10:00 AM – 10:25 AM
AI is moving out of the cloud and into the machine. Vision inspection, predictive maintenance, smart HMIs, machine builders everywhere are being asked to put AI inside their systems. But unlike software, the hardware you choose is locked in for the lifetime of the machine. Get it wrong, and you either pay for compute you never use, or redesign when the AI outgrows the box.
This session shows a practical way to size edge AI hardware: how latency, camera streams, and model size turn into real GPU and memory requirements and why a GPU that is great for training can be surprisingly slow for inference. We compare the real options like embedded NPUs, NVIDIA Jetson, and PCIe GPUs, with simple rules on power, cost, and lifetime. You will leave with a sizing method you can apply to your next machine design.
Speaker: Amol Jadhav – ADLINK on behalf of Alcom Electronics