我正在使用NVIDIA AGX Orin 64GB开发者套件推理 VAD.onnx 模型，但是性能很差（约3.5fps），我注意到jetson_release显示 DLA0 DLA1 off，我应该如何激活DLA0,DLA1，为了使用它们？同时，我可以使用什么工具，查看编译出的 VAD.engine 的性能瓶颈点，提高推理效率
I am using the NVIDIA AGX Orin 64GB Developer Kit to infer the VAD.onnx model, but the performance is very poor (approximately 3.5 FPS). I noticed that jetson_release shows DLA0 and DLA1 as off. How should I enable DLA0 and DLA1 to use them? Additionally, what tools can I use to identify performance bottlenecks in the compiled VAD.engine and improve inference efficiency?
VAD.onnx
Hi,
When deploying a model with TensorRT, please use --useDLACore=[ID] to use DLA for inference.
Please find more details in our document below:
Thanks.
我可以怎样分析模型的性能瓶颈呢？
I need to analyze the performance bottlenecks of the model. How should I do it?
Hi,
Do you want to profile a model that runs on the DLA?
If so, please find below link for a tutorial:
Thanks.
don’t find nvvp or Nvidia Visual Profiler on Jetson AGX Orin Developer Kit
sudo apt install nvidia-jetpack should install them. After installation, you can use nsys to profile your program and nsys-ui to view your report in GUI.
thank you for your reply
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.