Using Orin Nano and Orin NX kits, which ways to improve reasoning speed?

[2029562471]

一、Hardware
The Orin Nano and Orin NX kits have the following specifications:
Orin Nano is 40T,Orin NX is 70T。

二、presentation of condition

1. The model is not large and simply wants to improve the reasoning speed, which is similar to the scale of MobileNetV3.
2. Use mediapipe framework, tflite model.

三、Problem
Using the Orin Nano and Orin NX for reasoning, the Orin NX with higher computing power is 1 millisecond faster than the other two products.
How to speed up reasoning?

[2029562471]

[2029562471]

[kayccc]

Have you set the system perormance to MAX by NVPmodel?

[2029562471]

Performance improved, but not as much as expected

[AastaLLL]

Hi,

Please monitor both tests with tegrastats and check if the GPU utilization reaches 99%.
If not, please use our trtexec binary for benchmarking.

$ /usr/src/tensorrt/bin/trtexec --onnx=[file]

Thanks.

[2029562471]

OK，Thanks