CPU running at hot temps?

Got it. Those core temperatures are slightly higher than we would expect for that ambient temperature (~50-60C above ambient is what we’d expect).

So, installing thermald brings the temps down to about 80C (from ~90C), power down to 22.3W (from 28W), and the clock down to 3GHz (from 3.2-3.3GHz) sustained.

However, I’m a bit confused (mostly because I know nothing about CPUs). 3GHz is the advertised base frequency at 28W (for the 1185G7). But I’m getting 3GHz at 22.3W and seem to be able to sustain 3.3GHz at 28W. Is this expected (is my data bad?)?

@Stebalien That monitoring tool is so cool I had to try it!
I have the 4 core i5-1135G7 DIY version with 2x16G RAM. Running Kubuntu+RegolithWM. I haven’t seen any overheating issues so far. Here’s a screenshot of my stress test result with 8 worker threads running for over ten minutes. Quite stable ~60C. I had tlp enabled and AC power plugged in.

@Stebalien I think the base frequencies are intended by Intel to be minimum guarantees, but you can sustain all-core frequencies above that.

2 Likes

I’ve continued to wonder about the thermals of my system. The fan was on & temperatures were high more often than it seemed that they should be.

Today I took advantage of the ease of disassembly in my Framework Laptop to inspect the thermal interface material on my heatsink. I did one ThrottleStop 4thread TSBench run & got my typical result of ~29.3sec ending at 100C before I disassembled the system.

On inspection, there was a void in the thermal interface affecting aprox 35% of the CPU’s surface area. (I have a photo documenting the void.)

I cleaned the CPU & chipset die & package substrate, cleaned the heatsink, & applied fresh Noctua NT-H1 thermal paste before carefully reassembling the system. (Isn’t it great that we can do this! The total time was <<20min.)

Repeating the qualitative benchmark with ThrottleStop, the results are clear. The fans run slower & less often. The temperatures are cooler & the benchmark results are improved. (I have the before & after ThrottleStop log file & charts showing the monitored results.) In particular, temperatures rise more slowly & peak lower under the same test after the change.

If you’re experiencing high temperatures or excessive fan noise, first use the Custom Default BIOS settings & if that doesn’t fully resolve it replace the thermal paste in your system.

9 Likes

That sounds like the thermal paste pump-out issue that we have found on some early units before we switched to a different thermal paste formulation. Anyone who is seeing this on a Batch 1 unit can reach out to our support team to request a free thermal paste tube of the new formulation.

14 Likes

I’m thinking this is also a great example of the success of the DIY program. For myself, I would much rather have some thermal paste shipped to me and let me re-install myself, versus having to ship the entire machine back to the manufacturer. Lenovo once tried to get me to do an RMA on a ThinkPad for a keyboard driver issue - I of course refused & just waited for the updated driver release. :confused: Has anyone ever heard of Dell, HP, etc. shipping someone a tube of thermal paste to fix their issue? It would never happen. But for lots of us DIY’ers, that would be the preferred solution, so I can make sure it’s done right. And how much easier & cheaper is that for Framework? Genius!

5 Likes

To be fair I get why Dell, HP, etc wouldn’t do that. Would you trust your dad to open up a Dell and change the paste out? I’d bet money that my dad would break something and void his warranty.

This is kind of a different market so framework can do things like that. I wouldn’t claim it’s the fault of dell, it’s just a case of working with your targeted market.

I’m not at all suggesting that Framework treat all of their customers that way, but rather the DIY’ers that are willing to do such things. For anyone that can’t handle something like that, they should absolutely return the product to Framework and let the experts handle it. I see that as the potential differentiator between the “standard” edition Framework Laptop that comes pre-assembled (with full service warranty & support just like Dell, HP, etc.) versus the Framework DIY Edition (with more of a “send the parts, let the customer swap them out” mentality).

1 Like

That’s fair, I’m just defending the big box’s position on that one is all. It makes sense at the scale they operate in.

