M1 Air with Thermal Pad Mod

[Mike Boreham]

While we are waiting for others to participate...

Those figures, apart from the rounding to 2 minutes, sound reasonable. They indicate, as one would expect, that the actual performance on the last pass is inevitably less than the average performance over the whole test. The average never gets as low as the current performance, well not short of a very long test and even then only because of rounding error of the CB scores.

However, I was precise with my timings, taking them from the CineBench 'count-down clock' exactly, to the second, each time a pass was completed. As I indicated, I had a fair few dry-runs before my skills were up to speed.

If you repeat the test with precise timings you will find that (the first pass score + the calculated 2nd pass score according to my formula) divided by 2 = the CB given score after the second pass, and so on.

OK I am convinced ! You are correct.

I confirmed that CB does indeed report cumulative averages on multipass tests, by running five separate tests in quick succession and comparing them with the first five results in the 30 minute test in my post#89. With five separate tests there will be no averaging.

Column 1 is the 30 min test CB reported numbers
Column 2 is todays five separate quick succession
Column 3 is the cumulative average calculated from column 2

It is obvious that column 1 and column 3 are very close so the CB long tests report cumulative averages.

Thanks for straightening me out, now back to the main plot!

I need to reissue my two previous posts about throttling because I have drawn the wrong conclusions from CBs reported numbers. The actual throttling after 10, 20 and 30 mins is greater. IMHO a throttling test should report the amount of throttling at the end of the test not the average over ten minutes.....or both.

 

[Mike Boreham]

Not aware of a free one but TGPro is not expensive. I use iStat menus, more expensive.

I have done 3 x 10 minute continuous Cinebench runs in quick succession on my M1 MBA:
First run: 7492
10 mins: 7020 (-6%)
20 mins: 6774 (-10%)
30 mins: 6632 (-11%)

These were run with machine "as is", ie all background process left running.

Repeated this as a single 30 minute run. CB carries on starting runs within 30 minutes then finishes the last run it started, so total time was actually 32 minutes.

Very close to previous results. Not quite flat after 32 minutes. Note this graph does not have a true (zero) origin so makes throttling look much more than it is....only 10% after 30 mins with all eight cores on 100%.

The "pACC MTR 3" sensor fluctuated between 95C and 98C from about the third run onwards.

I guess Handbrake might load the M1 chip more than Cinebench.

Not planning on doing any throttling mods.



View attachment 1715569

The above results do not mean quite what I thought they meant. Thanks to Tim Millea in the above few posts I realise that Cinebench 10 min and 30 min tests report the average throttling during the 10 minutes or 30 minute test. They do not report the performance level after 10 mins or 30 mins, which is what I thought the numbers were,

So my number of 6% throttling after 10 minutes gives an idea of how much longer a 10 minutes task might take, but not the actual performance levels at the end of 10 minutes. I have done some more tests to find the actual performance level after 10, 20 and 30 minutes, not the averages:

After 10 minutes: -11%
After 20 minutes: -12.4%
After 30 minutes: -11.9%

The 20 and 30 min numbers are a bit odd, but thats what I saw. Maybe a background process affected the 20 min number. Anyway I think it fair to say that has flattened out by then.

I don't know how many other people have been misled by this Cinebench reporting, or whether the numbers that get bandied about are average over the period or performance at the end of the period. I guess it doesn't matter too much. The level of throttling is much lower than I expected.

 

[Tenkaykev]

I'm happy to run a test using my frankenkludge bathroom scales mod. Any opinions on an optimum siting for the M1 Air sitting above the cooling fan ? ( the rear of the 40mm fan opening in the glass is 25mm from the rear of the scales )
I'm thinking that lining up the rear of the Air with the rear edge of the platform would place the cooling fan pretty close to where the M1 processor sits.

 

[Mike Boreham]

Hi, this is my test run with Cinebench R23 and my M1 MacBook Air 8/7 with 16GB of RAM with the thermal pad mod. The numbers in the figure are the results recorded after individual, consecutive passes, so no averaging at all. The initial score was 7788 and the final score, after 20 passes (about 40 min), was 7129. It seems like the 10% performance droop observed by others, after a long periods of intensive use, is mostly what could be expected in the long run with no other mod than the thermal pad. With no mod I was getting a score of 6200 after the 30 min CB test (averaging).

