Power delivery issue or faulty board?. (Lags severely)

So i’m now about 40 hours hands on, I could use some guidance, because i’m about to use a hammer. lol
First used was 256gb MICROSD 100mb read 90mb write.
USB-C Charger for Motorola ONE (Plenty of Watts/Amperage)
Logitech C270 cam
Used 5w and 10w modes

then migrated over to this setup step by step to troubleshoot as I continue to try to get this to work.

128gb 170mb read 90mb write
Anker PowerPort PD+ 2 33W (used both ports on this, both with high quality cables)
j5create USB-C to USB-C 6ft (with speeds up to 480 Mbps and an output of 3A.)
Logitech C270 cam
Used 5w and 10w modes
Used 4gb swap 6gb swap 10gb swap 16gb swap

Also Decreased monitor frequency to 50hz, and resolution to smallest actually useable.
Put keyboard/mouse/cam/wifi on a (on/off) switch usb hub into the USB 3.0 port.
Created fan system powered outside to ensure proper cooling (tegra stats never showed hot)

Created controled DC output system (2.5A@5V) to connect TWO to the 5V pinouts on j6 40 pin header to power without usb-c.

Some code blocks take few minutes to initialize.
On merge to display Cam/Training/ETC camera has good flow until interaction.
Can get 1x of Thumbs up / Thumbs down, but takes 15 minutes due to severe lag.
Train is not possible, even let run overnight and I just wake up to a locked up system.

How do I know if my nano 2gb is actually bad or not?


Is this an performance issue or you cannot even tell the board is booting up or not?

It sounds like it is booting up but very slow?

Are you able to dump the log from serial console? If any kernel error appears, it will show here.

Boots fine, everything is good until I try to add images or process epoch in any image classification. Then it lags and hangs, CPU uses only ~29%, not seeing any cuda activity if I’m looking at it correctly or not on cuda I’m not sure. I’m providing enough power, I have wondered if I’m just slightly too high on power delivery, I’d like to find where I should check on board past the pinout for voltage/amperage to ensure my pinout connections are indeed delivering the power thoroughly. Changing to 5w shows no improvement, 1 epoch will not process even overnight.

Hi @ssgloganvann, from this image you posted in other thread, it appears like you are browsing JupyterLab from you Jetson. Instead, please browse JupyterLab from your PC. Disconnect the HDMI from your Jetson and reboot, this will help to save memory by running it headlessly. You can then SSH into your Jetson from your PC, start the container, and then navigate the browser on your PC to JupyterLab.

At a glance it doesn’t appear to be a defective board, but rather memory is low and SWAP usage is high. As per the instructions of the ‘Getting Started with AI on Jetson Nano’ DLI course, it is intended to run headlessly and not have a display attached. Browsing JupyterLab from the Jetson itself would be laggy and consume extra memory.

Alternatively, you can try these other training examples from here:

In my testing of Hello AI World, I was able to run those PyTorch trainings on Nano 2GB with display attached. I recommend using the jetson-inference container for it, as it comes with PyTorch pre-installed.

1 Like

It looks like that was the issue indeed. Thank you