I've tried voltageshift it does work to reduce the power but I don't think it can be used to 'increase' the limit.
Feel free to try though.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
2020 Air Heatsink Modification Thread
- Thread starter ilikewhey
- Start date
- Sort by reaction score
Feel free to try though.
Did you tried it on windows using boot camp? Cause I know that older models like the MBP 15" 2015 had configurations that were just a slight overclock from one to the other, and if you used Intel XTU on windows you could overclock that few megahertz. So yesterday I thought "and if we could do something similar but this time just increasing the max PL1 tdp to the 20W and get stable frequencies above 2.5ghz".
I have a Gpd win max that has the i5-1035G7 and if I lock the PL1 at 12W and PL2 at 25W with the time burst at 10 seconds I basically get the same result as a modded MBA air core i5.
Also, the GPU performance on the MBA Core i5 and i7 are suffering a huge penalty because it reaches max TDP to the package and it won't boost to the max frequency.
I did, yes - also 'this device is not compatible'. Throttlestop also won't let you change the TDP on Windows. These chips are locked down much more tightly than older chips (probably for good reason).
I can't tell you what to do, but I really wouldn't advise finding a way to up the TDP of a processor that's designed for burst performance at 12w. It's your money, but this computer (with mods, and for some people, without them) performs extremely well for a passively-cooled Y-series chip.
I don't know how many times this needs to be stated in this thread, but the Macbook Air is not the Macbook Pro for half the price 'if we can just unlock the performance'. The limits in the MBA's various Ice Lake chips are imposed by Apple and Intel for good reason (business reasons, from their point of view, but probably also 'we don't want people's passively cooled CPU's failing after six months.')
If you need or even want stable frequencies above 2.5ghz, you need a computer (Apple or PC, doesn't matter) that's designed to handle that. The Macbook Air definitely, definitely is not.
No frozen vegetables were hurt during testing 🙂
i5/16/256
0.4mm shim - gelid paste
gelid pad 1,5mm
Just wanted to reach 1200/4000 🙂
room temp ~20
fan on auto ~2800
machine on table
battery only ~90%
back to back
1180+
3950+
wasn’t a fluke
after idle for 5-10min
Thanks to all for the great collab in here! 👍
i5/16/256
0.4mm shim - gelid paste
gelid pad 1,5mm
Just wanted to reach 1200/4000 🙂
room temp ~20
fan on auto ~2800
machine on table
battery only ~90%
back to back
1180+
3950+
wasn’t a fluke
after idle for 5-10min
Thanks to all for the great collab in here! 👍
Last edited:
I know, I have a PC tower for my work on home, I am a student of eletronic engineering in Portugal, and I really need upgrade my laptop for outdoors work, I want a Mac because of the versatility of software and the awesome display, keyboard and speakers, but I have a huge issue in my country, the prices od MBPs are hugely inflated, the base spec 8th gen MBP 13 2020 is 1600 euro, and the core i5 10th gen MBA with 16gb ram is around 1250 euro student price. That is a huge investment and I need to consider all the variables, if the MBA could boost a little more and the GPU could start boosting as it should, than it would be perfect buy for my wallet and needs
[automerge]1592222731[/automerge]
Can you tell us Cinebench results? Cause I know that Geekbench isn't as long as Cinebench and if you break 1200 cb on MC than this mod could put MBA 2020 on par with MBP 13" 2019
If you are on a tight budget and dont need mac os just get a widows machine. Do the layout programms run on mac os? Layouting and FPGA programming ist mostly done on windows machines.
@Guilherme Oliveira im getting 1155 without reaching 100 degrees and throttle limit in cinebench with an i5 16 256
@Guilherme Oliveira im getting 1155 without reaching 100 degrees and throttle limit in cinebench with an i5 16 256
Last edited:
MacBook Air 2020 USER reviews
I just got my MBA 2020 i5 16GB RAM. And I applied the voltageshift with below value: ./voltageshift write 0x610 0x4380c800dd8038 438 0c8 00 dd 8 038 0c8 == 200 dec == (200/8) 25W 038 == 56 dec == (56/8) 7W basically limiting the turbo to 25W and normal to 7W. The kext and binary, I download...
forums.macrumors.com
Not quite 1200 but close enough
1182
i would say i get a pretty persistant
~1150
@nill1234 and @hmhmmhmmm? , that are very nice scores, but not what I'm looking for unfortunately. I am looking for some windows machines too but I'm tired to have always compromises on that machines, like the agressive tuning on gpu power from Dell, the buggy bios on HP's ultrabooks, the "low res 1080p" screens from all the base models, and the others haven't that perfect construction materials that Apple has. So I will try to wait for a promotion on the 8th gen MBP 2020 or maybe see a refurbished one.
The 8th-gen MBP 13 scores similarly or slightly better than the 2020 MacBook Air in many GPU tests (3D Mark and 3D Mark 11). The 8th-gen MBP 13 has a 15W CPU and a heat pipe connecting the fan to the heat sink.