As you can see in the figure, there are a slight disturbance after the fifteenth pass, that's because I raised the laptop for a few seconds to move it to another place. Its evident that slightly improving the flow below the laptop is enough to improve the performance of the computer (it's very temperature sensitive!). Using a fan, or even just raising the back of the laptop, may reduce considerably the thermal throttling.

Thanks, impressive.

1. A couple of observations, with a hot bottom plate due to the mod, it probably makes it more sensitive than the standard configuration to the kind of changes you mention. Some tests on the fanless 12" MacBook also showed different results for the machine resting on a granite surface compared to wood.

2. Your unmodifield 30 min average score of 6200 is quite a bit lower than my unmodified 512/16 result of 6777.

 

[Tenkaykev]

I'm not sure if my base Air has the thermal sponge. with regard to Cinebench R23, I've looked online but can't see any info on logging the performance as it is running the test. Is there a setting that I am missing? I was planning on running the Cinebench test for 60 minutes but would prefer to set it running and leave it, instead of constantly peering at the screen every couple of minutes. Iss there a preferred methodology for doing the test?
TIA
Tenkay

 

[denzdaniel]

Hey guys! I finally decided to do a thermal paste reapplication and used Thermal Grizzly and after doing it the M1 temp doesn't go pass 97C stays below hovers at 89-93C running 10 min CinebenchR23. Scores after 3 10 min runs averaging 7760! Points stays there it doesn't go below 7700....

I'm gonna do another Thermal Paste Reapplication and will use IC Diamond 24K Paste! It has 2,000-2,500 W/mK conductivity and will give better heat transfer...

I must say, our M1 MacBook Air is a great machine! Let's find a way to lengthen the life of it by making it cooler without a fan!

 

[timmillea]

My results are in.

Reminder - this is the base model 8GB/256GB M1 MacBook Air. I have conducted two sets of tests - pre and post thermal pad modification. In my case, this was a cheapo (£2.64 delivered)) 2W/m.K 2mm thermal pad cut to fit over the upper section of the heatsink only (as pictured). FYI, there was no pre-installed 'thermal sponge' attached to the lower case as owners of higher spec models have reported.

Just opened...

With thermal pad applied....

I ran CineBench R23 for a custom (minimum of) 60 mins noting the score, time, CPU and battery temperatures at each pass. The temperatures came from a beta version of Macs Fan Control compatible with the M1 - ironic as the MBA doesn't have any fans. The ambient temperature was 21-22 degrees C and there were no other apps, Bluetooth or wi-fi running. The MBA was plugged into the power and placed on my MDF desk top without any ventilation or additional heat-sink.

Here are the list of 15 (!) temperature sensors that Macs Fan Control picks up on the M1 MBA.

First here are the pre & post mod scores from CineBench over 61 minutes.

The headline is that even with this simplest, cheapest of mods, the MBA is 12.4% faster over a sustained period of 61 minutes.

As per my previous post, I calculated the per-pass performance from the rolling average that CineBench produces. This is a better indication as to the performance at any given point in time, during the 60+ minutes test.

Interestingly, with the thermal pad there was a second phase of throttling that occurred approximately 43 minutes in. The CPU temperature dropped from its regular 94ish degrees down to 85ish before performance began to rise again.

Finally here are the maximum battery temperatures as recorded by Mac Fans Control. This chart shows the temperatures through the testing and for another 20 minutes after.

Clearly the thermal pad causes the battery temperature to elevate around 6 degrees (almost certainly through contact with the bottom case) and this, and the other temperature sensors, are likely to be the cause of the second throttling back witnessed in the previous chart.

The take home is that you can gain a 12.4% improvement in sustained performance from an MBA by just adding a cheap thermal pad to the M1's heatsink and nothing else.

However the real potential must be from keeping the bottom case cooler either with an external fan, a desk with greater thermal inertia (e.g. marble, granite or steel), or simply popping the MBA on a raised, tilted stand to encourage natural thermal air dissipation.

I hope you have enjoyed my mod! Thanks again to all the other contributors here on this thread who made it all possible and so easy. All comments and questions welcome.

Tim.

 

[timmillea]

I thought I would listen to my own advice and dug out the stand I usually only use when travelling - it simply raises a laptop around 5" at the front and 5.5" at the back - for a more 'ergonomic experience' with Bluetooth keyboard & trackpad without an external display.