Personally I’d just buy a tube of whatever I want anyways and put it on. I don’t need bulk stuff from FW.

I didn’t check for pump-out on my unit as I’m traveling (and can’t repaste), but I did notice pretty high temps (batch 1) with not that high of CPU usage (bullet physics simulator with a static opengl scene).


Note that 100C for a few seconds is expected when boosting to PL2 power levels. This happens automatically for short periods of high load. If the package temperatures at sitting at 100C during sustained load for more than a few seconds, that likely indicates thermal paste pump-out.

1 Like

Yes, I think it settles around low to mid 90s. For those who want to test:

  1. Install psensor/stress (I use nix, nix-shell -p pensors stress tmux)
  2. Start sensors sensors --cpu 8 and psensor
  3. Watch the graph.

2 Likes

Could you reach out to our support team to request a tube of thermal paste?

Just read this thread – I’m Batch 1 so I checked to see if I had this issue, though I didn’t think so cause fans/temps didn’t seem out of the ordinary. After checking again, I still don’t think so. Though maybe with some fancy thermal paste I could eke out more performance/cooling, I’m quite pleased with stock!

tldr my i7-1165G7 (Fedora34):

  • Idles around 38-40C
  • 2.8GHz sustained: ~67C (thermald default mode) [1]
  • 3.5GHz sustained: ~90C (thermald performance mode) [2]
  • Managed to achieve even higher 3.6GHz-3.7GHz sustained oddly in [3]

Falls in line with NotebookCheck’s i7-1165G7 review (under Stress Test section):

  • 3.0GHz sustained (Windows, Prime95): 82C

@ me if you Windows numbers, else I’m too lazy to reboot for now :sweat_smile:

@Michael_Lingelbach or anyone that needs a data point for the i7-1165G7:
Sustaining 2.8GHz on all cores, mine settles at ~67C with stress --cpu 8 or s-tui’s stress test. [1]

@Stebalien I read your comment and decided to see if I could get my i7-1165G7 to sustain above its stock 2.8GHz base frequency, since it seemed to sustain 3.0GHz in NotebookCheck’s review (in Prime95 on Windows).

I then discovered that with thermald-set-pref, I could set thermald to PERFORMANCE mode instead of DEFAULT. On PERFORMANCE mode, it seems to sustain 3.5GHz (made sure by letting it run for over a minute), settling at ~90C. Sweeet! [2]

Interestingly, there also seems to be an ENERGY_CONSERVE mode.
Edit: There are also other settings such as CPU MAX temp, CPU PASSIVE temp, and CALIBRATE in thermald. Haven’t dove too deep into it, but:
I tried thermald-set-pref → 5 SET USER DEFINED CPU MAX temp80, oddly it seems like it’s performing better than the PERFORMANCE mode. It sustainins ~3.6GHz (bounces between 3.6 to 3.8GHz) :thinking: [3]

On Windows, I could probably use ThrottleStop to emulate thermald preferences.

[1]

[2]

[3]

2 Likes

We’re using Shin-Etsu X23-8117 now.

2 Likes

I haven’t noticed (but admittedly haven’t attempted stressing the CPU for sustained periods just yet) any cpu temp issues on my batch 1. However, is the nature of the pump out such that it is only a matter of time before we have an issue? IE: is it something you would suggest we do now?

If I bought something like Artic Silver, would that suffice? Would that be an improvement?

Pump-out should only happen after hundreds or thousands of temperature cycles, but we have seen instances of it occurring in what seems like less than that.

1 Like

I ran MemTest86 on mine. For some parts of the workload, the CPU temp hung in the 80s C. (I think the fans kicked on in the 60s or 70s.) The peak temperature I observed was 90 C (pictured below).

The 90 C peak is within spec and expectations, right?

Is there any available documentation that gives details on the thermal specs or expected behavior?

20210923_165803_2

I believe that it’s rated for 100C, but someone will correct me if I’m wrong.