My Cinebench Throttling tests M2 vs M1 MacBook Airs

[Unregistered 4U]

I’ve done 30 minute Cinebench runs with both my M1 and M2 MBA with a laptop cooler (two fans on a plastic shelf) and I found that the M1 doesn’t throttle (or very little) and the M2 still does throttle to an extent but maintains a higher performance than the M1. If you frequently run high intensity tasks but still want the portability advantages of the M2 MacBook Air, a laptop cooler will greatly improve sustained performance.

Edit: More details here: https://forums.macrumors.com/threads/how-hot-does-your-m2-mba-run-doing-what.2356187/post-31437451

Interesting information to know!

 

[Lounge vibes 05]

Other conclusion: The M2 is faster by 9% after 30 minutes. Cool. It’s also like 20% more expensive. Dollar-per-productivity the M1 is better.

not necessarily, the recent M2 Mac Mini is actually $100 cheaper than its M1 predecessor.
The extra $200 of the new Air is not related to the processor at all, and more than likely related to…
The bigger and brighter LCD.
The new 2022 enclosure.
The bigger battery.
The 1080P camera.
The new speaker design.
Plus inflation on all of the standard components.

 

[Mike Boreham]

I’ve done 30 minute Cinebench runs with both my M1 and M2 MBA with a laptop cooler (two fans on a plastic shelf) and I found that the M1 doesn’t throttle (or very little) and the M2 still does throttle to an extent but maintains a higher performance than the M1. If you frequently run high intensity tasks but still want the portability advantages of the M2 MacBook Air, a laptop cooler will greatly improve sustained performance.

Edit: More details here: https://forums.macrumors.com/threads/how-hot-does-your-m2-mba-run-doing-what.2356187/post-31437451

Thanks very interesting! I confess I am slightly surprised, because the bottom of the M2 Air is quite cool even after 30 mins of R23, so I would not have expected much heat to be removed by a fan, or ice pack. Unlike with my old 12" Intel MacBook which got very warm on the bottom, and resting it on an ice pack made a noticeable difference.

Your laptop cooler result has definitely piqued my interest! While I agree in principle with the guy who said "if you need a laptop cooler with your MBA you should have an MBP", I know temperatures will only be pushed in the chess analysis program I use sometimes. So maybe it makes sense to use the laptop cooler for that, and enjoy the benefits of small light and fanless the other 99% of the time.

The Chess analysis engine pushes CPU straight to 100% but I can change the number of threads. I am fine with the performance level even after throttling has kicked in, but I have a slight unease about whether say 30mins/week running at 95-100C would affect life of machine. I might reduce threads or use a laptop cooler....or maybe through caution to the winds and just use it full bore and rely on Applecare if it shortens life. Dunno, but definitely going to get a cooler as you did.

 

[jdb8167]

Your laptop cooler result has definitely piqued my interest! While I agree in principle with the guy who said "if you need a laptop cooler with your MBA you should have an MBP", I know temperatures will only be pushed in the chess analysis program I use sometimes. So maybe it makes sense to use the laptop cooler for that, and enjoy the benefits of small light and fanless the other 99% of the time.

The Chess analysis engine pushes CPU straight to 100% but I can change the number of threads. I am fine with the performance level even after throttling has kicked in, but I have a slight unease about whether say 30mins/week running at 95-100C would affect life of machine. I might reduce threads or use a laptop cooler....or maybe through caution to the winds and just use it full bore and rely on Applecare if it shortens life. Dunno, but definitely going to get a cooler as you did.

The cooler I bought was literally $20 on Amazon. Big and bulky (and relatively noisy) so it isn't anything you would want to leave your office but it is so cheap that just having one lying around seems worth it. I've never used it again after my testing though.

 

[Mike Boreham]

The cooler I bought was literally $20 on Amazon. Big and bulky (and relatively noisy) so it isn't anything you would want to leave your office but it is so cheap that just having one lying around seems worth it. I've never used it again after my testing though.

I lashed up a cooling system by using an angled open back laptop stand and two small desktop fans blowing across underneath from the same side. It made a substantial difference, but I don't fully understand all I saw.

After 17 runs nearly (30 mins) the R23 score had only dropped from 8632 (run 1) to 8077 (run 17) which is only 6.4% throttling But then it dropped to 7725 (4%) over the last two runs.

Another difference is that in the previous 30 minute runs the average CPU temp had dropped to around 95 by the end from 105 at the start. In this lash up fan cooled run they stayed up at 102-105.

I decided then to drop the fans idea (wife was complaining) and just use it normally with my chess engine running full bore. I particularly wanted to see what the CPU temps did over time. This was a surprise too. They started out at 102-105, same as the Cinebench test, but after about five minutes dropped to about 80C, but CPU was still high.

Not sure I understand all that but I am basically reassured that my occasional use of the chess engine does not result in non-stop 105C CPU, so not pursuing laptop coolers.

