How many concurrent AI models can an industrial AI edge gateway run at most simultaneously?

The number of concurrent AI models in industrial AI edge gateways is typically limited by computing power and memory, usually ranging from 2 to 8. When multiple models are running concurrently, the frequent data flows of each model access the DDR and PCIe buses, which can easily generate synchronous switching noise and lead to power integrity issues.

Countermeasure 1: Divide the core power plane and isolate each channel with independent ferrite beads (PBZ1608E600Z0T recommended).

Countermeasure 2: Connect a common-mode choke (CMZ2012A-900T) in series to the PCIe channel associated with each model to suppress inter-channel crosstalk.

Countermeasure 3: Add a surface-mount ferrite bead array (BLA31BD600SN4L) to the data line between the AI chip and DDR to reduce reflection noise.

Simultaneously, connect a TVS 5.0SMDJ33CA in parallel at the power input of the entire system to absorb surges. Actual measurements show that even with 4 models running concurrently, the radiated emission still meets the Class B margin of 4dB.