Hello friends, in the end you will see me around here because I have sold the mac for € 400 damaged and I have given Hauwei a chance to buy something cheap and with a Ryzen that is cooler than Intel ... hahahaha it is an option since in I always have Windows and Mac at home, I still have a Macbook 12 "the Iphones, the Apple wacth, 2 ipad ... in other words, I have not yet left Apple only to pay two laptops now ...
I have had a great time and I hope that my insistence and daring have served some purpose.
See you soon!!!!
I have had a great time and I hope that my insistence and daring have served some purpose.
See you soon!!!!
Attachments
I think I might reverse my 1.5mm pad mod and do the shim. I have weekly DnD games over Skype and the chasis got incredibly hot. I had it on a cooling pad to avoid contact with my warm legs but I didn't turn on the fans straight. Maybe next time I'll have them on from the beginning but I reckon maybe the shim might be more helpful when playing on my lap. Just feel uncomfortable doing it before the on year warranty but maybe I can reverse it easily.
@srkirt glad you were able to sell you MacBook Air!
@srkirt glad you were able to sell you MacBook Air!
So, I've finally got to repasting my MBA'20 i7/16 with Kryonaut and replacing my original shim with the one from @Loog.
Loog has sent me two versions, 0.5mm and 0.3mm, so I've tested both to see which performed better. Before the installation I've polished both shims to practically mirror-like state just for the sake of removing all the potential imperfections out of the equation.
If anyone wants to know, I've used P2000, P2500, P3000, P5000 and P7000 grit sandpaper one after another. I've also polished the heatsink, but just a little bit.
Kryonaut thermal paste is a bitch to work with due to viscosity, compared to Arctic Silver. It takes longer to apply thin even layer on top of the CPU die, and it gets everywhere around the CPU, but since everyone says it's the best, I had to try.
Testing process remained the same, clean macOS, nothing running except for the Intel Power Gadget and Geekbench 5.1.1. Waiting for the CPU to cool down to ~40°C and then run the test. I've run each test 3 times, so below is the average of 3 runs.
Now, the numbers. Here's the comparison table:
As you can see, 0.5mm shim performs slightly worse than 0.3mm, but all in all the difference in negligible. I do like that the top multi-core result is the whopping 50% increase compared to the pre-mod!
Thermally the new 0.3mm shim+Kryonaut paste performs better, which can be seen on the following Cinebench R20 temperature curves comparison:
In fact, during the whole Cinebench R20 run the temperature never reached 94°C, and stayed below 90°C for the most of it (with the thermal pad of course). Compared to the pre-modded version running at constant 100°C I'd say it's a huge improvement.
One thing that bothered me is that I couldn't get my Geekbench multi-core score above 4000, while even the i5 models have recently managed to climb over it. I've run all the previous tests on macOS 10.15.4, so I didn't update to the 10.15.5 for the retesting to make sure it didn't affect the outcome. But after all the tests were done, I've updated the macOS and ran Geekbench 3 again a few times. Below is the average of 4 runs.
But one of those runs gave me the following result:
Finally! And it looks like the single-core performance in macOS 10.15.5 has slightly improved as well.
Loog has sent me two versions, 0.5mm and 0.3mm, so I've tested both to see which performed better. Before the installation I've polished both shims to practically mirror-like state just for the sake of removing all the potential imperfections out of the equation.
If anyone wants to know, I've used P2000, P2500, P3000, P5000 and P7000 grit sandpaper one after another. I've also polished the heatsink, but just a little bit.
Kryonaut thermal paste is a bitch to work with due to viscosity, compared to Arctic Silver. It takes longer to apply thin even layer on top of the CPU die, and it gets everywhere around the CPU, but since everyone says it's the best, I had to try.
Testing process remained the same, clean macOS, nothing running except for the Intel Power Gadget and Geekbench 5.1.1. Waiting for the CPU to cool down to ~40°C and then run the test. I've run each test 3 times, so below is the average of 3 runs.
Now, the numbers. Here's the comparison table:
| Pre-mod | Original shim, Arctic Silver, thermal pad | New shim (0.5mm), Kryonaut paste, thermal pad | New shim (0.3mm), Kryonaut paste, thermal pad | |
| Geekbench (Single-core) | 1169 | 1265 | 1261 | 1269 |
| Geekbench (Multi-core) | 2621 | 3849 | 3804 | 3939 |
| Cinebench R20 | 903 | 1139 | 1144 | 1169 |
As you can see, 0.5mm shim performs slightly worse than 0.3mm, but all in all the difference in negligible. I do like that the top multi-core result is the whopping 50% increase compared to the pre-mod!
Thermally the new 0.3mm shim+Kryonaut paste performs better, which can be seen on the following Cinebench R20 temperature curves comparison:
In fact, during the whole Cinebench R20 run the temperature never reached 94°C, and stayed below 90°C for the most of it (with the thermal pad of course). Compared to the pre-modded version running at constant 100°C I'd say it's a huge improvement.
