Uneven CPU thermals!

that sounds pretty concerning. my fans stay at 0 rpm with it open, and that process is showing 0% cpu according to the task manager. i’m not sure what’s going on there

If I run stress -c 16 and each core is at 100c with a package power draw of 37W, do I have a cooling issue?

Batch 1.

What do you use for reading the Temps? I Experienced, that Linux is not Reporting all Coretemps each Out of the Box. But Yes if you hit 100C at 37W Power Draw You are definitly experiencing a Problem.

I was using htop to read cpu temp

htop only shows the hottest Coretemp for all Cores, but it doesn’t alter that your CPU is thermalthrottling.

that’s good to know, thanks

How much power has the CPU use?

1 Like

Power use is within specifications. I recently changed to NixOS and configured the power plan suggested on the wiki for FW16, I think it does a better job at the balanced power configuration than the default under Fedora 40, I’m getting more performance on that plan with not much downside.

In the other news, over the weekend I received a new heatsink I had ordered, it basically has the same increased thermal resistance for core/sensor4. I also tried out liquid metal, it does perform slightly better than the 7950 pads right from the start. I would still advice to use pads over liquid metal.
The pad takes some time, a few days or so to get settled in, but after that, there’s not much difference between liquid metal or the pad in thermal performance.
So if anyone is considering either option, go with the pad, less hassle if you need to service your machine and no risks for leakage, spillage.

I now have a spare heatsink which I will start modding when I get some shims in to play with.
I’m waiting to decide which one I will sacrifice, waiting for the pad to settle in to see which heatsink was performing better if any difference.
Note, on the package it said it was produced m02-y2024, in case this might get relevant.

1 Like

So I got the Grizzly PTM sheet in today.

Went ahead and applied it to my GPU for testing (much easier to mess around with). I will do some benchmarks over the next week or see to see how it runs.

Once complete, and pending some long-term results of using PTM instead of LM - Ill likely make the plunge to replace the LM with the PTM.

3 Likes

I had good results with replacing the paste with the thermal grizzly pad. I saw a significant fps increase in heaven benchmark.

1 Like

Edit: So I checked my previous post in this thread and it turns out I am a total drama queen and have no idea what I’m talking about. Performance actually looks better than it did 21 days ago, although the cores are somewhat more busy. The thing about performance doing worse than balanced for some reason still applies, as does the question of whether or not it is a heatsink issue or a mainboard issue.

I leave my original post below, as a form of entertainment. Read and LOL as some form of entertainment.

=========================

The performance degradation is real, sadly, at least for Batch 15 units. A system which used to run at 4.4ghz now only runs at 4.2ghz at best, and at worst now runs at 3.9ghz. On average it runs at about 4.1ghz.

Core    CPU     Avg_MHz Busy%   Bzy_MHz TSC_MHz IPC     IRQ     POLL    C1      C2      C3      POLL%   C1%     C2%     C3%     CorWatt PkgWatt
-       -       4130    98.33   4200    3793    1.58    27857   5       76      227     897     0.00    0.03    0.12    1.53    36.82   44.00
0       0       4129    98.27   4201    3793    1.64    1632    0       10      11      50      0.00    0.15    0.14    1.44    4.55    43.99
0       1       4123    98.20   4199    3793    1.63    1634    0       5       36      63      0.00    0.03    0.40    1.38
1       2       4137    98.51   4199    3793    1.55    1617    0       0       5       62      0.00    0.00    0.04    1.47    4.43
1       3       4113    97.96   4199    3793    1.62    1656    3       13      34      70      0.00    0.07    0.23    1.76
2       4       4146    98.71   4200    3793    1.62    1611    0       10      1       54      0.00    0.08    0.00    1.21    4.57
2       5       4094    97.53   4197    3793    1.52    1619    2       6       22      81      0.00    0.02    0.15    2.31
3       6       4103    97.73   4199    3793    1.46    1603    0       0       5       71      0.00    0.00    0.03    2.26    4.52
3       7       4145    98.66   4201    3793    1.60    1623    0       4       7       49      0.00    0.05    0.06    1.24
4       8       4133    98.41   4200    3793    1.56    1585    0       1       9       28      0.00    0.00    0.11    1.49    4.69
4       9       4155    98.89   4201    3793    1.61    3273    0       2       25      50      0.00    0.00    0.19    0.93
5       10      4107    97.84   4198    3793    1.62    1619    0       1       10      73      0.00    0.00    0.09    2.09    4.72
5       11      4146    98.69   4201    3793    1.53    1869    0       5       19      59      0.00    0.06    0.12    1.14
6       12      4129    98.35   4198    3793    1.63    1598    0       6       5       58      0.00    0.03    0.02    1.61    4.78
6       13      4158    98.95   4202    3793    1.63    1663    0       10      24      27      0.00    0.02    0.24    0.79
7       14      4129    98.30   4200    3793    1.65    1619    0       3       2       39      0.00    0.02    0.02    1.66    4.55
7       15      4127    98.24   4200    3793    1.43    1636    0       0       12      63      0.00    0.00    0.08    1.69

That being said, Linux benchmarks with 7z should probably be taken with a grain of salt, since it seems to perform better in balanced mode compared to performance mode, there I was seeing 3.9ghz regularly.

