Inference TCO

GPUs dominate highly parallel inference, but a CPU server can be the better purchase for small models, low traffic and consolidated workloads.