Exploring CPU voltages and undervolting

[DarkBahamut]

As part of my overclock testing a few days ago I also added OS level voltage control for the CPUs to aid that and to also look into possible undervolting. I attempted the same thing back on the XU4 but it never really had much headroom and the gains were quite limited. However much like the overclocking results, the N1/RK3399 seems to have a bit more headroom available and this can translate to some reasonable improvements in power usage and heat.

I don't have the larger HK passive heatsink myself to be able to test but I think it might be possible to avoid nearly all throttling on it if you can get the voltages down a bit. This is how I think I'll look to run my N1 long term as I don't think the fan will be needed much at all with a little tweaking

I ran a bit of a test/benchmark below as a proof of concept and to gauge the performance gains available. It takes a while to work out and test the lowest stable voltages but at least it translates to something this time

Run 1:

Stock CPU voltages
Stock active cooler - fan disabled (running passive)
test: Compiling N1 kernel


First throttling occurs after 171 seconds
Average clockspeed: 1586MHz
Average Temp: 80.2C


Run 2:

Undervolted to lowest stable voltages for my sample (-87.5mV to -112.5mV on A53 cluster / -75mV on A72 cluster)
Stock active cooler - fan disabled (running passive)
test: Compiling N1 kernel


First throttling occurs after 302 seconds
Average Clockspeed: 1796MHz
Average Temp: 78.4C

[rooted]

Nice work, I have the passive heatsink but no thermal paste currently to test it.

[bronco]

Did you try out testing for reliability of your undervolting with something like this Linpack: https://github.com/ehoutsma/StabilityTester

Background information/discussion: https://www.raspberrypi.org/forums/view ... 9&p=927615

The above test script using this Linpack is direct result of the Linpack behaviour (to show data corruption already before the system crashes)