Uneven CPU thermals!

This thread has gotten to over 400 pages long. To help new-comers catch-up quick, I generated an AI summary of the thread and manually edited or removed sections that I believed to be hallucinatory.


Summary Report: Uneven CPU Temperatures in Framework Laptop 16 (AMD 7040 Series)

1. Background and Scope

This document is a detailed, community-driven forum thread chronicling user experiences, diagnostics, and technical investigations into uneven CPU core temperatures and thermal throttling on the Framework Laptop 16 (AMD Ryzen 7040 series, especially 7840HS/7940HS). The thread spans from initial user reports in August 2024 through early 2025, capturing evolving community knowledge, troubleshooting, and Framework’s official responses.


2. Core Problem: Uneven CPU Thermals

  • Symptom: Users observed large temperature deltas (10–40°C) between CPU cores under load, with one or more cores rapidly hitting 100°C and others much cooler. This leads to thermal throttling and reduced sustained performance (often 30–40W CPU package power, sometimes lower).

  • Benchmark Impact: Cinebench R23 multicore scores for affected systems dropped from expected 15–16k+ (for 7940HS) to as low as 11–13k, with power draw and clocks well below spec.

  • Fan Noise: Many users reported excessive fan noise even at idle or light loads, attributed to poor heat transfer and rapid thermal ramping.


3. Community Diagnostics and Experiments

A. User Testing and Data Collection

  • Benchmarking: Users ran Cinebench R23, 3DMark, and stress tests, sharing detailed results (scores, power, per-core temps).

  • Monitoring Tools: HWInfo (Windows), amdgpu_top, s-tui, and lm-sensors (Linux) were used to track per-core temps and power.

  • Comparative Analysis: Users compared Framework 16 results to other laptops with the same CPUs (e.g., Razer Blade 14), finding Framework’s cooling underperformed.

B. Hardware Interventions

  • Thermal Interface Material (TIM) Swaps: Many replaced the factory-applied liquid metal (LM) with:

    • PTM7950/7958 phase change pads (Honeywell or Thermal Grizzly PhaseSheet)

    • Conventional thermal pastes (Noctua NT-H1/H2, Arctic MX-6, Gelid GC-Extreme, etc.)

    • Copper shim “sandwich” mods (placing a lapped copper shim between die and vapor chamber, with PTM on both sides)

  • Heatsink Modifications: Some users lapped (polished) the heatsink or even desoldered and replaced the factory shim for better contact.

  • Fan and Chassis Mods: Blocking dGPU vent holes with tape or vinyl to improve CPU airflow and reduce noise.

C. Results of Interventions

  • PTM7950/PTM7958: Widely reported to restore or exceed original performance, with sustained 50–57W CPU power and Cinebench scores back to 15–16k+. Core-to-core temperature deltas dropped to 5–10°C.

  • Copper Shim Sandwich: The most effective (but risky) mod, enabling sustained 60W+ and even higher scores, but requires advanced skills.

  • Conventional Paste: Improved over degraded LM, but not as effective as PTM or copper shim mods.

  • Heatsink Quality: Users found significant variance (“heatsink lottery”)—some heatsinks performed much better than others, even after mods.


4. Root Causes Identified

A. Factory Liquid Metal Issues

  • LM “Pump-Out” and Runoff: The indium-tin-bismuth LM pad used by Framework can migrate away from the die over time, especially after many thermal cycles or if the laptop is stored vertically. This leads to poor contact and rapid performance degradation.

  • Insufficient LM Application: Some units had too little LM, causing immediate poor thermal transfer.

  • Removal Difficulty: The LM pad solidifies at room temperature, making removal challenging but less risky than conventional liquid metals.


5. Framework’s Official Response and Actions

  • Acknowledgement: Framework acknowledged the issue in late 2024, citing both LM pump-out and heatsink bonding as root causes.

  • Production Change: As of November 2024, Framework switched new production to Honeywell PTM7958 (a phase change paste/pad) instead of LM.

  • Remediation for Existing Customers:

    • PTM7950/PTM7958 Kits: Framework is offering free PTM kits to affected customers (request via support form), with a detailed installation guide forthcoming.

    • No Recall: No full recall; RMAs are still processed, but replacement boards may still have LM until old stock is depleted.


6. Community Recommendations and Best Practices

A. Diagnosing the Issue

  • Run Cinebench R23 multicore in “High Performance” mode, plugged into a 180W+ charger.

    • Expected: 15k+ score, 50–54W sustained, core temps within 5–10°C.

    • Affected: <14k score, <45W sustained, one or more cores at 100°C, others much cooler (delta >15°C).

B. Fixes and Workarounds

  • If under warranty: Request a PTM kit from Framework support.

  • DIY (out of warranty or advanced users):

    • Replace LM with PTM7950/PTM7958 or Thermal Grizzly PhaseSheet.

    • For best results, add a lapped copper shim (20x20mm, 0.5–0.8mm thick) between die and vapor chamber, with PTM on both sides (“PTM sandwich”).

    • Lapping the heatsink and ensuring even pressure can further improve results.

  • Fan Noise: Blocking inner dGPU vents can reduce high-frequency fan noise and improve CPU cooling.

