Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

mi7chy

Suspended
Original poster
Oct 24, 2014
10,686
11,350
Keep sustained load under 30 minutes or get the version with fan.

1658293805062.png
 
Although I know nothing about Tomb Raider it is good to see some solid data. I expect similar results apply for other types of sustained load.

No doubt some people will see this and think the M2 Air is a disaster ("real computers don't throttle"), and anyone doing frequent sustained load work should indeed not get the M2 Air.

For anyone who does occasional sustained load work and doesn't mind waiting a bit longer, the M2 Air is brilliant. All that performance without the hassle of a fan.
 
15W GPU + 15W CPU + thermal insulation

The bottom chassis won't get super hot, but it also means quicker throttling.

Screen Shot 2022-07-19 at 11.27.52 PM (2).png
 
So the passively cooled M2 loses 30% off its peak performance after 30 minutes and it’s a disaster and an engineering fail. But the 6800U loses over 40% off it’s peak performance after seconds and it’s a marvelous product and the best CPU ever. Double standards and misdirection like usual.
 
Any video showing it?

Compare the base clock to the boost clock. Base clock is the sustained performance, boost clock is peak performance.

The thing is, if M2 in the Air were an x86 CPU with an advertiser base clock of 2.5 and boost clock of 3.5, everyone would be raising their hands in admiration of having such an amazing passively cooled system. But since Apple doesn’t advertise the frequencies or TDPs people just invent their own ridiculous metrics to fit the common narrative, but they dint apply this narrative systematically.

The entire discussion of how much M2 throttles etc. is meaningless, because again, it does not apply the same standard to Apple as it does to the rest of the industry. The only relevant thing should be relative performance. How does M2 perform under burst/sustained load compared to M1 or to other similarly priced ultraportables? What is the best product for the specific user? These are constructive questions that have real-world value.
 
Boost clock is optional on 6800U. If you're comparing M2 base clock with 6800U boost clock then I can see why you'd think 8GB=16GB.

M2 does not have any advertised base clock or boost clocks, so what you say doesn't make any sense. If you are only willing to limit the 6800U to base clock only when discussing benchmarks feel free to cut most of it's published scores in half.

What I am saying is that you are completely ignoring that the M2 has a dynamic performance range just like any other contemporary CPU and that M2 in the MBA is not the same as the M2 in the MBP. The "base" clock equivalent for MBA might be 2.5-2.7Ghz for all we know while for the MBP it might be closer to 3.0-3.2Ghz. The data-mined 3.5Ghz clock is the "boost clock". Not that these concepts mean much for Apple Silicon which operates differently from the mainstream x86 CPUs. But if we ignore these differences for a moment you are basically complaining that the "base" for the M2 Air is 30% lower than the "peak". Which is utterly ignorant and opportunistic given the fact you are happy with 40-50% difference between the "base" and the "peak" on the x86 CPUs. Not to mention that the MBA can still operate at the "peak" on battery, while most x86 systems will see their peak performance cut by 20-30% just by unplugging.
 
It is precisely as Leman describes. Intel CPUs have two power limits PL1 and PL2, the first one is typically referred to as TDP, because PL1 is typically configured to match the thermal dissipation capacity of the chassis. Along with the power limits there are according frequency limits defined - the base frequency, which Leman is referring to, is the guaranteed frequency under PL1 power limit for a particular workload. Conceptionally we can apply this also to non-Intel devices.

That having said, the PL1 or TDP for passively cooled devices is below 10W (or 7-9W in actual devices in a laptop or tablet form factor) - this also applies to the Macbook Air. This means, after a certain amount of time the power limit drops from PL2 down to PL1 and the frequencies are reduced accordingly. As I said, conceptionally this is applicable to any device/CPU. Also to give you an idea, for phones the PL1 is usually below 4W.

Of note here is, that if you are running Geekbench, then you almost certainly running under PL2 and the result is the peak performance of the device and never the sustained performance, which typically is much lower. Therefore the Geekbench score of the MBPro and the MBAir are more or less identical as both running under the same PL2. However ther PL1 limit is quite a bit lower on the MBAir compared to the MBPro due to cooling differences.

