I’d feel comfortable guessing it’s not genuine honeywell… but if it meets your performance needs then it’s good enough anyways
Well that settles it then, I’m buying some damn PTM7950 and getting rid of the LM.
That looks tipycal for PTM. For Burnin the Materials needs to be at above 45C and Cool down to Room Temp again. Over about 10Cycles it gets its End thickness and Distribution.
For anyone interested I bought this one:
https://www.amazon.com/Thermal-Grizzly-Performance-electrically-Electronics/dp/B0DB84CQW6
Not “technically” PTM 7950 but reviews are pretty sure its just relabled 7950… seemed safer to get this than some unknown from amazon.
Going to test this on the steam deck and the dGPU first. Then after a couple months maybe Ill put it on the CPU.
Ill be interested to see your score in ~4 weeks.
You have the 7940 right?
My 7840 has dropped about 1000 points in that time - I would hope that we dont see the same degradation with the PTM.
I’m starting to feel set on getting rid of the LM if this is why it’s happening, too. What do I have to be worried about to make the replacement go smoothly? I’ve never played with liquid metal before and I really, really, REALLY don’t wanna go and short my nearly-1k board.
So the interesting thing about the liquid metal solution Framework decided to use here is that it’s actually a phase change pad. It’s kinda like PTM7950, except made out of an indium-tin-bismuth mix that melts at around 58°C, as described by Framework themselves:
Instead, on Framework Laptop 16, we used a liquid metal thermal interface on the CPU. Liquid metal is exactly what it sounds like: a 100% metal sheet made up of indium, tin, and bismuth that turns from solid into liquid at around 58°C as the CPU heats up, filling any air gaps completely. Being metal, it is extremely thermally conductive, rated at 72W/mK, substantially better than the 5-10W/mK that is typical for traditional paste. It also doesn’t dry out or pump out over time. We’re using Coollaboratory’s Liquid MetalPad through their Taiwan-based partner CCHUAN.
As far as safety, this is what they had to say:
You might be thinking, is it safe to have highly electrically conductive liquid inside of a portable device? The answer in this case is yes, as we’ve designed the thermal system to contain it at multiple levels. First, because AMD’s processors have small capacitors directly next to the CPU dies, we dispense an insulative glue layer using a robotic fixture during Mainboard assembly to cover them. Next, an etched pattern in the surface of the vapor chamber holds the liquid metal through tension. Finally, a dual foam barrier around the CPU is squeezed between the processor package and the vapor chamber copper plate, preventing any liquid metal from escaping. All of this results in a very cool thermal solution and great CPU performance.
Both quotes were taken from this Deep Dive post from the company: Framework | Framework Laptop 16 Deep Dive - Liquid Metal
Multiple users in this thread have talked about how they removed the LM pad and it doesn’t seem to be too much of a hassle. Just make sure to clean the die and heatsink thoroughly before applying a new thermal compound and reinstalling the heatsink.
Mine is 31w max and 12983 R23 score (7840), I suppose it’s pretty bad, right?, anyone with a small guide of how to remove the LM and if I changed it for thermal pads, how thick it need to be?
@knipp30 same. Yes, I have the 7940.
I hope it won’t change.
A cold, single run resulted in 16944 points today,
More details:
Edit:
A second test resulted in a max of 70.725 W for the CPU Package Power and 16752 pts.
What operating system? In Windows, I found that even though the wattage and temps were no better, when running on “balanced” power mode vs. “Best Performance,” my scores were lower. When I switched to best performance, my scores went up from around 12,800 to just over 14,000. Granted, I sent Framework all that info and screenshots and they are still replacing my mainboard. But it’s just a thought.
The Cpu is throttled to max 50w in Balanced, as to why it doesn’t run that hard into the Thermal Throttle. It needs time to recover thus you get less Points because it tries to suck as much Power as it can until it throttles again hard.
18 days later. i had a health day good enough to sit at my desk and twiddle with laptop internals. ran cinebench r23 before and after i attempted retensioning screws. results were essentially the same both sessions today, about 13.7 k after it stabilized from extensive runs. all testing was in performance mode.
weirdly looking at the power consumption figures in turbostat and lact (it’s the only thing i found that could give me a graph without trying to parse turbostat data into a spreadsheet or something), it SEEMS to be using less power while getting a higher score than 18 days ago.
the major observable differences in circumstance today are the laptop is on a desk instead of my laptop tray (which is attached to a vesa mount gas spring arm) so there is slightly better airflow (i haven’t figured out how get a spacer to stay in place to lift the laptop from the tray securely), and it’s in the same room as the air conditioner, so ambient temp is a bit lower. but i feel like if my problem was the thermal interface in the cooler, a half degree C of ambient temperature wouldn’t have a substantial effect.
still haven’t found a way to get per-core temperatures to show up. i still have compiling a newer kernel on my task list, it’s just a bit on the daunting side.
Okay well this is unexpected. That’s a crazy good Score with a formerly pretty bad Result.
i’m about 90% positive the difference is due to airflow underneath the chassis. i did a quick run (i’m tired and didn’t want to spend too much time staring at it) with it propped up in a… less than sustainable fashion, in the laptop tray, and the initial scores were looking similar to the initial scores in the office.
so i’ll need to figure out a spacer of some kind… if i can come up with something that’ll fit within the locking mechanism, and won’t get easily wiggled loose.
amdgpu_top --gui
should give some nice graphs, but it looks like it requires wayland to run. I’m on X11 currently (openSUSE tumbleweed), and all I get is
$ sudo amdgpu_top --gui
[sudo] Passwort für root:
Failed to set up a graphics context (OpenGL).
winit EventLoopError: os error at /home/abuild/rpmbuild/BUILD/amdgpu_top-0.9.2/vendor/winit/src/platform_impl/linux/mod.rs:786: neither WAYLAND_DISPLAY nor WAYLAND_SOCKET nor DISPLAY is set.
thread 'main' panicked at crates/amdgpu_top_gui/src/lib.rs:236:9:
explicit panic
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
The version of amdgpu_top is v0.9.2 from the openSUSE repository (source package amdgpu_top-0.9.2-1.1)
it looks like 0.9.1 runs in gui mode fine for me in xorg on xubuntu 24.04-1 lts. graph length is a little short tho, wonder if that’s configurable somehow
note that you’re trying to run it with sudo
, which might have an issue trying to automatically inherit the DISPLAY
environment variable from your user account. try it without that?
edit: i had this app installed already because someone suggested it earlier, but i am slightly dubious of at least some of its stats, because it e.g. violently disagrees with all the other tools i have for measuring per-core clock rates
Ah okay, that’s it. Thank You!
Without sudo, it does run, but sadly it doesn’t visualize the CPU values.
I can’t tell how good the values are, but my fans start spinning full tilt as soon as the gui variant is running. It uses 93% CPU and the power consumption jumps from 20W to 50W according to my USB power meter.
Unfortunately, testing thermals on a new mainboard is not going to happen this weekend. I installed the new mainboard, and the keyboard and macropad didn’t work. I checked everything, inspected connectors, removed and reinstalled everything multiple times, but nothing. The trackpad worked in any position, but the keyboard and macropad didn’t work in any position. After a ton of testing and troubleshooting I switched back to the original mainboard and everything worked. I reached back out to support. We’ll see where it goes from here.
Well, after coming across this thread and seeing the issues others were having, I decided to try it out myself on my 7840HS batch 10 unit.
I ended up getting a R23 score of 14344 with the best being a 14355.
Hottest core was Core #6 at 100 C and the lowest being Core #0 at max 82.9 C.
Wattage started at 56.6W and at the end of the test, it was in the mid to upper 30W range.