Yeah you are Right it was Coolermaster.
And That the Heatpipes are not getting Hot is the Main Problem. They have to heat up in Seconds. If not the Thermal Resistance before the Heatpipe is to high. They have to Transfer the Heat away.
2 Likes
Hello Framework 16 owners!
I wanted to share my experience with my uneven CPU temps and the framework support team.
Since I received my F16 (with graphics module), I’ve been very annoyed by the poor thermal management of the machine. Full blasting the fans whenever I would put windows 11 in anything else than “power saving” mode. I was certain it couldn’t be normal and had to be a hardware issue.
After finding this thread I immediately did some proper test and found out that one of the core of my CPU was jumping to 100 degrees whenever I would use the cpu more then 25%. And that’s enough for the motherboard to push the fans to the max.
I contacted support and followed the (long) procedure to justify a test&repair shiping ticket. It tonk almost 3 weeks to come back (they are overwhelmed in the Germany repair center apparently) but it’s now returned all fixed. You probably guessed it but they simply changed the faulty liquid metal with thermal paste.
I encourage most of you to do the thermal managment test (the one with Cinebench and HWinfo) and contact Framework support if needed. It really is night and day difference when it’s fixed. Sure it takes time and energy to do all the back and forth with the support team but in my experience it’s just procedure and they are not trying to minimize the issue.
Thanks
7 Likes
Just got my PTM7958 from Framework, followed the guide for replacement. My Liquid Metal had mostly run to one corner with a super thin layer with what looked like scorch marks on the CPU. Was meticulous about getting the CPU die and the heatsink clean. While I didn’t actually look at temps per core in HWInfo before, I do notice the fans ramp up more gradually. Running Cinebench R23 on my first run i got around 15500 (didn’t take a pic). Second run I had HWInfo running and got lower. But I noticed there are still a couple of cores with a pretty high temp delta. Something I should contact support about? Even though I’m prob out of warranty period?
The ghostly looking mark on the CPU die is before I did the final cleanup, it’s gone now.
4 Likes
Nicely done, much shiny!!!
How much power is it drawing? HWiNFO shows my CPU Package Power peaks near 60W when starting a benchmark and steady-state at 45W (at least, it’s steady-state at 45W for my 7940HS. I assume that’s the same spec for your 7840HS?).
If you aren’t getting 45W then I’d definitely contact support. If you are getting 45W and you’re getting roughly expected performance for a 7840HS, I don’t see much point in contacting support.
Anyone able to show the cpu frequency when doing a stress-test?
When I launch a stress -c 16, and doing so - my frequency stays around 3370Mhz for all cores.
Command I use is:
watch -n.5 "grep '^cpu MHz' /proc/cpuinfo "
Sadly sensors does not show me the core temp per se.
Just want to evaluate if I replace the metal-liquit with the replacement pad.
Try “amdgpu_top”.
It is on github.
Are you sure you’re in max power mode? Idk what it’s called in Linux. Nothing you’re showing so far indicates any thermal issues to me. But that’s because you aren’t looking at the right metrics.
If you aren’t getting 100C on any cores and/or aren’t drawing at least 45W in steady state, then I believe that means you’re in an efficiency or power saving mode. Could be due to using a 3rd-party charger too, at least for my system on Windows I won’t get max performance even in Windows Performance mode unless I’m specifically plugged in with the Framework charger and power cord. Third party ones should work with proper ratings, but you need proper ratings on not just the charger but also the cable.
I’m pretty sure, yes.
Using balanced mode provides power when required (my day to day use case when on power).
System is connected to a Dell WD19TB dock, with 180W power supply delivering 90W to non Dell systems. So that should be plenty enough.
I can try this WE (long WE) to plug in the 180 provided Framework power supply and putting it all in performance mode. But I remember doing that already with the same results some months ago.
Looks like CPU sometimes gets stuck in some sort of low power mode - #10 by Obasav is recommending to use s-tui to see thermals as well as temps.