Also I agree to Leman, with the assessment of the 6800U situation, clocks have chance to be reduced to base clock under PL1 (which can be set between 15W and 28W for the 6800U) - and they would have to be reduced much more if you (hypothetically) set the PL1 of the 6800U to 9W - in order to put it into a passively cooled device.
 
If you're a professional gamer, better get M2 MBP then.
No, better get an M1-Pro or Max for their more powerful GPUs and two fans.

No one who is serious (enough) about games, and also wants a Mac, is going to get the computers with the weakest GPUs and cooling.

Another take away is that the M2 Air with its slower SSD and supposedly inferior cooling did better than the equivalent M1 Air.
 
Last edited:
Once again people are using their computers wrong and then getting all pissed off at the manufacturer.

The MacBook Air is designed for basic users doing light tasks. Anyone pushing the Air into thermal throttling is on the wrong laptop.

Even without a fan, the M1 and M2 have more processing headroom than PRO COMPUTERS THAT ARE STILL IN USE. My Classic Mac Pro is more processing power than most people ever need (myself included), and the M1 and M2 are even more powerful than that.

The M2 under maximum throttling is still faster than a cold M1.

There is no thermal problem. End of line.
 
the 6800U loses over 40% off it’s peak performance after seconds

This is the only benchmark I have found that shows the performance decay of the AMD Ryzen 7 6800U throughout the test. It does not show a 40% drop as you were saying.
6800U.png


 
  • Like
Reactions: JPack
This is the only benchmark I have found that shows the performance decay of the AMD Ryzen 7 6800U throughout the test. It does not show a 40% drop as you were saying.
View attachment 2032275


It throttles by 16.7% after 10 minutes but it also has a fan that gets louder by 7.5dB or 150% unlike the MB Air. If it was fanless it would throttle by 41.7% according to the chart.

Skärmavbild 2022-07-20 kl. 21.58.06.png
 
  • Like
Reactions: souko
This is the only benchmark I have found that shows the performance decay of the AMD Ryzen 7 6800U throughout the test. It does not show a 40% drop as you were saying.
View attachment 2032275


It’s because it throttles instantly. At peak power (peak possible clock) that CPU would consume over 60-80W which is why the PL2 limit kicks in after few milliseconds. Also, your graph doesn’t show the performance. We know from other examples that between 30W and 10W TDP the 6800U suffers around 30% reduction in performance. Now imagine if it could sustain its peak 80W.

 
Or buy a gaming laptop AKA one with Windows?.... I bought this really expensive wrench but it just sucks for hammering in nails. Do you think it's defective and I should return it? That's how these reviews with silly benchmarks are coming off as.

Exactly. Reviews like this do not make sense. The Air is designed around someone who has long periods of low intensity activity with infrequent and short bursts of high intensity activity. If you need sustained high intensity then the Air is not the right tool...

Also according to the chart the M2 consistently does better than the M1, but that isn't what is being talked about...
 
It’s because it throttles instantly. At peak power (peak possible clock) that CPU would consume over 60-80W which is why the PL2 limit kicks in after few milliseconds. Also, your graph doesn’t show the performance. We know from other examples that between 30W and 10W TDP the 6800U suffers around 30% reduction in performance. Now imagine if it could sustain its peak 80W.


According to this review PL1 is even 25W for the Zenbook S 13 OLED - so the PL2->PL1 throttling is not as drastic. However PL1 is lowered to 16W in battery mode and 12W in silent mode - which results into 35% performance drop. This 35% drop can also be seen in the video linked by Xiao_Xi.
The overall performance drop from PL2 = 30W to PL1 = 12W in silent mode is even close to 45%.

Now imagine giving this poor 6800U only 9W, which it will get best case in a fanless laptop. :)
 
Last edited:
This feels a lot like another form of the old "Apple is going to make MacOS as restricted as iOS because they're moving to Arm" and "iPad with M1 should run MacOS" threads... People get so caught up in the component that they're not thinking about the system.