The CPU % and CPU Core Average temp plots from iStat are below. Red is System, blue is User. So it looked like the User CPU dropped for last two runs, corresponding to the extra drop in Cinebench score. The Chess phase looked like it never had 100% user. And note that the sharp drop in CPU temp occurred at 18:17, but there is no corresponding change in the CPU used.

My head is spinning. Think I am just going to enjoy the machine, use the chess engine to the full and trust that Apple have done it right. I have only had the machine six days, so still in the new device test/poke/prod/understand phase.

 

[jdb8167]

I lashed up a cooling system by using an angled open back laptop stand and two small desktop fans blowing across underneath from the same side. It made a substantial difference, but I don't fully understand all I saw.

After 17 runs nearly (30 mins) the R23 score had only dropped from 8632 (run 1) to 8077 (run 17) which is only 6.4% throttling But then it dropped to 7725 (4%) over the last two runs.

Another difference is that in the previous 30 minute runs the average CPU temp had dropped to around 95 by the end from 105 at the start. In this lash up fan cooled run they stayed up at 102-105.

I decided then to drop the fans idea (wife was complaining) and just use it normally with my chess engine running full bore. I particularly wanted to see what the CPU temps did over time. This was a surprise too. They started out at 102-105, same as the Cinebench test, but after about five minutes dropped to about 80C, but CPU was still high.

Not sure I understand all that but I am basically reassured that my occasional use of the chess engine does not result in non-stop 105C CPU, so not pursuing laptop coolers.

The CPU % and CPU Core Average temp plots from iStat are below. Red is System, blue is User. So it looked like the User CPU dropped for last two runs, corresponding to the extra drop in Cinebench score. The Chess phase looked like it never had 100% user. And note that the sharp drop in CPU temp occurred at 18:17, but there is no corresponding change in the CPU used.

My head is spinning. Think I am just going to enjoy the machine, use the chess engine to the full and trust that Apple have done it right. I have only had the machine six days, so still in the new device test/poke/prod/understand phase.

View attachment 2147525View attachment 2147527

It's important to understand that the various applications that display SoC temperatures are just guessing on what part of the SoC the sensors are measuring. There is no documentation and the M2's sensor names are much more obscure than they were on the M1. I have my own menu bar temperature sensor application that I've played around with and there still seems to be no public information on which sensors are P-Core, E-Core, GPU etc. Like I said there are guesses but nothing concrete. That's why comparing temperatures coming from the SoC sensors between the M1 and M2 isn't very useful.

BTW, if anyone has come across a good reference on the M2 sensors, please post. I haven't had much luck finding a good reference since the M2 came out.

 

[Mike Boreham]

I have been doing some more extended testing with my chess engines and, together with the Cinebench testing, feel I have a reasonable understanding of how the machine behaves under this load, which I will do maybe once a fortnight, and am quite happy with the conclusions.

1. The machine protects itself by limiting the CPU core average to about 80C after 5-6 minutes, and stays at that level. (at least for 40 minutes I ran). It takes about 20 mins to slowly drop to 80C, faster initially.

2. Peak performance at the very start is e-core 2.4GHz, p-core 3.2Ghz

3. Fully throttled performance is e-core 2.2GHz, p-core 2.2Ghz (see screenshot of Mx Power Gadget after 30 minutes running)

4. The performance at the fully throttled state is about -15%, deduced from Cinebench CPU testing. (This is faster than my 2017 iMac 3.5GHz, which I had until M1, then M2 MBA, replaced it).

During the return period for my M2 MBA, I have thought about returning it for a 14" M1 MBP, but not any more. I could have more chess engine capability with a cooled MBP, but I am happy to have the advantages of the M2 MBA and give up the extra performance. Magnus Carlsen would probably have gone the other way

EDIT The above chess engine numbers have been done with the M2 MBA resting on an ordinary open back laptop stand, no fans. Prompted by @jdb8167 's post I am repeating with the M2 MBA flat on a wooden desk. The results are different. I did the original Cinebench test with Mac on wooden table.

 

[Mike Boreham]

EDIT The above chess engine numbers have been done with the M2 MBA resting on an ordinary open back laptop stand, no fans. Prompted by @jdb8167 's post I am repeating with the M2 MBA flat on a wooden desk. The results are different. I did the original Cinebench test with Mac on wooden table.

The steady state p-core values I see are:-

Flat on wooden desk 1.49 GHz

After putting on stand 1.97 Ghz

NB the 1.97Ghz should have been the same value as in my previous post ie 2.21Ghz but it never got there.

Anyway conclusion is that raising it up off the desk does make a difference, even without a fan.

 

[ilikewhey]

Clearly, if that is all you're concerned with and don't care about the better display, speakers, etc. The M2 was a redesign for which you got a lot. Now if you don't care, well you don't care, and on a dollar per productivity basis, you could do way better getting an older used machine.

fixed