One thing that bothered me is that I couldn't get my Geekbench multi-core score above 4000, while even the i5 models have recently managed to climb over it. I've run all the previous tests on macOS 10.15.4, so I didn't update to the 10.15.5 for the retesting to make sure it didn't affect the outcome. But after all the tests were done, I've updated the macOS and ran Geekbench 3 again a few times. Below is the average of 4 runs.
| macOS 10.15.4 | macOS 10.15.5 | |
| Geekbench (Single-core) | 1269 | 1300 |
| Geekbench (Multi-core) | 3939 | 3994 |
But one of those runs gave me the following result:
Finally! And it looks like the single-core performance in macOS 10.15.5 has slightly improved as well.
Last edited:
That is not entirely true. The base macbook pro can be unlocked to 28W (using CPU+GPU tasks) without overheating. The 15W are for CPU only tasks. Other aspect is that many GPU accelerated workloads are very long and the base MBP can easely be two times faster than the MBA core i5, as I said, the G7 graphics are hugely bottlenecked by the termals, Cpu frequency and TDP limits for the package. For an example I asked on the emulation discord server if a friend of mine could do Wii U emulation using CEMU on his core i5 MBA 2020, and it started very well cause the Package was boosting to 25W, but when it hitted 100ºC it limited to 12W and the performance dropped to a point that we couldn't make the emulator stable to run simple homebrew code. The channel "The Phaw" has a video teaching how Ice Lake is a Power dependent architecture, cause when compared to Kaby Lake, when a Kaby Lake was running at 5 to 7W it could manage the tasks in a way that the performance could get a hit but you still had stability, on Ice Lake if you try to run the dual core ate the same 5 to 7W you lose almost all stability and in the graphics workloads you could see it running worst than weaker Kaby Lake iGPUs.
Intel when was making this new graphics architecture with Raja Koduri, was thinking on defeating AMD's Vega iGPUs and it came with a cost that is power consumption. A unlocked G7 graphics on windows machines with sufficient cooling can scale performance beyond 50W, Intel them selves made a graph explaining that you need at least 20W to it perform near Vega 8 and at 60W it could beat a 1050 Max-Q
Nice! Thanks for the detailed results on shim thickness and thermal paste. Btw, what was the thickness of your original shims? (I'm guessing 0.3mm)
I've also quickly played around with thermal pads of different thickness: 1mm, 1.5mm and 4mm. The 1.5mm is the clear winner.
[automerge]1592258757[/automerge]
[automerge]1592259467[/automerge]
[automerge]1592258757[/automerge]
Yeah, original ones were 0.3mm, 15x15, one full, one cut to size.
[automerge]1592259467[/automerge]
I have to admit, this does sound kinda nice 😎
Last edited:
Would one of this be possible to install on MBA? I think the screws layout is not the same, but, it could be possible to drill the cooler.
[automerge]1592264788[/automerge]
On the other photo is missing this black thing that allows better heat transfer
[automerge]1592264788[/automerge]
On the other photo is missing this black thing that allows better heat transfer
I have the same model as you and my GeekBench 5 scores consistently come out at ~1200 single and 3350 multi. Best is 1250 and 3519. That is with no modifications. How come your pre-mod multi-core is so low?
Last edited:
He mentioned he ran them 3 times back to back and took the average. Are you doing the same?
Are you responding to my post?
2019 and 2020 MBP 13s have Intel i5-8257U processors. They have a thermal design power (TDP) of 15W. The i5-8257U does not have configurable TDP and is not unlocked. The CPU and its integrated GPU share power. A manufacturer might set one of the power limits to be the same as the TDP, but it does not have to. How much power a processor actually draws at any given time depends on the power limits, the temperature and other circumstances. For example, in the MBP 13, the processor can draw 40W briefly in a CPU load, but its TDP is 15W.
Last edited:
Bin lottery. Some chips just work better than others. My un-modded scores are the same as yours.
There has also been some speculation that the “early” run of these computers had a rather less than efficient thermal paste application. Some of the disassembly pictures show a thick paste, with cracks even.
My unit is recent, delivered new less than a week ago. My shim, 0.3mm, from @Looq, arrived today. I hope to do the mods Friday, when I have a block of time to work with it. Probable shim polishing on Thursday.
Loving my MBA, although I have not done any Zoom meetings on it yet.
There has also been some speculation that the “early” run of these computers had a rather less than efficient thermal paste application. Some of the disassembly pictures show a thick paste, with cracks even.
My unit is recent, delivered new less than a week ago. My shim, 0.3mm, from @Looq, arrived today. I hope to do the mods Friday, when I have a block of time to work with it. Probable shim polishing on Thursday.
Loving my MBA, although I have not done any Zoom meetings on it yet.
did you fully tighten the screw? i am worried that it might pressure the heatsink/motherboard even for the 1.5mm thickness