Search for: All records

Abstract During and after recent La Niña events, the decline of the eastern East African (EA) March‐April‐May (MAM) rains has set the stage for life‐threatening sequential October‐November‐December (OND) and MAM droughts. The MAM 2022 drought was the driest on record, preceded by three poor rainy seasons, and followed by widespread starvation. Connecting these dry seasons is an interaction between La Niña and climate change. This interaction provides important opportunities for long‐lead prediction and proactive disaster risk management, but needs exploration. Here, for the first time, we use observations, reanalyses, and climate change simulations to show that post‐1997 OND La Niña events are robust precursors of: (a) strong MAM “Western V sea surface temperature Gradients” in the Pacific, which (b) help produce large increases in moisture convergence and atmospheric heating near Indonesia, which in turn produce (c) regional shifts in moisture transports and vertical velocities, which (d) help explain the increased frequency of dry EA MAM rainy seasons. We also show that, at 20‐year time scales, increases in atmospheric heating in the Indo‐Pacific Warm Pool region are attributable to warming Western V SST, which is in turn largely attributable to climate change. As energy builds up in the oceans and atmosphere, during and after La Niña events, we see stronger heating and heat convergence over warm tropical waters near Indonesia. The result of this causal chain is that increased Warm Pool atmospheric heating and moisture convergence sets the stage for dangerous sequential droughts in EA. These factors link EA drying to a stronger Walker Circulation and explain the predictable risks associated with recent La Niña events. 

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× (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 behaviour to parameterize extended roofline performance models in future rounds.