I’m opening this thread (quickly) as a placeholder as I couldn’t find another suitable, and furthering some discussion from the PTM7950 thread.
To start off, this is for reports of differences between different thermal materials (stock, PTM7950, liquid metal, etc.).
I eventually plan on putting together a cross-platform script that can automate testing and recording temperatures, battery life, and performance with the goal of keeping it as simple as possible, yet standardized enough that the community can compare to see the impact of various configurations as well as a quick device health check. I’m daily driving Fedora on the AMD 7840U, so it’ll start with that (though I still have my old Intel i7-1165G7). Not sure when (quite busy ATM), and if anyone else wants to tackle this in the meantime, please do and I’ll help out!
Mods please turn this into a wiki if possible, TIA. Or delete if this is too much noise or merge if there’s another thread or whatever
This is now a wiki, I’ll be editing/updating this original post and feel free to add whatever’s useful!
Ooo, intriguing. Any insight/updates on this/the negatives of increasing the power limit? /is there another thread you’re posting insights in? Also wondering if disabling STAPM might resolve some of that weird behavior people have reported though I have a hunch that the limits might be tuned for the stock paste (could be completely wrong, untested). Anyone know if the weirdness also happens with liquid metal?
So far have not noticed any drawbacks, idle power is within margin of error and any differences are more likely to be software updates.
What werid behavior?
The stapm isn’t a problem with the stock paste since the default paste can’t do more than 30W sustained anyway. PTM may have quite a bit of headroom above 30 but I switched to lm before figuring out how to bypass stapm so can’t really test that.
Nice, could you post/point to how to increase the power limits? No rush or anything, I haven’t looked into it/just for the future
These:
click to open
kind of circled back from this (click to open)
Though I just reread from the above quote that the behavior does also happen with liquid metal. So I’m wondering if disabling STAPM disables/messes with the “sensor that no longer heats up as quickly as it does when stock”, and/or there’s somewhere where the threshold can be changed. (also TBD on my end)
I forgot to bring my smokeless stick so I can’t really check, I know it was a combination of settings that did it. May be a good idea to confirm that they don’t handle vrm thermal protection with that (I’d hope they use prochot but you never know and I don’t got a schematic) before I tell people how to do it. I’ll do it eventually though, hit me up next week if I forget.
Doubt that has anything to do with stapm since the ec sets the power limit to 30 on ac anyway. the stapm limit only comes into play when you mess with ryzenad to temporarily raise your power limits above 30 and have temps above 80 otherwise it doesn’t do much. Part of those is probably messed up application of the ptm, if you get a hole in the pad while applying you can easily get a void that cockblocks the rest of the chip and the fan being weird is probably easier to attribute to ec software or something, though different thermal behaviour may have an impact.
Yeah probs a good idea to confirm nothing blows up haha
Cool, will do!
Hm, I was thinking that since perhaps the surface/skin temperature sensor doesn’t get as hot as quickly with PTM/LM due to it dumping heat more quickly, it causes the fans to not ramp up as quickly, resulting in a quicker time to throttle due to cooling the chip at a lower fan speed at a specific power. And perhaps this early throttling causes overall reduced performance, until the surface/skin temperature sensor gets hot enough that the fan catches up. Maybe disabling STAPM ignores the surface/skin temperature sensor, allowing the fan to ramp up “normally”.
STAPM is a key feature within their mobile processors. STAPM extends the on-die power management by considering the processor’s internal temperatures taken by on-chip thermal diodes and the laptop’s surface temperature (i.e. the skin temperature). The primary goal of STAPM is to prevent laptops from becoming uncomfortably warm for users, allowing the processor to actively throttle back its heat generation based on the thermal parameters between the chassis and the processor itself.
stapm doesn’t have as much to do with surface temperatures as it sounds like, it’s mostly just a bit of fancy math. for my observations the fancy math does seem to boil down to “if temp > 80 decrement”. Afaik there isn’t actually a skin temp sensor that would inform it about the actual skin temp.
While ptm certainly causes the fan to spin up slower, it should still have about the same ammount of fan at the same temp so it should not cause any more throttling.
Pretty sure there is no skin temp sensor but either way the skin temps are lower with lm/ptm compared to stock paste (as long as the fan is on).
Yeah but they pretty much do that by limiting the time above 80c and tune the coefficients to the particular laptop, which are of course off when using ptm or lm. But with the fw13 tune it’s kinda pointless anyway since the stock power limit is the same as stapm fully throttled.
