I am using Jetson Nano on the official Developer Kit carrier board. I am powering it via the header pins with 5V@6A. I measure stable 4.99V on the header pins with a multimeter.
The sys logs show OC ALARM 0x00000001, and the GUI shows system throttled due to over-current.
Running sudo cat /sys/bus/i2c/drivers/ina3221x/6-0040/iio:device0/in_voltage0_input reads 4616 with no periferrals attached.
The Jetson Nano eventually shuts down, I guess due to being under 4.75V.
How do I solve this? Should I just apply 5.25V to the header pins? Is there any configuration required on the board I am missing? Any jumper?
I don’t know, it could be undervoltage causing shutdown, but consider that over-current is throttling when the given performance model is drawing too much power for that model. Imagine that you are in some mode other than max performance, and the system is drawing too much power; in that case, it might throttle clocks to lower power consumption, and the log note would be the alarm you saw and a note of system throttle due to over-current. I can’t say, but it is quite possible at the point you measured that 4.75V is valid.
However, Jetsons are particularly sensitive to power regulation quality. I’ve powered via a bench supply many times, and tend to set to 5.20V (Nanos take lower voltage than other models, be careful to not use another model’s power adapter). I don’t think it would hurt to set to that, but I don’t know about doing so through the header pins. I’ve used the barrel jack when I did that.
What do you see from: sudo nvpmodel -p --verbose | grep -i 'pm_config'
Then try running “sudo nvpmodel -m 0” and see if you get the same overcurrent warning. In max performance mode it can still throttle clocks due to overcurrent, but in that case it is the safety limit and preventing too much heat rather than as a convenience to run in a lower power mode. You’d be less likely to see the OC warning. If it is shutting down, then one possibility is (A) the power supply is not regulating well as power consumption goes up, or (B) total heat is too much, e.g., a bad heat sink setup, or (C) the power supply is indeed too low of a voltage (if a component on some individual bus is failing and making one bus unstable, then I could see the external voltage being of good quality and still shutting down for that reason).
Someone else would have to tell you if the voltages you measured are out of tolerance or not, but I suspect it is an issue of the power supply having inadequate regulation under load.
The 5V input that I have used is the barrel connector. I’ve not applied power directly to the pins. This is basically the same spec as the USB2 5 volt spec. I think the spec allows a 5% tolerance, but for Jetsons, I would not recommend dropping below 5.0 V. 5% gives you a maximum value (without violation of standards) of 5.25 V. If you have a bench or other supply which can provide 5.20 V to 5.25 V you might try that.
However, that isn’t where the story ends. Regulation stability as power load is applied still matters. When under load, does the power drop substantially? Quick changes can cause shutdown. One can put a large capacitor right next to where the power supply enters to aid with power spikes (e.g., 2000 uF; I tend to also suggest a 1 uF tantalum capacitor at the same location).
Then you have to consider long term voltage drops. Short spikes are taken care of by a capacitor, but if the supply drops voltage under load, and stays too low, then the supply probably needs something of higher quality.
Then, within the Jetson itself, there are power regulation components. If the input power is of proper voltage, and if spikes and droops in power are within tolerance, then the internal components should not care. It is possible for regulator components within the carrier board to be out of spec though. Someone from NVIDIA will have to tell you if the points you measured are within spec or not.
Note that MAXN will not show an OC alarm until it reaches its design limit. At that point you are seeing clocks being reduced to prevent power consumption from exceeding maximum thermal limits. This is not an error, nor is it something of any concern. However, what it does point out is that at this point when it hits OC at MAXN, that the system is also consuming the highest current it can consume. The external power supply is at its highest limit at that point, and so if that power supply is inadequate, then this is when it would fail.
The 2000 uF should help rather than harm. However, if the power supply itself has trouble with the inrush current, then perhaps voltage is not stable soon enough. Can you add the 2000 uF to the power supply cable near the Jetson, and turn the power on before connecting to the Jetson? I’m assuming you started power to both the capacitor and the Jetson at the same time, but if we change the timing so that the 2000 uF is charged prior to the Jetson seeing this, then it should help. Whether or not it succeeds this is a useful test to find out more about the power source.
Someone from NVIDIA would have to comment on whether or not your /sys monitor of voltage is in spec or not. Internally I would expect some voltages to be a bit lower due to various circuitry, but I have no way of knowing if that voltage is valid or not.
