An official website of the United States government
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.
more »
« less
DeepFlow: A Cross-Stack Pathfinding Framework for Distributed AI Systems
Over the past decade, machine learning model complexity has grown at an extraordinary rate, as has the scale of the systems training such large models. However there is an alarmingly low hardware utilization (5-20%) in large scale AI systems. The low system utilization is a cumulative effect of minor losses across different layers of the stack, exacerbated by the disconnect between engineers designing different layers spanning across different industries. To address this challenge, in this work we designed a cross-stack performance modelling and design space exploration framework. First, we introduce CrossFlow, a novel framework that enables cross-layer analysis all the way from the technology layer to the algorithmic layer. Next, we introduce DeepFlow (built on top of CrossFlow using machine learning techniques) to automate the design space exploration and co-optimization across different layers of the stack. We have validated CrossFlow’s accuracy with distributed training on real commercial hardware and showcase several DeepFlow case studies demonstrating pitfalls of not optimizing across the technology-hardware-software stack for what is likely, the most important workload driving large development investments in all aspects of computing stack.
more »
« less
- Award ID(s):
- 2231097
- PAR ID:
- 10541688
- Publisher / Repository:
- ACM
- Date Published:
- Journal Name:
- ACM Transactions on Design Automation of Electronic Systems
- ISSN:
- 1084-4309
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Diffractive optical neural networks have shown promising advantages over electronic circuits for accelerating modern machine learning (ML) algorithms. However, it is challenging to achieve fully programmable all‐optical implementation and rapid hardware deployment. Here, a large‐scale, cost‐effective, complex‐valued, and reconfigurable diffractive all‐optical neural networks system in the visible range is demonstrated based on cascaded transmissive twisted nematic liquid crystal spatial light modulators. The employment of categorical reparameterization technique creates a physics‐aware training framework for the fast and accurate deployment of computer‐trained models onto optical hardware. Such a full stack of hardware and software enables not only the experimental demonstration of classifying handwritten digits in standard datasets, but also theoretical analysis and experimental verification of physics‐aware adversarial attacks onto the system, which are generated from a complex‐valued gradient‐based algorithm. The detailed adversarial robustness comparison with conventional multiple layer perceptrons and convolutional neural networks features a distinct statistical adversarial property in diffractive optical neural networks. The developed full stack of software and hardware provides new opportunities of employing diffractive optics in a variety of ML tasks and in the research on optical adversarial ML.more » « less
-
null (Ed.)The data partitioning and scheduling strategies used by DNN accelerators to leverage reuse and perform staging are known as dataflow, which directly impacts the performance and energy efficiency of DNN accelerators. An accelerator micro architecture dictates the dataflow(s) that can be employed to execute layers in a DNN. Selecting a dataflow for a layer can have a large impact on utilization and energy efficiency, but there is a lack of understanding on the choices and consequences of dataflow, and of tools and methodologies to help architects explore the co-optimization design space. In this work, we first introduce a set of data-centric directives to concisely specify the DNN dataflow space in a compiler-friendly form. We then show how these directives can be analyzed to infer various forms of reuse and to exploit them using hardware capabilities. We codify this analysis into an analytical cost model, MAESTRO (Modeling Accelerator Efficiency via Patio-Temporal Reuse and Occupancy), that estimates various cost-benefit tradeoffs of a dataflow including execution time and energy efficiency for a DNN model and hardware configuration. We demonstrate the use of MAESTRO to drive a hardware design space exploration experiment, which searches across 480M designs to identify 2.5M valid designs at an average rate of 0.17M designs per second, including Pareto-optimal throughput- and energy-optimized design points.more » « less
-
Abstract The incorporation of high‐performance optoelectronic devices into photonic neuromorphic processors can substantially accelerate computationally intensive matrix multiplication operations in machine learning (ML) algorithms. However, the conventional designs of individual devices and system are largely disconnected, and the system optimization is limited to the manual exploration of a small design space. Here, a device‐system end‐to‐end design methodology is reported to optimize a free‐space optical general matrix multiplication (GEMM) hardware accelerator by engineering a spatially reconfigurable array made from chalcogenide phase change materials. With a highly parallelized integrated hardware emulator with experimental information, the design of unit device to directly optimize GEMM calculation accuracy is achieved by exploring a large parameter space through reinforcement learning algorithms, including deep Q‐learning neural network, Bayesian optimization, and their cascaded approach. The algorithm‐generated physical quantities show a clear correlation between system performance metrics and device specifications. Furthermore, physics‐aware training approaches are employed to deploy optimized hardware to the tasks of image classification, materials discovery, and a closed‐loop design of optical ML accelerators. The demonstrated framework offers insights into the end‐to‐end and co‐design of optoelectronic devices and systems with reduced human supervision and domain knowledge barriers.more » « less
-
Reasoning about storage systems is challenging because these systems make persistence guarantees even if the system crashes at any point. To achieve these crash-safety guarantees, storage systems include recovery procedures to restore the system to a consistent state after a crash. Moreover, large-scale systems are structured as multiple stacked layers and can require recovery at multiple layers of abstraction. Formal verification can ensure that crash-safety guarantees hold regardless of when the system crashes. To make verification tractable, large-scale systems should be verified in a modular fashion, layer-by-layer in the software stack. Layered recovery makes modularity challenging because the system can crash in the middle of a high-level recovery procedure and must start over from the low-level recovery procedure. We present Argosy, a framework for machine-checked proofs of storage systems that supports layered recovery implementations with modular proofs. The framework is based on combinators for transition relations that are inspired by Kleene algebra, which provides a convenient formalism for specifying and reasoning about crashes and recovery. On top of this framework, we implement Crash Hoare Logic (CHL), the program logic used by FSCQ. Using the logic, we have verified an example of layered recovery featuring a write-ahead log on top of a disk, which itself runs by replicating over two unreliable disks. The metatheory of the framework, the soundness of the program logic, and these examples are all verified in the Coq theorem prover.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1145/3635867