And as you can see from other posts in the forum eg 7700S won't go past 35W - #8 by Jimster480, it’s no guarantee that any 3rd-party supply will guarantee full power. There’s even a 240W supply that causes power issues. Only the 180W supply is fully tested and functional, even if most other stuff should theoretically work.
Yes, I ended up returning the laptop unfortunately. Power supplies past 100W are all iffy (even if there are a few good ones on the market today) and they don’t all work with all devices that technically support them. So really with the FW16… you are basically stuck with the power supply it came with, unless one of the newer models from Anker or GearIT or MINIX work in some capacity.
However there is no 240W power supplly I have seen on the market that does 240W in a single cable, only 140W and 180W outside of framework. So please keep this in mind because the laptop will power limit you as my posts mentioned and there is no way around this. It essentially gimps the laptops power down to almost nothing.
I ended up buying a laptop that would allow 100W draw from the USB-C with power controls (to an extent) and that worked better for me… I still dream for a future FW16 model with better compatibility but the USB-C PD world above 100W is really a dice roll still today.
I am using a delta electronics 240w power supply right now.
1 Like
I searched for that and didn’t find it until I typed in “Delta Electronics 240W USB-C” and I cannot find any link to buy it. Delta is an OEM for laptop PSU’s and the 2 articles I found which were talking about this PSU have it being the same size / bigger than the factory one that came with the FW16…
A few examples of where it may be acquired:
I grabbed it off a Reddit post, so I can’t confirm if they have stock or not. The model for the PSU is “ADP-240KB BA”
3 Likes
It’s possible to manually set settings for the PD controller to help bypass some of the issues. When I run “sudo ectool chargecontrol idle”, the laptop will stop trying to charge the battery and dedicate the charger+ battery to the workload. It prevents the stutter and frame time issues from the DGPU constantly changing power states.
1 Like
I requested and received the framework PTM and changed my FW16’s LM to the PTM7985. The process took 15 minutes and went just like the instructions. Most of the liquid metal was pumped out next to the CPU. It was relatively easy to clean, I used a hairdryer on hot to soften and clean both the CPU and heatsink with good effect.
I have a 240W delta charger that I bought from Digikey.
I then used my laptop in the usual fashion without overtly stressing it for a couple of days. Then I ran cinebench R23 and R24 several times. Before the TPM7985 I scored low 14000s in R23. Now I regularly get high 15000s!!! The top score that made me do a take a double take was 16099!!! This stuff is magic. With the LM the CPU package power never went above 45W. Now it regularly peaks at 67W with sustained 45-47W while doing the CB R23 multiprocessor bench! It used to get 100 degrees Centigrade for most of the cores with that well described large delta between the cores. Now they are at most 5-7 degrees between the lot. With the PTM the hottest it gets is 93!!!
I did not run prime95 before the swap but now with the PTM in place, i can run it with the second option (small FFT) to really get the burn going! The CPU never gets above 93-94! The fans goes hard at this stresstest but smoothly ramps up and down.
I know this does not add anything new to the discussion but just let people know how easy the process was and the benefits of doing it!!!
11 Likes
Running stress (through s-tui), my all-core, full-load performance is locked at 4.1 GHz and 50 W for at least 3 minutes continuously. I haven’t tested anything longer than that.
This is after replacing the LM TIM with the PTM TIM, as I was starting to see thermal issues. Before the upgrade, I would have been closer to 3.2 GHz on all cores. I would recommend doing the replacement. It’s not too difficult–just takes patience.
EDIT: In more practical usage (all-core compiling from the AUR), I see full usage with 4.6 GHz, 94 °C, and 59 W sustained
Folks, don’t try this at home, you’ll probably shorten component lifetimes…
I ran prime95 (small ffts for the burn!) as well a 2 instances of furmark and CB23 at the same time AND STILL GOT A SCORE OF 15500!!! Iknow that Furmark most likely has a small effect on the test but heck, THE TEMPS STABILIZED AT 95 degrees Centigrade!!! For 30 minutes, just to check it wasn’t fibbing! GPU 7700S and the 780m both sat at 76degreees.
The fan was going well and ramped down smoothly after I stop torturing my poor FW16.
FWFTW! Framework for the win!!!