High Thermals on a 7840HS

:Summery:
About a 23c difference in temp from hottest to coldest core when under heavy load. Working with support to troubleshoot.
:Summery:

When stress testing the CPU thermals with Prime95 I noticed I was getting some high numbers and weird fan behavior, based on sound of the fan as I don’t have monitoring software for fan speed installed.

One core gets hammered and jumps to 100c almost instantly and stays. The clock speeds on most cores hover anywhere from 3.9ghz to 4.1ghz. With the one core that’s at 100c dipping down to about 3.7ghz.

The temps when getting hit hard range from 100C on the hot core to 77C on the coldest core. I haven’t looked into the physical layout of the 7840HS but a almost 23C difference seems a bit high from my other systems I’ve run in the past.

Yes this is not a below base clock speed, however the CPU temp from core to core is almost 20C difference, possibly indicating that its not balancing the load over the CPU, or maybe a dry spot with the TIM?

In non-multi threaded things this seems to not be an issue as it boosts a core to the listed max.

In multi-threaded tasks however it started to throttle about 45min into a heavy task, and fails to recover.

I’m currently actively working with support on this to hopefully figure out why its behaving like this.

Config

Software
Clean Windows 11 Pro with latest updates. Microsoft telemetry disabled as much as possible, with a few extra services turned off.

Latest release of Prime95
Latest update of Steam running in background.
No other software installed, as I want to keep things to a minimum.

Hardware
CPU - 7840HS
GPU - 7700S
RAM - 16gbx2 5600mhz
NVME gen 4 2Tb

Drivers

Tried both OEM package from Framework and updated GPU/Chipset from AMD.
Expansion cards have been removed with power plugged into the system directly.

The thermals don’t change with cards installed or removed. Card in used

1 - USB C
2 - USB A
3 - USB A
4 - Ethernet/USB C when no internet is needed.
5 - USB C
6 - USB A

I have the laptop offline when I’ve been trying to troubleshoot this so I don’t get a random updates coming in messing with the testing method.

I will update when I can.

1 Like

That sounds like something wrong with either the thermal interface material or with the mounting pressure of the cooler. I only see an ~5°C difference between my hottest and coldest cores when running an all core load (haven’t tested Prime95 but I’ve hit it with plenty of CPU renders).

3 Likes

I would suggest the mounting pressure of the CPU heatsink is uneven. With all the time pressure to expedite the laptops in a short time, with all the delays already, the QC suffered. In my case I already had it apart a few times :joy: But I am comfortable to do these things, as I did it many times. Even my old Legion 5 (4600h and RTX2060) has liquid metal instead of thermal paste, an it did wonders with the cooling. The mounting points should be numbered. When tightening start from number one, when unmounting start from the highest number. Hope this helps. There are enough youtube videos for this process.

So after multiple emails with support they said it was normal expected operations for the system?

With the wide temp range from core to core, I am not convinced that that is normal or expected for a correctly mounted cooler. This is going off my past laptops/desktops.

I’m planning on pulling the laptop apart and making sure that the cooler is evenly torqued down. I can’t find a clear answer on spec so ill try and find spec for bare die or IHS and go that.

I will update one I get it all said and done. Fingers crossed that it’s just a QC oversight and that evens out the thermals.

I would definitely concur. A 23C difference sounds utterly ludicrous. 4-5C sure. But in the past a 10C difference would immediately suggest to me to yank the heatsink and repaste it, a very clear indicator of a bad paste job. I would push to escalate the issue with support, that just can’t be normal.

1 Like

Except that the FL16 uses liquid metal instead of paste. I would be very wary of dismantling that interface as it would be all too easy to mess it up and void your warranty.

Quite frankly a core to core temperature difference is going to be very dependant on exactly what software is running. It would be all too easy to run a program that doesn’t use more than a couple of threads and have the rest of the processor essentially idle resulting in exactly the scenario described.

1 Like

Generally speaking if you are using Prime95 chances are you are stress testing the rig, and that does not involve a heavy load on a single core.

Doubt it. You are not going to have a load that pounds one core to complete oblivion while the rest of them sit at 23C below that single core.

Lastly,

nice job of taking my comment completely out of context.

As far as I am aware that would not void the warranty. Any damage from improper removal of the old liquid metal pad/heat sink or poor/incorrect re-installation of a new one would be on me as it would be user caused damage.