C. Risks

  • Liquid Metal Removal: Framework’s LM pad is solid at room temp, but care is still needed to avoid board damage.

  • Copper Shim Mod: Advanced mod, risk of damaging heatsink or board if not done carefully.


7. Broader Lessons and Reflections

  • Design Flaw: The combination of LM pump-out and poor heatsink/shim bonding is a fundamental design flaw, not just a manufacturing defect.

  • Community Power: The Framework community’s technical depth, experimentation, and documentation were crucial in diagnosing and publicizing the issue.

  • Framework’s Response: While Framework is now addressing the issue, some users are dissatisfied with the pace and scope of the fix, especially those who went through multiple RMAs.

  • Modularity and Repairability: The situation highlights both the strengths (user repairability, transparency) and challenges (early adopter risk, DIY burden) of modular laptops.


8. Key Takeaways

  • If you own a Framework Laptop 16 (AMD 7040 series) and see poor CPU performance or uneven core temps, you are likely affected.

  • Switching to PTM7950/PTM7958 (or Thermal Grizzly PhaseSheet) restores most or all lost performance and is now the official fix.

  • Advanced users can further improve results with a copper shim “sandwich” mod, but this is not required for most.

  • Framework is providing PTM kits.


References and Further Reading

7 Likes

Not that I don’t salute the effort but did you know there is a summary button? Well, something that only shows top replies anyways. Not the same thing as a summary I’ll grant you

2 Likes

I did not! I don’t see a button labeled summary. Are you talking about this one?

It never occurred to me to use it for a summary, I never understood that button’s purpose.

1 Like

That’s the one! And yeah, it’s not really a summary, I misspoke. But it does highlight the top replies and gives a good sense of what’s happening in a thread and if someone wants more context they can drop out of the top replies and look at all of them

2 Likes

I suspect my thermal solution may have degraded despite having switched to the PTM, as my cpu is hitting 100c again while the coldest core may be 10-15c lower. It’s particularly noticable when playing Battlefield 6. Has anyone else noticed this? I used to be able to run 55W through my CPU, but now it caps at 45W.

I’ve seen this on BIOS 3.05, 3.07, and 4.01.

PTM needs some heat cycles to perform properly. How long ago did you apply PTM?

Also, looks like poor application. ∆15° is way too much. Try reapplying and give it some Cinebench/ turnoff cycles.

Quite a while ago, not sure how many months. I should probably rerun cinebench to see if that’s dropped as well or if this is just normal for a different kind of work load.

Even when I switched my 7940HX to PTM (not Framework-given, had some leftover thermal grizzly branded PTM from another laptop I repasted) it hits 100 on Cinebench and when compiling code. When looking at the temps while gaming though, it’s usually hovering somewhere in the 80s (specifically have been playing Shadow of War and Black Mesa, used the Steam performance overlay to keep an eye on temps and framerate). I’ve wanted to undervolt the CPU to see if I could bring down the temps and maintain high performance but my CPU gets very unstable when undervolted. Bad luck with the silicon lottery, I think, or I didn’t do a good job replacing the shim on the heatsink.

1 Like

Same here with me! As I don’t game nor regularly do other intensive tasks and, my FW16 still runs fast enough for me. I’ve decided to accept how it behaves. I’ve had enough of chasing the event of 10 to 15 degrees difference between a couple of cores.

Cheers

that support form link always redirects me a language localized sub-url to that original and never the less even tried with english lang settings or even the TOR browser, it always ends up as the topic of the page shows “unauthorized” but otherwise completely blank empty white page nothing gets rendered in the browser. what gives? what is the original entry page or place to request for the PTM phase change thermal pad etc? thanks.

It gives me that as well, I wonder if they took the request form down.

1 Like

What link are you using to contact support with? Happy to check it out.

I believe they’re references the PTM Request for cited above. The end of that URL looks odd admittedly. This is a copy paste from this comment

https://knowledgebase.frame.work/phase-change-thermal-pad-kit-request-form-BkrOJPCfye

edit: oh, that’s my comment. Let me see if I got the link wrong.

Oh yeah, that form has closed. You just need to open a support ticket request. Here you go: https://frame.work/support#contact

@abittner

3 Likes

I just reran my cinebench test and got a score of 13479, which is quite low for a R7 7840HS w/ the official PTM. Originally it was trying to boost to ~58W, but that sent it to 100c and it eventually settled at pulling 35W ± 0.01W. The last score in my log was (nearly) 15k. I’m debating whether I should submit a support ticket for this or not.

I’m currently on BIOS 4.01 beta, but I had also noticed this on 3.05.

I have a Ryzen9 7940HS and my score 13k → 15880 with PTM. I will run it again once 4.x BIOS is officially released. I have a 9100 Pro 4TB (running at PCIe 4.0 speed) but the cache has doubled compared to previous 2TB…curious to find out what new score would look like!

1 Like

Just incase I was unclear, what I’m highlighting here is that my score went from 15k → 13.5k both scored with PTM. If I’m lucky it’s just a BIOS issue.

I will try to test tomorrow with 4.01 Beta and see if my scores go down as well…

Sadly, probably not a bios issue.

Im on 4.01 and get mid 15s still

Thanks for the double check. I’ve decided to put in a support ticket, as I’ve also been having other issues related to power delivery.