Intentionally throttling my Jetson Nano

I am trying to study the effect of throttling on the Nano and am running multiple CNNs from jetson-inference GitHub repo. My setup and questions about throttling are mostly the same as the OP in this thread Trying to thermally throttle the Jetson Nano but it gets throttled by overcurrent but I have some lingering questions:

(1) I can only get the device’s temperature up to around 85C. But according to the thermal specs here Tegra Linux Driver, throttling will kick in at temperatures >= 97C for the Nano unless I’m misunderstanding something. Why am I unable to get the temperature to go higher? Is my device not getting enough power? I am using this power supply

(2) I am not seeing any drop in performance (CNN latency, throughput), temperature, or GPU frequency even when I get the “throttling due to over current” over several hours. The answer in the thread linked suggests that I need to run tegrastats at shorter timescales (i.e., 500ms vs ~2 seconds). But if I need to do that, then the drops must be quite insignificant and the clock rates/temperatures must bounce back quickly? If so, is there even a point in throttling to such a small extent if it has such a negligible impact?

Thank you for your time, I appreciate it.

Please remove the heatsink for a try. And execute sudo tegrastats to check the system status.

Hello @mr_broccoli,

I see you referenced my thread.

To throttle my Jetson Nano, I had to remove the heatsink and run a TensorFlow multiplication code in a loop.

For the overcurrent issue, it also shows up when there is less current and most commercial 3A power supplies are trash, they don’t provide exact 3 amps.

I bought a CanaKit 5.1V, 3.5A Type-C charger and my overcurrent issue was resolved.

Kind regards,

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.