Yolo causes jetson nano devkit with GPU to power off

I have a jetson nano devkit with a wifi card (waveshare AC8265 Wireless NIC Module for Jetson Nano Supports 2.4GHz / 5GHz Dual Band WiFi and Bluetooth 4.2), usb mouse, usb keyboard, camera (Raspberry Pi Camera Module V2-8 Megapixel,1080p), and a fan (Noctua NF-A4x20 PWM, Premium Quiet Fan, 4-Pin (40x20mm, Brown)). I compiled and test Yolo with no GPU or NVCC and was able to get some image detection. The detection was slow but it shows that Yolo will work on the Nano. I then attempted to build and use Yolo with GPU enabled. As soon as Yolo finishes the initialization and the camera is about to show on the screen, the nano shuts off. The green light on the nano board goes dim. The only way to get it going again is to unplug and re-plug the board back into power. For power I tried two new power plugs. First I tried SMAKN DC 5V/4A 20W Switching Power Supply Adapter 100-240 Ac, and when I saw the power turn off during the Yolo run I suspected the power supply might not be sufficient. I ordered and tried a 5V 15A 75W Power Supply 100V-240V or 110V - 220V AC to DC Adapter 5V 15 amp Switching Converter Charger 5.5x2.1mm Plug for WS2811 WS2812B WS2813 2801 LED Strip Pixel Lights. I was hopeful this would work but it resulted in the same behavior.

I created an 8g swapfile to try and head off an out-of-memory condition. That didnt work. I set the nano to use max power using the settings provided. That didnt help. I tested with tegrastats - although I do not know how to read the output, it ran and ran without crashing.

Can someone tell me how to tell what is happening that creates this behavior? The Nano just shuts off I have no way to debug what is happening. I read a topic that claims if too much power is requested by the GPU, the nano will shut off. I think the power supplies I am using have sufficient energy for the board to consume. Yolo works on the regular CPU, but it halts the Nano when GPU is accessed. Maybe popping up the camera viewer in addition to the Yolo activity is causing some nasty condition?

Has this happened to anyone else yet?

First, try the various power modes, including the low-power modes, to see if any of them works.
You can see them all in /etc/nvpmodel.conf, and apply them with “sudo nvpmodel -m X” for X is 0, 1, …

Actually, the Nano only has two modes, 0, and 1, where 1 is low-power, and 0 is high-power, so try with -m 1.
You can also add more modes that you tweak yourself in the nvpmodel.conf file.

Then, you can run the “tegrastats” tool in a separate window, ideally over SSH, which will tell you what the temperature and power draw is for each sensors. This only prints a value once a second, so an instant change will not get caught, but it might give some clues.

Then, you can add a serial cable to the serial console, and monitor that from another computer, to see if the kernel gets to print something.

Then, you can hook up an oscilloscope to the various rails on the board – 5V and 3.3V are easily available on the 40-pin GPIO header; see if a change on those rails correlates with starting the GPU.

Also, it may be as simple as the module not being properly seated in its holder. Try jiggling and re-seating it. If the contacts aren’t firm, then there will be voltage loss when power draw goes up, which can cause a shutdown.

un-seating and re-seating the GPU did the trick. Thank you.

Sometimes it’s the simplest things :-)