The one hour CB test came in at 7446, now over 16% faster than the unmodded MBA.

GeekBench scores are 1739 single core and 7661 multi-core. That makes this humble base-model, fanless, MacBook Air faster than any Mac recorded on the GeekBench website in single core and stepping on the toes of the 2019 Mac Pro 8-core (at £5499) in the multi-core tests.

More controversially, this multi-core GeekBench score is 4.4% faster than the M1 MacBook Pro's.

 

[denzdaniel]

I thought I would listen to my own advice and dug out the stand I usually only use when travelling - it simply raises a laptop around 5" at the front and 5.5" at the back - for a more 'ergonomic experience' with Bluetooth keyboard & trackpad without an external display.

The one hour CB test came in at 7446, now over 16% faster than the unmodded MBA.

GeekBench scores are 1739 single core and 7661 multi-core. That makes this humble base-model, fanless, MacBook Air faster than any Mac recorded on the GeekBench website in single core and stepping on the toes of the 2019 Mac Pro 8-core (at £5499) in the multi-core tests.

More controversially, this multi-core GeekBench score is 4.4% faster than the M1 MacBook Pro's.

That it's true! Our M1 Air is Fast! And we made it even faster after the MOD.

Congrats!

 

[Tenkaykev]

I've just done the heatsink pad mod to my base M1 MacBook Air. I used a 2mm pad sourced from Amazon UK which cost just over a fiver for a 100x100x2mm pad. It is also black in colour so it blends in somewhat with the existing heatsink.
I then ran the Cinebench R23 ten minute test and took readings after each run. It completed six runs so went slightly over the ten minutes as Cinebench will complete the existing run if it reaches the end of ten minutes midway through a run.
All apps closed except TG Pro and Cinebench of course. Before starting the tests the CPU/GPU temps were showing 28/30 C
MacBook on battery power.

Run 1 score 7746 - CPU/GPU 70/30 - Battery temp 26 C

Run 2 Score 7759 - CPU/GPU 77/30 - Battery temp 29 C

Run 3 Score 7766 - CPU/GPU 83/30 - Battery temp 31 C

Run 4 Score 7773 - CPU/GPU 84/30 - Battery Temp 33 C

Run 5 Score 7775 - CPU/GPU - 87/30 Battery temp 35 C

Run 6 and final score 7776 - CPU/GPU 90/30 Battery temp 36

This was mostly an academic exercise, I don't push my MacBook at all, just general streaming Netflix / Amazon , browsing and a little light photo editing. It took me ten minutes to do and cost me just over a fiver so all in all a worthwhile experiment.

EDIT: My Air did not have a thermal sponge fitted as on some models.

I'd be interested to hear any observations.

 

[PinkyMacGodess]

No I did not touch the thermal paste. Just cover off, apply thermal pads to heatsink, close cover. This mod is only for the thermal pad and is already beyond expectations.

So these pads fill up the space between the heatsink and the case? Does it make full contact? I would have thought replacing the thermal paste with the pad would have been a better choice. It would seem to me that packing the empty space in a notebook with a pad that might not make contact with the case is just potentially causing more issues. *shrug* If it's working, good, but it seems questionable to me.

 

[PinkyMacGodess]

The hot back shell with the thermal pad mod is not an indicator that battery is also hot, it is only showing that heat is escaping from the inside to the outside.

One more thing in response to the battery heat theory. The batteries get hot already when charging and you can feel that near the middle of the laptop. This thermal pad mod does not heat up that part of the back shell meaning the batteries are unlikely to be impacted by this mod.

But if the heat is just being transferred to the back, and it's not being actively removed, it's just getting hotter, and that heat will spread across the surface, eventually. It's just transferring the heat problem to another surface that needs to be cooled. Swapping paste for pads, and upping the fan speeds would be better to expel the heat, rather than just transfer, and retain, that heat. It's not getting rid of the heat, it's transferring it to another surface that can still spread the heat to other parts of the system.

