CPU cooling temperature

I want to keep CPU running at high frequency, so I have changed trip temperature to 95.

But during the stress test, cpu temperature never goes up to 90(very very rare), frequency got automatically degraded, and temperature is around 85.

So where else I can do to let it not to lower the frequency so quickly?

There is hardware protection preventing the CPU to burn anyway, what you want is not increasing the max temperature but dissipating heat so that the CPU never reach this max value (heat sink, fan…). I suggest you to look at existing topics:

https://discuss.96boards.org/t/the-cpu-frequency-of-hikey960
https://discuss.96boards.org/t/cooling-with-mezzanine
https://discuss.96boards.org/t/cooling-the-hikey960-a-better-solution-than-most

Yes, we are looking for cooling solutions,(with fan, it’s able to let big cores run above 1.8G), at the same time, try to adjust limit a little. So it’s not possible in this way ?

Increasing the temperature is definitely not the right solution for increasing performance. Not unless you don’t mind destroying the CPU.

These phone CPUs really aren’t made to operate at maximum performance for extended time periods. They’re made for responsiveness – jump to maximum for a quarter of a second to complete a job and then back down to idle. The PoP configuration also makes it difficult to remove heat from the CPU, since the CPU is insulated by the RAM. You can try a heat sink on the top, combined with a thick thermal pad and heat sink on the bottom, and a fan blowing across the board. This will let you hold a higher average speed, but it will still throttle under load.

I believe that there are more things that can be done in software to reduce heat generation as well, but don’t get your hopes up that they will be implemented soon.

thanks for your patient answer. yes, it’s definitely not the right solution for safe use, just curious about the result, and I guess several degrees celsius more doesn’t help much.
also plan to experiment with a thermoelectric cooler.
thanks again.