Nvidia triples and Intel doubles generative AI inference performance on new MLPerf benchmark