MLPerf HPC: A Holistic Benchmark Suite for Scientific Machine Learning on HPC Systems

• Main Authors: Farrell, Steven, Emani, Murali, Balma, Jacob, Drescher, Lukas, Drozd, Aleksandr, Fink, Andreas, Fox, Geoffrey, Kanter, David, Kurth, Thorsten, Mattson, Peter, Mu, Dawei, Ruhela, Amit, Sato, Kento, Shirahata, Koichi, Tabaru, Tsuguchika, Tsaris, Aristeidis, Balewski, Jan, Cumming, Ben, Danjo, Takumi, Domke, Jens, Fukai, Takaaki, Fukumoto, Naoto, Fukushi, Tatsuya, Gerofi, Balazs, Honda, Takumi, Imamura, Toshiyuki, Kasagi, Akihiko, Kawakami, Kentaro, Kudo, Shuhei, Kuroda, Akiyoshi, Martinasso, Maxime, Matsuoka, Satoshi, Mendonça, Henrique, Minami, Kazuki, Ram, Prabhat, Sawada, Takashi, Shankar, Mallikarjun, John, Tom St, Tabuchi, Akihiro, Vishwanath, Venkatram, Wahib, Mohamed, Yamazaki, Masafumi, Yin, Junqi
• Format: Journal Article
• Language: English
• Published: 21.10.2021
• Subjects: Computer Science - Distributed, Parallel, and Cluster Computing; Computer Science - Learning
• DOI: 10.48550/arxiv.2110.11466

Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning applications that are representative of real-world scientific use cases. MLPerf is a community-driven standard to benchmark machine learning workloads, focusing on end-to-end performance metrics. In this paper, we introduce MLPerf HPC, a benchmark suite of large-scale scientific machine learning training applications driven by the MLCommons Association. We present the results from the first submission round, including a diverse set of some of the world's largest HPC systems. We develop a systematic framework for their joint analysis and compare them in terms of data staging, algorithmic convergence, and compute performance. As a result, we gain a quantitative understanding of optimizations on different subsystems such as staging and on-node loading of data, compute-unit utilization, and communication scheduling, enabling overall $>10 \times$ (end-to-end) performance improvements through system scaling. Notably, our analysis shows a scale-dependent interplay between the dataset size, a system's memory hierarchy, and training convergence that underlines the importance of near-compute storage. To overcome the data-parallel scalability challenge at large batch sizes, we discuss specific learning techniques and hybrid data-and-model parallelism that are effective on large systems. We conclude by characterizing each benchmark with respect to low-level memory, I/O, and network behavior to parameterize extended roofline performance models in future rounds.