[dogma1138]

For servers no, but we’re talking about endpoints here. Also this isn’t only about reducing the existing vector bandwidth but also about not increasing it outside of dedicated co-processors.