What are you using to pull the info in the screen shots above?
Its turbostat.
1 Like
I think your CPU was somewhat screwed from the start, I have stats from my two month old laptop stressed using 7zip’s benchmark functionality and it looks similar to your system when it was healthy while using far less power on the balanced setting.
Or I could be misunderstanding something. That would be par for the course.
Thanks for your share. I suspected that all FW16 share this fate. Thats why i started modifying my Heatsink. You could test to loosen your 6 Mainboard Screws and just slightly torque them again. This can possibly help. Thats one Idea a User shared higher up in this Tbread Here yesterday.
The System should stay at about 50-53w on a Multicore CB23 Run. If its not its hard Throttling. And the Number 1 Reason on this System is the Heattransfer to the Heatsink.
1 Like
finally got around to getting cinebench r23 going, ended up with 11.9k to 12.5k for multi-core with the power profile set to performance, with turbostat results typically looking like…
of course, being on ubuntu 24.04 lts, there’s some awkward with my current kernel (i don’t have properly functioning temp sensors, for instance), but this does look fairly below what some other folks have achieved, so maybe i’ll fiddle with screws later.
I did a little bit more experimenting.
When I screwed down the screws again, I didn’t hit the 16 anymore.
This led me to believe something is going on with the pressure and or angles of the heatpipe and mainboard.
I tested with a pad I had laying around ~25mm*30mm that I got from an nvme m.2 one of those grey rubberish slabs.
I put it on top of the heatpipe and the top plate.
still sub 16k nothing else noticeable.
Then I tried putting the slab between the bottom and the mainboard.
I didn’t hit the 16k but I did see another anomaly, the thermals for sensor 1-3 had a 3-5C lower delta. sensor 4 still hit 100c.
The slab was too small to actually have good contact or all round pressure.
I am going to order some more of these slabs and going to try different shapes and locations to see if I can find the sweet spot for improvement of the delta and all round temps.
it is a different route @PSierra117 took lapping the heatpipe and PTM7950.
My hunch is that the cpu is not getting the optimal mounting pressure for the heatpipe and ihs to get all the cores to have good enough indirect contact with the heatpipe.
1 Like
Thank you for your Tests and further Tinkering. Thats the Value of a Community willing to experiment 
We should inform the Framework Support of this Findings. Maybe @Destroya can Help?
I will buy a second Heatsink and 2 PTM LM Pad from Framework in mid to late October. They wrote to me in my Support Request, that the Warehouse is stocked with the LM Pad, but they will realease it for Sale when the Replacement Guide goes online and this is presumably in Early October.
When i have a second Heatsink i will try to modify my Current one further and than Test it against a "Stock Setup and i will experiment with different Mounting Pressures and maybe Shims underneath some of the Mainboard Fasteners.
1 Like
I have ordered some of those thermal pads, hopefully they will get delivered tomorrow so I can proceed testing.
Depending on the results, I might end up doing what you did @PSierra117 I don’t mind tinkering, I have been lapping CPU’s since the Celeron 300 days.
lapping bare dies is a lot riskier these days, also not really worth it at these power levels.
I know, if you read back a little, you see in FW16 case, it is the heatsink/heatpipe that’s being flattened.
I’m just saying I’m no stranger to the process.
I realize we are talking about AMD procs here. But on the newer Intel chips with Performance and Efficientcy cores, would you ‘expect’ there to be a thermal/speed difference between the cores running the same tests?
@taezea I wonder if this is because of the thermal transfer between the main board and the chassis. By adding the thermal pad the chassis acts like a big heatsink.
There are a couple of things that we should consider when doing this, but, I may be wrong.
- We should make sure that we won’t add extra heat to different components under the mainboard. Mainly because we’re moving heat away from the CPU.
- We need to make sure that the thermal pad isn’t too thick as we could add some extra stress on the mainboard and it could be a risk of damaging it. Also, I think that the thermal pad should be a bit wider in order to avoid adding a stress point.
For the second point, I’d say that we may need to calculate the thermal pad thickness by checking the distance between the chassis and mainboard and getting one that’s thicker by 0.10/0.15mm.
Re:
Not enough pressure or, not enough liquid metal.
I wish someone from the framework team could confirm that if we have the right thermal pad we can in fact use the chassis to improve the thermal transfer from the mainboard.