 

[PinkyMacGodess]

I've just done the heatsink pad mod to my base M1 MacBook Air. I used a 2mm pad sourced from Amazon UK which cost just over a fiver for a 100x100x2mm pad. It is also black in colour so it blends in somewhat with the existing heatsink.
I then ran the Cinebench R23 ten minute test and took readings after each run. It completed six runs so went slightly over the ten minutes as Cinebench will complete the existing run if it reaches the end of ten minutes midway through a run.
All apps closed except TG Pro and Cinebench of course. Before starting the tests the CPU/GPU temps were showing 28/30 C
MacBook on battery power.

Run 1 score 7746 - CPU/GPU 70/30 - Battery temp 26 C

Run 2 Score 7759 - CPU/GPU 77/30 - Battery temp 29 C

Run 3 Score 7766 - CPU/GPU 83/30 - Battery temp 31 C

Run 4 Score 7773 - CPU/GPU 84/30 - Battery Temp 33 C

Run 5 Score 7775 - CPU/GPU - 87/30 Battery temp 35 C

Run 6 and final score 7776 - CPU/GPU 90/30 Battery temp 36

This was mostly an academic exercise, I don't push my MacBook at all, just general streaming Netflix / Amazon , browsing and a little light photo editing. It took me ten minutes to do and cost me just over a fiver so all in all a worthwhile experiment.

EDIT: My Air did not have a thermal sponge fitted as on some models.

View attachment 1720737


View attachment 1720738



I'd be interested to hear any observations.

Is there a fan in these new M1 systems?

 

[Tenkaykev]

Is there a fan in these new M1 systems?

Only on the M1 "PRO" . None of the M1 Airs have a cooling fan fitted. I have a 2020 i5 Air, that has a fan fitted and it regulary kicks in if doing any sort of intensive work. The M1 Air in contrast runs cooler without a fan and with about double the benchmark score

 

[Tenkaykev]

But if the heat is just being transferred to the back, and it's not being actively removed, it's just getting hotter, and that heat will spread across the surface, eventually. It's just transferring the heat problem to another surface that needs to be cooled. Swapping paste for pads, and upping the fan speeds would be better to expel the heat, rather than just transfer, and retain, that heat. It's not getting rid of the heat, it's transferring it to another surface that can still spread the heat to other parts of the system.

There aren't any fans in the M1 Air. At the moment the "heat sink" is an elongated metal plate that makes contact with the processor an dissipates the heat into carcass of the Air where thermal conductivity draws it out into the surrounding air. ( as I write the M1 processor is running at a temperature of 21 C, the temperature in the room is 19.5 C.)
What the pad is doing is speeding up the transfer of the heat from the processor into the case of the MacBook and into the surrounding air.

 

[timmillea]

Tenkaykev and I have taken the advice from earlier in this thread from downshiftdre,"I used 3mm for all of it. There is a step down in the heat sink so I cut 2 seperate pieces. If I were to do over again, I would use 2mm for the higher portion of heat sink and 3mm for the lower portion."

"But if the heat is just being transferred to the back, and it's not being actively removed, it's just getting hotter, and that heat will spread across the surface, eventually. It's just transferring the heat problem to another surface that needs to be cooled."

Thermodynamics dictates that even a passive cooling area, such as the lower case of the MBA, transfers heat according to size x temperature difference. No active cooling is required to make a huge difference.

"It would seem to me that packing the empty space in a notebook with a pad that might not make contact with the case is just potentially causing more issues. *shrug* If it's working, good, but it seems questionable to me."

See my charts above. On a stand here is the chart. A 16% improvement in performance over a sustained period of maximum load (61 mins +) is not to be sniffed at.

This is with just a cheapo 2mm thermal pad cut to fit over the heatsink in clamshell mode on a stand. I found today that by opening the MBA, i.e.not in clamshell mode, I can gain few more percent.

I am compressing my entire Blu-ray collection with Handbrake in HQ H.265 so these figures matter to me. GeekBench tells me that this base model MBA is almost 15 times faster than my 2015 MacBook it replaced. 15 times!

 

[Mike Boreham]

I've just done the heatsink pad mod to my base M1 MacBook Air. I used a 2mm pad sourced from Amazon UK which cost just over a fiver for a 100x100x2mm pad. It is also black in colour so it blends in somewhat with the existing heatsink.
I then ran the Cinebench R23 ten minute test and took readings after each run. It completed six runs so went slightly over the ten minutes as Cinebench will complete the existing run if it reaches the end of ten minutes midway through a run.
All apps closed except TG Pro and Cinebench of course. Before starting the tests the CPU/GPU temps were showing 28/30 C
MacBook on battery power.

