Why has the system break down when I loading the tensorrt model?

I have created the tensorrt model,the system was broken dowen when I loading the model . in the same time I am staring at the memory station, it has enought memory; and this result is not happend in every time,it is randomicity. could I get some useful suggestion for this problem, thank you.

Hi, 327985736

Could try to increase the work space size in your code? The default size may not work well for your model.
In the same time kindly try with FP16 enabled to save more memory cost and get a better performance.

Hi,

May I know what kind of power supply do you use?

It’s known that Nano may shutdown directly if starving for the power.
TensorRT is a highly GPU load library and may cause the system break down.

It’s recommended to use the power supply mentioned here:
https://elinux.org/Jetson_Nano#What_You_Will_Need
https://devtalk.nvidia.com/default/topic/1048640/jetson-nano/power-supply-considerations-for-jetson-nano-developer-kit/

Thanks.

Dear alanz,I didn`t described the problem clearly. The system was power down when I was running the model; and this result is not happend in every time,it is randomicity, So,I was puzzled.

And Dear AastaLLL, MY power supply is enough for nano, it can supply more than 10w-power,and this program as same code is running well on tegra-TX1.
thank you for your help

Dear alanz,I didn`t described the problem clearly. The system was power down when I was running the model; and this result is not happend in every time,it is randomicity, So,I was puzzled.

And Dear AastaLLL, MY power supply is enough for nano, it can supply more than 10w-power,and this program as same code is running well on tegra-TX1.
thank you for your help

Hi, 327985736

As AastaLLL described, a random power down should caused by starving for the power in most cases.
So may know which kind of power supply are your using? USB cable or DC input? 10W is definitely enough for Nano dev kit.

thank you ,the problem has been settled by change a good DC-power。Thanks for alanz and AastaLLL。