Is this a mainboard issue or a heatsink issue?

=====================
Edit 2 for ease of people scrolling through thread:

Again. Drama queen.

4 Likes

Hey all, on my second motherboard so far and it was fine for a day or so and now it performs quite poorly ~30w sustained max multiple cores at or close to 100c 12k-13k ish cinebench scores, I had a peak of 15.5k when I first got the replacement board, then my new board started behaving even worse than my original after a few thermal cycles.

Second Board (~30w): https://www.youtube.com/watch?v=4_zKF2CkCOs
First Board (~40w): https://www.youtube.com/watch?v=AIBizvVRaac

As you can see I’ve had pretty terrible luck, not even sure what to do at this point.

Either open a new support ticket or you could start tinkering first, remove the LM and replace it with a pad or thermal paste, see if it improves.

I speculate that the liquid metal has started seeping away in the recesses around the die.

Run the cpu hot in a benchmark, flip your laptop upside down and shut off, and let it cool down.
It should be easier to remove the liquid metal that way, it will stick to the heat-sink more and less to the die.
Then the left over flakes you can scrap off with the framework tool backside.
Put in your own thermal solution as replacement for the LM.

Hope it helps.

2 Likes

My moddiy genuine PTM 7950 arrived today, and I had promptly installed it once I got home from work.

Here’s the application, which admittedly, wasn’t the best but I was too excited to try it and get it tested:

I applied it without using any guides online. I peeled one side of the plastic off initially, then laid the other plastic side (with the PTM7950 attached) down onto the CPU die to measure it up, left a few marks on where I needed to cut, and then cut accordingly.
Once cut to size, I took the remaining plastic off and used the spudger side of the Framework screwdriver to move it into a slightly better place once I had laid it down using my fingers.

Alright with that waffling over, time for the results! Aaaaaand… it sucked? 45w (stable) by the half way mark of a 10 minute run, and the score I got was roughly 14800 - just a little higher than my original scores pre-RMA and pre-repaste. For comparisons sake, Cinebench R23 last gave me a score of 15330 after a day of work, no cooldown period, using the Gelid GC-Extreme I had previously applied.

Now, I knew that people had uttered the so-called ‘break-in period’ but I was doubtful it was going to make a difference. I promptly shut the laptop off for 20 minutes and had a shower, only switching it on and immediately running another 10 minute run of R23 once I had dried myself. The score now? 15767! And by the end of the 10 minute sprint it was at 50w with occasional bounces to 53w.

Of course, CPU temps were still at the 100c mark but that’s of no concern to me. Core 0 had a delta of about 11c, with Core 5 being my hottest, with Core 3 and Core 4 being runners up.

I also just wanted to say that I’ve been checking this thread religiously every day, and have enjoyed reading all of your guys’ experiences and testing results.

9 Likes

I made a mask of paper for both the dGPU and CPU so I can easily cut the size out of the pad sheet.

Grab a piece of paper and a pencil and go over the cpu die sides like a detective outlining the die.
Cut that out, now you have a perfect reference for size for whenever you need to reapply a new piece.

4 Likes

I ended up repasting with regular paste, Antec FORMULA 7 Nano Diamond, because I was getting as low as 25watts.

New Results for 10 minute cinebench, its likely slightly lower than normal because I have 4800mhz ram instead of 5600mhz.
image

Here are some pictures from the install

5 Likes

There are just a few steps if you want to reapply it.

  • Make a template using some paper ( make sure that the template is slightly bigger )
  • After cutting, put that piece in the freezer for 4-5 minutes
  • Remove the first layer ( plastic )
  • Apply it
  • Use the spudger to press on the sides
  • Remove the plastic cover with everything that’s extra

Some photos with my application.

Step 1, pressing on the sides

Step 2, remove final layer

Thanks for sharing the score! This is quite close to my fake PTM.
In my case, I have the same hot core, core 5 and that’s strange as before it was 4.

I just did a new test, and I have 16288pts on a 10-minute test from cold (20C room temperature).

New Test

2 Likes

Hello, I need help, what Can I do with this:


should I remove the shield so I can remove the remaining LM or I leave like that and install the pads?

1 Like

Framework wants me to ship the entire laptop back for the repair. Being a dysgraphic college student that’s not going to happen until winter break.

Thinking of installing a KryoSheet if perf craps out AGAIN after this. I keep hearing bad stuff, though - is it really just PTM7950 or bust as far as most can tell? I can’t find where to get the real stuff, maybe moddiy?

3 Likes

Moddiy for the actual PTM 7950

Im using Thermal Grizzly Phasesheet PTM - which is basically the same thing.

1 Like