No doubt. Especially for GPU inference at scale. We overclock/overvolt for training and tune way down for inference.