I'm glad they're not caving to market ignorance and doing this, but Apple could probably sidestep half this crap if they continued using exactly the same part in different products just as they are now but labeled the SoC package with M2-U for the ultralights and M2-M for the iPads and M2 for the entry MBP and so forth. "That's the M2-M, it doesn't run MacOS". "That's the M2-U, it doesn't need a fan, you're confusing it with the M2".
 
That having said, the PL1 or TDP for passively cooled devices is below 10W (or 7-9W in actual devices in a laptop or tablet form factor) - this also applies to the Macbook Air. This means, after a certain amount of time the power limit drops from PL2 down to PL1 and the frequencies are reduced accordingly. As I said, conceptionally this is applicable to any device/CPU. Also to give you an idea, for phones the PL1 is usually below 4W.
For what it's worth, Apple Silicon doesn't seem to have anything akin to PL1/PL2. The one behavior which maps well onto Intel power/frequency control concepts is that maximum P core clock speed is permitted only while just one P core is active, and it drops off as more P cores wake up. However, unlike Intel, this is governed on a per-cluster basis rather than the whole chip, and the worst case frequency penalty with all cores active is quite small - only about 6% for M1 family chips.

Other than that, Apple's primary (only?) speed regulation mechanism seems to be a temperature feedback loop. As long as the chip's cool, you get full speed. As cores get close to whatever Apple chose as the maximum operating temperature, the feedback loop reduces clocks to keep temps from rising higher. It's at the very least a well tuned PID loop, maybe something more sophisticated - I haven't ever seen evidence of overshoot or ringing.

Because it's a feedback loop, exactly how much clocks roll back isn't predetermined, it's just whatever the loop's reaction to the current circumstances happens to be. You can actually watch a M1 Air get heat-soaked by plotting its frequency over time while running an all-cores CPU load - there's never a massive single step down in frequency akin to changing from PL2 to PL1, it just ramps down smoothly until the system reaches thermal equilibrium.
 
For what it's worth, Apple Silicon doesn't seem to have anything akin to PL1/PL2. The one behavior which maps well onto Intel power/frequency control concepts is that maximum P core clock speed is permitted only while just one P core is active, and it drops off as more P cores wake up. However, unlike Intel, this is governed on a per-cluster basis rather than the whole chip, and the worst case frequency penalty with all cores active is quite small - only about 6% for M1 family chips.

Other than that, Apple's primary (only?) speed regulation mechanism seems to be a temperature feedback loop. As long as the chip's cool, you get full speed. As cores get close to whatever Apple chose as the maximum operating temperature, the feedback loop reduces clocks to keep temps from rising higher. It's at the very least a well tuned PID loop, maybe something more sophisticated - I haven't ever seen evidence of overshoot or ringing.

Because it's a feedback loop, exactly how much clocks roll back isn't predetermined, it's just whatever the loop's reaction to the current circumstances happens to be. You can actually watch a M1 Air get heat-soaked by plotting its frequency over time while running an all-cores CPU load - there's never a massive single step down in frequency akin to changing from PL2 to PL1, it just ramps down smoothly until the system reaches thermal equilibrium.

Great explanation! Unfortunately, folks are used to Intel marketing and are applying their preconceived notions to Apple's hardware platform without discussing the details or relative merits of both approaches. And this results in a nonsense where a chip that has maximal power draw of 30W running at TDP of 9W is considered "bad throttling" where a chip that has maximal power draw of 130W running at 25W is considered a "marvel of engineering"
 
The MBA Mx has never been marketed as a gaming machine and that’s why it’s always been updated along with the base MBP Mx which is recommended by Apple ‘for those who need sustained performance’.

But honestly, serious pc gamers shouldn’t game on macs.

If you have enough $$$ to buy a MBA in the first place, you have enough money to buy the top of the range Xbox, PS or even the steam deck.

I mean it’s great that RE village is coming to the Mac but you have to think that this is more about triple game studies getting ready to work on the metal tech stack for the forthcoming Apple VR product (or maybe Apple TV is going to finally transform into a good cheap gaming console 🤷‍♂️).
 
  • Like
Reactions: russell_314
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.