On 11/07/2017 05:41 AM, Michael Mrozek wrote:
On Di, 2017-11-07 at 02:57 +0200, Grazvydas Ignotas wrote:
One thing would be interesting: Is it the battery itself that heats up that much (because it has to provide a lot of power) or is it the CPU board that heats up the battery?
I keep the charger connected, so it's not discharging the battery much.
The latter one should be fixed with the aluminium plate we have between the CPU board and the battery (at least on ptitSeb's unit it never got warm running with full CPU power at 1GHz).
1.5GHz full CPU power seems impossible though, the OMAP5 heats up way too much for such a small system. Short 1.5GHz peaks in normal operation are no issue, but using a lot of CPU definitely gets too hot within seconds.
Like said I'm doing all my tests at 1.2GHz, so too-hot-to-hold battery was with that. In fact yesterday I had a thermal shutdown with 1.2GHz also.
Additionally, the throttler seems to be one of the worst I've ever seen... It seems to throttle to 300MHz suddenly and stays there for a while (don't know for how long)
A good throttler would dynamically lower the CPU power to keep the heat below a certain temperature, but it shouldn't suddenly throttle to lowest CPU speed...
I guess there's work to be done here.
Yeah that seems to work.
But my board boots reliably at 1.35V even without any tweaks, and I've had a 2+ hour dual memtester run at no-tweaks 1.35V before crashing, so it's really hard to tell if something helps because it's so unpredictable. It's also hard to do longer tests because I don't want to leave the device unattended on stresses as it heats the battery so much.