Does the industrial AI edge gateway support multi-model pipeline collaborative inference?

Industrial AI edge gateways support multi-model pipelines, such as object detection + classification + tracking collaborative inference pipelines. Frequent data transfer between GPU-NPU-DDR can easily lead to power rail collapse. Solutions include:

1) Parallel connection of a TVS 5.0SMDJ33CA and a 220μF electrolytic capacitor to the 12V input;

2) Constructing a low-pass filter by connecting a PBZ1608E600Z0T ferrite bead and a 10μF MLCC to the core voltage VDD_GPU;

3) Adding a common-mode inductor CMZ2012A-900T to the PCIe Gen3 signal to reduce radiation mutual coupling.

Actual measured inference speed for a 3-model pipeline is 120ms, with EFT fluctuation <5ms at ±4kV.