The whole Framework thing is for sustainability and being able to repair the device you own.

OK, back to the topic at hand!

So from the " Framework Laptop 16 Deep Dive - Liquid Metal" email/post it links to the manufacturer of the sheet.

English instructions - Installation of liquid metal pad

It sounds like it would be fairly easy and quick to remove the old one and install a new one if I can get my hands on one as the liquid metal does not phase change until about 58c. So in theory it should be a solid at room temp and come off cleanly?

I have never used this type of thermal material so I’m going off what little info I found.

Again I will update when I hear back from support. I started this post in the hopes of documenting my issue so others might get something from it.

3 Likes

I wonder if framework pulled a similar blunder to xmg and only applied the lm to one side.

Maybe without the telling the manufacturer they know better how to use the stuff than the manufacturer bit XD.

1 Like

I just got an update from support.

“Thank you for your patience, this case was escalated for review with our Hardware and Engineering team who confirmed that the delta between core temperatures is not expected behavior. We would like to arrange for replacement hardware in this case so that this can be returned for further analysis…”

I’m glad that they support team is actively working with the early adapters of the FW16 and listing to issue and not just brushing it off as user error immediately.

I will update once I get the RMA hardware in hand.

4 Likes

There’s a deep dive on it : Framework | Framework Laptop 16 Deep Dive - Liquid Metal

If someone was to take off the cooler it would be very interesting to see pictures of the application.

Only applying it to one side has been a problem in a bunch of minipcs that used liquid metal too.

Given it comes as a module (for want of a better phrase) that should prevent issues relating to uneven application with Framework?

Even with a bad application would it even itself out with temperature cycles on a level surface?

That bit is what makes it more likley to be wonky, liquid metal has a crapton of surface tension so both sides need to be whetted for it to work right.

Or seep out the sides.

I really do wonder what it looks like now.

What does “both sides” mean here? The iFixit overview has video of the heatsink being removed from the CPU (I realize it’s a prerelease unit and one of the things they were still polishing at this point is the enclosure of the liquid metal but since I’m not sure what is meant by “both sides” I don’t know if there’s enough there to go on).

That’s why there’s a gasket around it…

That doesn’t make it cool the parts where it’s missing any better XD

Helps with not causing catastrophic damage though.

Update time!

They are sending me a new motherboard/main board to sway out and send the problem one back to them.

I am not going to remove the heat sink as they sound like they want to troubleshoot/root cause the issue.

After reading up on the material in use in the FW16, it seems it is not the same as other liquid metal solutions that stay liquid at room temperatures.

From the Framework post on it.

“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”

From the OEM instructions linked before. Maker CCHUAN/Coollaboratory

“If you want to remove the Liquid MetalPad, you can peel it off the contact area carefully. Depending on the structure of the surface it should peel off very easy and may leave small residues. You can remove these quickly and simply with a metal grinding pad. Coollaboratory sells this high-quality metal grinding pad in larger packs of MetalPad”

When the liquid metal pad/sheet is up on the marketplace I plan on buying a couple to play around with as well as to pull apart my FW16 to see what it looks like after some real world use.

Final update to this tread.

New board showed up in 3 days once support informed me it had shipped. Shipped from New Jersey to California. FedEx was the slow mover once it was in there hands.

Installing the new board was easy, followed the guide that Framework made just so I didn’t forget any steps/miss screws that needed to be loosened/tightened. It took me about 20 min to swap boards.

After booting back into my Win 11 install I ran Cinebench R23 multicore.

The delta from core to core is now less than 7c from hottest at 98c to lowest at 91c.

The improved thermals netted a jump from 14k to 15k, about a 7% jump. I ran it multiple times both with the old board and new and the results are solid.

So if you’re seeing a wide range on temps from core to core, reach out to support. I’m very happy with the help I ended up getting. It did end up taking about 3 weeks total to get a replacement in-hand, about 90% of the time was emailing support on it. Hopefully that time shrinks as they get through the pre-orders.

5 Likes

Well as it seems i also think i have a problematic ThermalSetup with my 7840hs Batch20 As i get a core Deviation of ~14C its not as abnormal as your 23C but i did write to the Support Team to see what they think. My CB23 Score is 14740 MC 1674SC i think thats still a good score. But this one Core being that far of tells my Gut its not right.