Comment by liorn

It's not a one time thing though - onprem inference requires continuous monitoring and tuning to get the most out of your GPUs. Looking at 25-50% performance benefits. Today, it does require dedicated Inference Engineers...