I’m encountering a persistent “system throttled due to over-current” error when using nvcr.io/nvidia/vllm:25.12.post1-py3 to deploy the Qwen3-VL-8B model on my Jetson Thor for multi-image inference tasks.
The issue occurs consistently across all available power modes ( 120W 、MaxN mode) whenever the workload increases during inference. Currently, I am powering the device via its Type-C port.
My questions are:
Is it safe to ignore this error and continue operation, or does it indicate a risk of hardware damage?
Could this be a power supply limitation? Would switching to the official barrel connector (DC) power supply likely resolve or mitigate this over-current throttling?
Any insights or suggestions would be greatly appreciated. Thank you.
Hi, I am using the Jetson Thor devkit with JetPack 7.1.
When running 5 concurrent requests, each processing 4 images of 1280*720 resolution for inference, the over-current error occurs consistently.
I would like to know if using DC power supply would yield better performance. Where can I purchase a compatible DC power adapter? Why does Thor retain the DC power interface?
Both the DC input and USB-C port can serve as power sources, though their power delivery capabilities may differ.
To achieve optimal performance , you should configure the power settings to match your specific workload and requirements."