Intel, Nvidia and Google have made significant strides in recent months that enable faster LLM training results in the MLPerf Training 3.1 benchmarks.