I pulled it apart every 6-8 months to clean out the heatsinks and check how it was doing. I never saw temps go up over time but I did have to add a bit more every other time to make good contact. If I was not pulling it apart I doubt I would have had to add more.
Ok then I will try it without nickel plating first. If it doesnt work, I can still do it later
I have been using liquid metal on both my GPU and my old laptop. One thing I recommend when applying liquid metal on GPUs and CPUs that have caps close to the GPU/CPU is applying clear nail polish on the caps. That saved me from accidentally shorting one of the caps when I spilled a small amount on one of the caps.
When I watch folks on YouTube using liquid metal I’m aghast they use a q-tip. Quite probably leaving cotton fibres etc. in the mix.
Just looks wrong to me.
1 Like
Had some stuff come up, but here is the results from my testing.
Testing setup:
All tests were performed with the Framework laptop flat on a wood table.
Battery fully charged, connected to 60w Framework charger.
Used canned air to get any dust out of the heatsink before starting.
Windows 10 Pro 21H2, power profile set to max performance
i7-1165G7
32GB (2x16GB) 3200 running at 2666
Samsung 970 Pro 1TB
Stock Paste: Testing was done before removing the heatsink for the first time.
Idle: 43-45C, 2.6-5W, 1300MHz. let sit for ~15 minutes with only aida64 open to establish baseline. (did not have the CPU power line enabled for this screenshot)
CPU burn: 12.5 minute CPU only stress. temp leveled out at 91C, CPU clock held 4100MHz the whole time. CPU reported 23-24W no throttling.
let sit for ~10 minutes to come back down to base temperature.
CPU + FPU burn: 12 min CPU and FPU stress. quickly hit 100C but never went over, clock bounced between 3100-3600mhz, CPU reported 26-28W, heavy throttling once the temp hit 100C
Kryonaut Paste: Its just thermal paste, super easy to apply. cleaned the stock paste off with a paper towel and then cleaned the dies and heatsink with alcohol pads.
Idle: 40-42C. After applying past I let the laptop sit for ~15 minutes with only aida64 open to establish the baseline.
CPU burn: 13 minutes of CPU only stress. leveled out at 78-80C. CPU clock held 4100 MHz, 24w. no throttling.
let sit for ~10 minutes to come back down to base temperature.
CPU + FPU burn: 12 minutes of CPU and FPU stress. Quickly ramped up to hit 100C and throttled, once the fan caught up it dropped to 90-95C and stopped throttling hard. when the temp hit 95C it would throttle slightly and drop back to 90C. CPU was doing 3400-3600MHz, and 27-28w.
Honestly very impressive improvement over stock without the risk of damage from liquid metal. If you lift the laptop up slightly to improve the air intake you can get zero throttling once the fan ramps up under full load.
Conductonaut Liquid Metal: TBH I forgot how much this stuff sucks to apply. took me an hour to get it right and both the CPU and GPU die with the proper amount.
idle: once again, let it sit for a bit to get an idle baseline. 40-41C
CPU burn: Ran for 15 minutes. Leveled out at 72C, 4100 MHz.
let sit for ~10 minutes to come back down to base temperature.
CPU + FPU burn: 12 minutes, hit 95C and throttled, maintained 82C for the rest of the run. pretty much the same story as the Kryonaut but it only hit the limiter once at the start of the run.
I’m a bit surprised by the last result. The jump from the stock paste to the Kryonaut was 10C better on the CPU only run and 10C again on the CPU and FPU run. The jump to the Conductonaut for the CPU run was almost another 10C down, the CPU+FPU run was down 8C. I guess this might be the upper limit of the cooler itself or my application of the liquid metal was not perfect.
After all this I will be removing the liquid metal and putting Kryonaut back on. I’m not really going to be using this laptop hard in the long run.
5 Likes
Did you control Fan speeds (is There even an Option on the framework)?
If not, how was the difference in Sound volume between the different TIMs?
Thanks for your work. Maybe I will settle for kryonaut, it seems to be a great improvement over the Stock paste
@devryd I let the stock fan profile do its thing. I’m not sure how to adjust it yet because I have not had the need. I’m in a fairly noisy environment and have no good way to gauge the sound level. At idle the fan is pretty much inaudible, any time the temp is over 70C for ~30 seconds its ramped up to full speed as far as I can tell.
Anyone know what the fan curve and delay is? Is it in the EC code?
As far as I know, the fan doest run at idle at all.
I was wondering if there was a difference at the high end, because fan speed is often related to temperature. If the fan was spinning slower with the liquid metal, the delta in temperature would be even higher if the fan speed was fixed.
And what was your ambient temperature?
Warm-ish, I don’t have a way to measure it with me. Its been constant all day though.
That explains it.
I have a 20°C ambient temperature and mine idles at around 37°C with the stock thermal paste.
Edit: This is the temperature for the package, the cores idle at around 32-33°C
Edit 2: I just noticed, that you have the i7 variant. That explains the difference in temperature
So I just switched my i5 framework to using conductonaut as the laptop was overheating after playing daggerfall unity for too long. Unfortunately I couldn’t find a good benchmark suite on gentoo to capture good data like AIDA64 so my data is pretty much just running sensors with a stress-ng burnin.
Anyways with stock I was getting thermal throttling with 1 core hitting 100c and the others around 85c-90c. Obviously with the thermal throttling the fan was autoed to 100%
After switching to liquid metal I’m getting much better temps. This was the first time I’ve ever done a liquid metal install (Gamers Nexus has pretty good videos on what to do) and I’m quite impressed with the results. With liquid metal and the same burnin test I’m seeing all cores around 75c and fan speeds are noticeably lower and quieter.
If you are willing to risk bricking your mainboard for better temps and quieter loads, I would recommend taking the plunge
1 Like
Its a little wierd, that your framework runs so hot. I cant get mine even close to throtteling with synthetic loads. Gaming should be less stressful that stresstest tools. From which batch is your framework?
You are running linux, if I see that correctly. Did you change any power settings for the CPU?
I just ran prime95 and furmark and never got above 80°C with my i5 framework
I was in Batch 4, the existing thermal paste appeared to be applied correctly when I replaced it with liquid metal. If anything there was more paste than necessary but that’s way better than not enough paste.
My kernel only has ‘performance’ and ‘schedutil’ cpufreq governors built and ‘performance’ is enabled by default. P-states are enabled and stress-ng is taking advantage of that with the --ignite-cpu flag
I am not familiär with that Tool. How much Power is the CPU allowed to draw during the stress test?
Performance will pin the cpufreq at max which I guess is 4.2GHz for these i5’s. As far as I’m aware there is nothing on this machine that would prevent the CPU from running as fast as it can
I assume stress-ng with ignite is more aggressive than prime95. Probably closer to IntelBurnTest
Linus Tech Tips made a video a few weeks ago where they tested different Stress test tools. Why Overheat your CPU on Purpose? - YouTube
They found that prime95 put out the most heat on intel. This test was done on windows though, which means I have no comparison for stress-ng.
Can you get a power reading from your system?