Run 1 score 7746 - CPU/GPU 70/30 - Battery temp 26 C

Run 2 Score 7759 - CPU/GPU 77/30 - Battery temp 29 C

Run 3 Score 7766 - CPU/GPU 83/30 - Battery temp 31 C

Run 4 Score 7773 - CPU/GPU 84/30 - Battery Temp 33 C

Run 5 Score 7775 - CPU/GPU - 87/30 Battery temp 35 C

Run 6 and final score 7776 - CPU/GPU 90/30 Battery temp 36

This was mostly an academic exercise, I don't push my MacBook at all, just general streaming Netflix / Amazon , browsing and a little light photo editing. It took me ten minutes to do and cost me just over a fiver so all in all a worthwhile experiment.

EDIT: My Air did not have a thermal sponge fitted as on some models.

View attachment 1720737


View attachment 1720738



I'd be interested to hear any observations.

Thanks for doing and sharing.

I don't understand how the result actually improves on each of six runs? These are the CB reported scores after each of the runs in a 10min test, so each is the average of all the previous runs which means the actual improvement is greater.

No one else has shown results actual improvements with consecutive runs.

 

[timmillea]

I got similar results after the mod with the first three CineBench passes:

It did not make any sense to me either. The hour-long tests behaved more as expected so I did not draw attention to this.

 

[Tenkaykev]

Thanks for doing and sharing.

I don't understand how the result actually improves on each of six runs? These are the CB reported scores after each of the runs in a 10min test, so each is the average of all the previous runs which means the actual improvement is greater.

No one else has shown results actual improvements with consecutive runs.

Hi Mike,
I never gave that a thought. My methodology was to select the Cinebench10 minute stress test, open iStat menu so I could check the various temperatures and start the test. At the end of each run as the test picture screen went blank before commencing the next run, I hovered the mouse pointer over the left hand rankings window where my processor was displayed and noted down the score that appeared.
happy to run it again using a different methodology if you point me in the right direction. 👍

 

[Mike Boreham]

Hi Mike,
I never gave that a thought. My methodology was to select the Cinebench1o minute stress test, open iStat menu so I could check the various temperatures and start the test. At the end of each run as the test picture screen went blank before commencing the next run, I hovered the mouse pointer over the left hand rankings window where my processor was displayed and noted down the score that appeared.
happy to run it again using a different methodology if you point me in the right direction. 👍

Not doubting you but don't understand the physics. And Tim M saw same in his first three runs.

You could try doing five separate runs in quick succession, launching immediately after the last run completes. Should show a slightly bigger rise as there will be no averaging effect.

 

[Tenkaykev]

Not doubting you but don't understand the physics. And Tim M saw same in his first three runs.

You could try doing five separate runs in quick succession, launching immediately after the last run completes. Should show a slightly bigger rise as there will be no averaging effect.

I’m on it 😁

 

[timmillea]

After the calculations to 'de-average' the per pass CineBench scores:

 

[Tenkaykev]

OK, as Mike suggested 5 individual runs, each started just a few seconds after the previous ended so I could note down current temperatures.
Ambient CPU 20 C

Run 1 - 7805 CPU - 48 C, Battery - 26 C ( graphics stayed at 30 C for every test so I won't bother writing it down.)

Run 2 7801 CPU - 50 C, Battery - 29 C

Run 3 7806 CPU - 56 C, Battery - 31 C

Run 4 7814* CPU - 55 C, Battery 33 C

Run 5 7801 CPU 83 C, Battery 34 C

In the time I've taken to write this the ambient CPU is 25 C and the Battery temp is 29 C

* the score of 7814 in the league table of processors had an asterisk next to the number for some reason.

My Air is on my glass frankenscales, no forced cooling. the front feet are higher than the rear feet by about 5mm

 

[darkharbour]

I am keen to try this on my MBA M1 16/1TB once my thermal pads arrive next week, I'll be watching this thread closely.

 

[denzdaniel]

Hi Guys! Awesome thread! Can someone post your Temperature using an app? GPU/CPU/Battery and so on! Just curios! I am testing mine and doing more MODS adding Thermal pads to almost all parts of the logic board that draws heat!
I would like to know what is your temp. Thank you