I just redid some of my own testing. Noticed that PTM7950 was actually performing better than Conductonaut Extreme and was ~200 points higher in Cinebench R23. Temps were very similar with both TIMs with PTM7950 actually running cooler by the end. I do think that the sensor controlling fan rpm isn’t directly based on power draw and isn’t on the cpu itself. I think it’s based on the sensor for skin temperature plus cpu temperature (maybe). Noticed that some cpu sensors were running at 100 C for a while on PTM7950 before the fan decided to kick on.
During both of my tests, the cpu was drawing 35 W (which is stock behaviour) and fan speed was variable. PTM7950 is probably the way to go over liquid metal. I do think liquid metal may be marginally better and the interesting test results I got could be due to insufficient liquid metal however, adding more would increase the risk of it seeping out and shorting things so I didn’t do that. An interesting note with PTM7950 is that in some games, the power draw would suddenly drop from 35 W to 30 W despite the temps being in the 70 C range. I didn’t experience this with liquid metal so this may be due to the high temps during the initial period where it’s beginning to liquify and the fans aren’t on.
Totally agree, it performs pretty much the same with none of the drawbacks of lm.
Even if it did perform a little better ptm is still the much better option.
30 does sound a bit like stapm which does seem to be set to 30 by default and even if you raise it using ryzenadj will come down to 30 if you stay above 80c for long enough.
I guess this would be one advantage of liquid metal or PTM7950 atm however, I do think this could be mitigated or entirely avoided with a custom fan curve based on processor temps. I do notice that with PTM7950, the fans will take up to 2 minutes before kicking on. The developer of x86 universal tuning utility has expressed some interest in developing an app to control power limits and fan curves.
On liquid metal, I wasn’t running into this issue. Temps would climb to around 90 C however, fans would quickly ramp up and bring temps down to around 70-75 C. I’ve attached my cinebench results with temps which should illustrate that.
there is a separate skin temp sensor and it was being used by STAPM
the fans ramp up in connection with that due to STAPM
the fans only ramp up when that separate sensor crosses 80C (or if the power limit is over 30)
then that would mean possibly the fans are kicking in too late with PTM/LM if those materials lead to overall cooler chassis temps and thus a cooler skin sensor temp. If with PTM CPU temps were 100C and the 2nd skin temp sensor was only 70C, the fan RPM might not be as high as if with stock it was 100C and 80C. And if with PTM, at the same CPU temps, RPM was relatively lower because of the cooler skin temp sensor, that would lower the potential power limit/performance at that moment. So if disabling STAPM disconnects reliance on the 2nd skin temp sensor and allows the fans curves to connect to just CPU temps as usual, it may allow overall better performance/resolve the wonkiness.
But yeah that would only make sense if there was a separate sensor and the above were true? I haven’t gotten around to testing this myself though I was thinking of blowing at spots on the mainboard and seeing if the skin temp went down lol
Interesting that PTM with higher temps resulted in a higher score albeit by not much. I’m starting to wonder if the “phase change” properties could be tangibly impacting results (probs some useful info in that LTT vid).
Though I also think the solution to these “issues” is the ability to set custom fan curves/power limits.
With the default power limits I would not expect huge differences, especially for a burst test without heat-soaking.
The ptm does not perform particularly worse than stock or lm even at the 50 or 60C temp limited tests so I kinda doubt the phase change bit has that big an influence.
I believe the phase change properties might be better at filling any microscope gaps that can form between the heatsink and die compared to liquid metal, leading to the lower temps and higher cinebench score. Adding more liquid metal might help it reach parity or do even better but that would dramatically increase the risk of it seeping out.
Honestly I doubt it’s that sophisticated. A slight difference in fan behavior could explain the difference just as well.
If you don’t have any spill out you probably have voids so that’s not good either. which is why you gotta protect the surroundings somehow (nail polish is still the best method I know of even if it can be a bit messy).
But anyway just use ptm XD, especially if you are not messing with the power limits
Maybe but we won’t know at this point since there’s no way to monitor fan speed.
Well it shouldn’t seep out after application period. During my last application, I already put quite a lot as it was starting to pool up. As for the nail polish method, I’ve been using that to protect the SMDs however, there’s also the risk of it shorting other components on the mainboard.
though I have a hunch that the limits might be tuned for the stock paste (could be completely wrong, untested). Anyone know if the weirdness also happens with liquid metal?
I do plan on taking a deeper look into it as well as 