1 Like
When you test the new thermal pads, can you please take some photos with the mainboard?
I want to see what we have under the CPU.
1 Like
I totally agree with you points, I cannot confirm or deny anything yet as I’m still in the collecting data and testing phase for any conclusions.
I doubt much heat transfer would take place from the small piece I used. I will do more extensive testing when I get the larger pads in and feel/touch with and without modifications.
At the moment there’s basically a very good insulator between the mainboard and the underside of the body. And that is the air pocket between it. (in the original configuration)
The main purpose for the padding would be to counter the top pressure from the top plate and keyboard and force a squeeze type of force on the mainboard and heatsink/heatpipe.
The type of padding I ordered is designed to squeeze in ~25% of it’s original thickness. It is very soft and designed for electrical components, I’m not to worried about the stress involved as it’s flat surface stacked on flat surface on flat surface with space to squeeze out of sides of the pressure, as long it’s not too thick it’s shouldn’t be more than the same type of stress of you pushing on the keyboard pushing towards the top plate, it’s probably even less.
I think pictures here would explain more than anything I type so that’s next.
When remove the mainboard again to measure and install the padding I will shoot some pics. @Alex_Uta
I cannot comment yet on if there’s enough liquid metal used or not, I wont open and remove the heatsink/heatpipe until I also have a replacement ready to go. I can tinker for a few days but I need a working laptop after that.
The mounting pressure is just the working theory that I’m going to work on to prove or disprove.
3 Likes
I’ll be watching this thread carefully for the foreseeable future. I already had my board replaced and, despite CBR23 scores initially being within the target range, they had already dropped over 10% within a month. I shared the numbers and such right here:
I spoke to Framework Support again and now they’re asking me to send the entire laptop in, which is honestly irritating. I’m fortunate enough that not having the laptop on hand for some time won’t be affecting my college work or anything like that, but not every one can afford to have their laptop out of commission for who knows how long. I’d much rather keep having my Mainboard replaced every once in a while until the problem is finally addressed rather than having to send in the entire unit with the caveat that the problem might still not be fixed after that.
1 Like
What would be the reason to take the whole laptop in, would they be checking if the body or top plate is warped or something like that? Something wrong with tolerances of the screw towers?
That’s really wild.
1 Like
I’m honestly not sure. Frankly, I don’t see how it could be an issue with any of the other components when the board worked fine upon arrival. So unless there’s something that’s slowly warping the mainboard or the heatsink assembly over time, the only potential culprits are the board itself or the heatsink.
I’m pretty damn sure the deck isn’t warped. However, I will say that I have observed that the body has a slight bend depth-wise. That may have something to do with it.
I installed some padding. I used TP-3 Arctic 1.5mm height.
I used a piece of paper and a pencil to make a mask to use for the padding and cut it to size.
Then I made a opening for the stud so the padding doesn’t stack with the stud putting uneven stress on the CPU area. Note the dot near the middle.
The mainboard has black studs supporting it. So it is screwed in with the 6 screws but also supported by the black studs all over the bottom of the mainboard.
This means that normally there’s a area that only has air which is a good insulator if it’s not moving. It must be around 1.5mm.
I don’t have calipers at hand to measure the studs, so if anyone else has a chance to measure them that would be great info. (The height)
The 1.5mm padding doesn’t seem to interfere with the installation of the mainboard back in it’s normal placement. The material property allows for it to squeeze in by design,
I will make some runs with CB R23.
3 Likes
i am really excited for your Results. After i loosened all 6 of my Mainboard Screws as you did. and just gently tightening them to when they just start to stop. I got reduced my “Spike” during Boot and its staying silent when in Desktop. And i am hitting more stable High 15ks in CBR23. Funny Think the “Temp” Delta increased again, but it didn’t hurt Performance.
I Get 63W pulled from the CPU in the Start of a BenchRun falling to 51W over the 10min Cycle.
2 Likes
I tried that screw loosening trick, but it didn’t seem to have any effect. Needless to say, seeing 37W max, 34W sustained is very concerning.
Edit: I also tried tweaking the tension on the heatsink retention screws, and now the core-to-core delta is more than 20°C (with Core1 now being the coolest core at 79.8°C max and Core4 hitting 100°C)
1 Like