My hoodie doesn’t seem to cut it as a blanket, it’s still doing 34w 15minutes in, I’ll try just turning off the fan to up the skin temperature and see if that helps. I am sitting at 49C skin temperature at 21W and 2.7GHz right now maybe that’ll go down further.
That is interesting, most forms of throttling I encountered did not go below 1100MHz since the min frequency in the kernel was changed to that.
This indicates the throttling is neither prochot (would reduce to 545MHz) nor STAPM (would reduce to a minimum of 28 to 30W)
I have been trying but I can’t get skin temp over 50, fan off/ fan on/ fan on low, it’s doing at least 22W and not going over 50. I am running liquid metal but I would think that would make the skin temp situation worse not better.
Might need hotter ambient but I am not creating a sauna just to test that. The ec has been entirely silent apart from my fan changes so idk how involved it is with skin temp stuff.
I am going to stop now cause I don’t like how 50C skin temp smells.
These throttles were the weird mystery throttle from setting the peak power limit too low and setting stapm/short/long power limits to <5W using ryzenadj, I can’t trigger prochot for long enough to see what it actually throttles to.
Strange. Mine stays above 28W/3.5 GHz on all cores when the skin temp approaches that range. Without my plushy dustcloth under the laptop, the time to reproduce this varies greatly from 20 minutes up to 4 hours. This highly depends on room temperatures and other unknown factors.
FIY I’m using the “performance” power mode of my desktop environment, without any ryzenadj tweaks at all.
Never had any smell. Its either the shirt, or may be your very first time you hit 50ºC.
If I turn the fan on high it easily did 30W but then the skin and core temps dropped a lot and I tried to go above the magic 52.
That is quite possible, I also have a dbrand skin on, could be that of-gassing. Either way not a fan.
I guess your machine is not affected. People with my issue may be in the minority and this problem could be caused by variance in production quality. Otherwise more people would complain. But it is also possible that other people either don’t push their machines as hard, or they simply don’t care. Outside of gaming, there are not many other workloads where this problem would be noticable. Benchmarks don’t run long enough, and tasks like compiling code don’t require the user to actively stare at the screen.
Whether my system throttles or not, depends completely on workload I’m running. If the workload allows the skin temp to rise slowly towards 52ºC, the APU will gracefully lower its wattage until the skin temp stabilizes at ~50ºC. Such workloads run forever with no issues.
My only conclusion is that my previous benchmarks increased the skin temp faster than STAPM samples them. Otherwise I would see the same plateauing in wattage as I’ve described above. And being only 2ºC away from a full-brake response explains why it happens so infrequently yet so abruptly.
I’ve done more testing with different STAPM limits. Those limits are, as the name implies, just upper limits. The actual wattage still depends on core temperatures, airflow etc. This explains why I’m not seeing any difference beyond a certain point. Setting slow and fast limits to 43W and 53W respectively (maximum) will simply tell the the APU to pick whatever wattage is possible with the current temps/airflow. The gist I’ve linked earlier is now superseded by a much shorter script which does not need ryzenadj anymore.
Latest solution: GitHub - AlxHnr/amd-ryzen-ignore-stapm: Disable throttling based on chassis (skin) temperature