Demystifying the MLPerf Benchmark Suite