Leveraging Deep Learning Accelerators on NVIDIA AGX Platforms

Deep learning accelerators (DLA) are a fixed-function inference accelerator hardware integrated into NVIDIA state-of-the-art systems-on-a-chip (SoC). This talk presents a high-level overview of the DLA hardware and software stack. We demonstrate how to leverage the DLA software stack to accelerate a deep learning-based perception pipeline and discuss the workflow to deploy a ResNet 50-based perception network on DLA. Such a workflow lets application developers offload the GPU for other tasks or optimize their application for energy efficiency.

DLA is part of the NVIDIA Orin architecture of the Jetson AGX Orin platform for robotics and the DRIVE AGX Orin platform for safety-critical autonomous vehicle applications.

The topics covered in this talk will get you started with using DLA on your NVIDIA Jetson or DRIVE platform.