Using Thermal Grizzly Carbonaut Thermal PAD on 2018 Mac Mini 6C i7 64GB Ram. Performance and Thermals.
I love the idea that I never have to change paste until the computer dies.
I have seen people test all kinds of paste and different heat pads with their Macs but never seen anyone use Thermal Grizzly Carbonaut.
So I have used Carbonaut before on different occasions with great success.
Mac Pro 2013:
I have used Carbonaut in my Mac Pro 2013 on both the CPU and 2 x D300 GPU with great thermals.
(I had to sand down the heatsink on the thermal core CPU side due to the fact that its poorly made (the copper pillars stood out to much) so a heat pad was not "optimal" but it really doesn't have to be perfect for it to work well enough.
See below what I did when "sanding" the CPU side of the thermal core: https://egpu.io/forums/postid/92644/
MacBook Pro Retina 2014 15 inch.
I have also used it on my MacBook Pro (both CPU + GPU) also with great thermals.
Here the CPU heatsink did not apply pressure (was not flat) across the whole CPU die so I sanded the heatsink flat.
Although it worked as well as paste without flattening the heatsink, the return in temperature after flattening the heatsink was not as high as expected so for me it was not really needed, but it evened out the temps a little.
Mac Mini 2018:
I have a Mini 6 core i7 8th gen CPU with 64GB ram.
I have seen a video on YT where someone tries to use a heatpad on a Mac mini, but he fails (not using it correctly) and its a different brand.
I needed windows (Parallels) with EGPU so M1 Mac was not for me at this time.
My usage case is very much a Geekbench one, mostly need single core with some shorter multicore processing for AutoCad, windows etc.
So I opened the Mini, cleaned it, cut a strip with the same size as the CPU die and reassembled it again.
After 2 weeks of use I can now give my verdict.
Geekbench scores are really nice and sustained frequency and temps for me are all good.
Now, your usage case (rendering, gaming, work), application, heatsink can all be different so it all depends, but for me I have always had great success with this PAD.
Geekbench results are without applying a fan custom curve (with iStat menus)
Original score with "Apple" applied paste.
After change to using Carbonaut PAD:
Run 4 = 1200 single core!
That is the 8th fastest Mac in the world right now in single core (using Geekbench anyhow)
And almost 1000 point extra in multicore (comparing to my original Apple applied paste), not bad at all!
Intel Power Gadget CPU Tests:
So for anyone thinking about paste / pad etc, for me this is a very nice option to use!
Now I have to go to never think of replying any paste for the rest of this computers life. That's nice
@itsage, Happy to share!
A tip, I put the nMP on a pillow on a table so the thermal core could lay on the table at an angle, so the CPU part of the thermal core is flat.
The PAD doesn't like to stay in place, so I found this to be a good solution when assembling (both CPU and GPU sides).
I have the 4C (did also test on a 8C cpu) and 2x D300.
I never got above about 100W for the CPU (and with that amount of cm2 area on the CPU die there where no issues and will work for 12C too is my expectation)
Don't know how much more power the D700 consumes vs D300, and the GPU die is a lot smaller cm2 area.
I did have to change my thermal Core on the nMP because the stock paste had caused the heat to burn a hole in the vapour chamber!
Found this image of Ebay where someone is selling this thermal core (that has a hole in it!) as "working" with the same issue as me.
My guess is that this is what happened to all failing GPUs that we have seen, perhaps the smaller power usage of the D300 managed even with a broken vapour chamber.
About "sanding" the thermal core of the nMP.
I did not sand it all flat, I just evened the copper pillars out, the copper is so thin it seems so I did not want to go too deep or doo to much.
I used the CPU and did a figure of 8 directly agains the thermal core, that way you can see how "bad" it is and "pair" it with your CPU, (The CPU isn't flat ether), the copper is so soft the CPU works as a fine "sandpaper" if you are careful..
In my case it didn't have to be perfect, with this big surface area on the CPU I found that the heat spreads a lot.
Perhaps the GPU side of the thermal core could use some flattening (depending on how it looks) but for me and the D300 it was not needed.
The GPU die I assume is flat.
Se how the pillars of the vapour chamber looks on the CPU and GPU side of the thermal core.
That heatsink is very different to a vapour chamber one, so it doesn't have the "pillars" and has a flat copper insert directly agains the CPU die.
See this image for a closeup of the heatsink from @itsage
It is taken from this post:
I just eyeballed it, it looked flat and I was prepared to disassemble the mini and adjust / test again if it didn't work properly, but it worked for me 🙂
@anotheregpuser, I wish Apple had done a better job with the thermal core on the 6,1. It's still one of the highest performing computers by volume. I think all three graphics card choices are effected by the insufficient cooling system. It's more on D700s than D500s than D300s. Graphics card B is always the culprit because it has to do many tasks at once (hosting a stick drive and all monitor outputs).
Adding an eGPU to take the load off the dGPUs is a perfect scenario for this aging Mac. The thermal core is more than capable of handling the CPU by itself. With better cooling materials in your case, I hope the nMP lasts a long time.
I am going to put carbonaut into my laptop. Currently have kryonaut on it but worried it has pumped out. Guess I'll find out. Will try to get some before/after data and post here.
@itsage, Yes I agree, I really like the 6.1 design and size, so too bad the development did not go forward.
I think it could have been made to be modular with both GPU and CPU.
I'm happy with the nMP+EGPU but actual performance in ex Parallels + Autocad does not reflect benchmarks.
As I need AutoCad for work it's being replaced by the Mac mini, there is a BIG difference in performance, even though benchmarks are the same, weird.
Otherwise I would have kept the nMP, I like it a lot!