling in more ports, more bandwidth, more RAM...

I don't know. That sounds unrealistic to me.

We'll see. The thing I wonder about is the higher end 13". I expect that to be the weakest in battery if it has more performance but the same battery as the M1 version.

 

[Henk van Ess]

Thanks for continuously posting these real results. Wish there was a way to mute all these other hypothetical discussions....

One guy even said rudely to his girlfriend : don’t touch this computer

 

[Henk van Ess]

Another glorious waste of bandwidth

 

[syndr0me]

One guy even said rudely to his girlfriend : don’t touch this computer

He told her to keep her fingers socially distanced from the M1 MBP! 😂

 

[torxcrossword]

Think the mini will run faster than the pro?

 

[syndr0me]

I’m not sure what’s worse, the early videos or the dick waving in this thread.

 

[bill-p]

I never said the M1 wasn't faster than the A14, only that it was a relatively small difference. You were saying every higher end device needs to be better in every way than the ones under it, all I said was that isn't even true now. The difference in the Intel 13" and 16" is a bigger percentage than the A14 and M1. it's not the same kind of difference.

Yes but not that much more power assuming they don't crank the clock speed and ruin their efficiency.

28-35W sounds great to me, the GPU alone is 50W in the current 16" Pro. The single core will easily be 30% faster than the i9 in single core.

When I said it will have better battery life I meant in the ways Apple advertises, web browsing and video playback. If 2.5-4 hours is the worst case I will be disappointed, but keep in mind people report that they can drain the current one in about an hour or less. Thats a big improvement.

We'll see. The thing I wonder about is the higher end 13". I expect that to be the weakest in battery if it has more performance but the same battery as the M1 version.

I don't think you can say benchmark scores are a matter of percentage... because the higher you scale, the harder it is for you to see gains unless you switch to a whole new architecture. I'd take the absolute score difference, but maybe that's just me. Also, that's not to mention the iPhone 12 Pro does thermal throttle quite often. It heats up very easily... and I encounter this every day because I do video calls with my wife on a regular basis. The phone heats up like its life depends on it. So the chip is clearly not meant to be in this body, or else Apple is too obsessed with looking good in benchmarks that they have scaled the chip too high. It probably doesn't hurt to scale it down a little.

And it's unfortunate that Geekbench is the only benchmark we can use to compare the iPhone and MacBook. The Air and Pro don't look different in Geekbench either, until Cinebench is engaged.

But anyways, yes, the current 16" is not great in worst-case. It really can drain in an hour flat. The GPU is not 50W, though. At max load, it's about 55 - 60W (I have the 5500M), and the CPU, while rated at 45W, can draw 120W when Turbo-boosting.

In regular use, it does last about 7-8 hours thanks to the massive battery, but do anything slightly more intensive like... say... "watching Youtube video" and that figure drops pretty rapidly.

So I've had enough of loud fans and short battery run time. But I'm not going to fall into the "fanless" trap of the old days anymore. Even if the fan in the 13" Pro is useless, at least it'll help when ambient temps get toasty. Last summer in Cali, I've learned that no device I own can work at full speed at 110F. The iPad Pro and iPhone hung on for dear life, and the 16" didn't even bother. I'm hoping the M1 Pro will do better next summer.