CNS Core: Medium: Real-time Energy-elastic GPUs for Embedded Autonomous Systems

CNS 核心：中：用于嵌入式自治系统的实时能量弹性 GPU

• 批准号: 1955650
• 负责人: Daniel Wong
• 金额: $ 119.99万
• 依托单位: University of California-Riverside
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-10-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1955650&HistoricalAwards=false
• 关键词: CNS; Core; Medium; Real; time

Graphical Processing Units (GPUs) provide massive computational power that enables modern transformative applications such as computer vision, machine learning, and scientific discovery. However, these benefits are typically confined to large-scale computer systems, such as cloud computing and supercomputers. Embedded resource-constrained environments, such as autonomous systems and aerial drones, can make limited use of GPUs due to constraints on available energy, computing and communication resources, and timing requirements (real-time constraints). This project fundamentally rethinks the design of GPU-enabled computer systems in embedded resource-constrained environments with real-time behavior. This research consists of three main goals: (1) to develop energy-efficient and elastic microarchitectures for computational resources and storage structures, (2) to develop timing-aware GPU hardware schedulers that coordinate with an operating system real-time task scheduler, and (3) to dynamically balance timing requirements and energy-elastic microarchitectures by holistically coordinating across the entire software-hardware computing stack.These next-generation real-time embedded GPUs provide more computational power to enable future embedded autonomous systems to become safer and more energy-efficient. This proliferation of computing power can also lead to myriad impacts on society such as improved portable medical devices that are smaller and with expanded capabilities, automobile collision detection systems with improved accuracy and safety, and lower electricity consumption for internet-of-things and even for larger-scale GPU-based computer systems. This project will offer research opportunities to undergraduate students, and broaden participation of underrepresented minorities, providing training to a new generation of computer engineers who will meet the design challenges of future embedded systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

图形处理器(GPU)提供强大的计算能力，支持计算机视觉、机器学习和科学发现等现代变革性应用。然而，这些好处通常仅限于大型计算机系统，如云计算和超级计算机。嵌入式资源受限的环境，如自主系统和无人机，由于可用能源、计算和通信资源的限制以及时间要求(实时限制)，可以有限地使用GPU。该项目从根本上重新考虑了在具有实时行为的嵌入式资源受限环境中支持GPU的计算机系统的设计。该研究包括三个主要目标：(1)为计算资源和存储结构开发节能和弹性的微体系结构；(2)开发与操作系统实时任务调度器协调的时序感知GPU硬件调度器；(3)通过整体协调整个软硬件计算堆栈，动态平衡时序需求和能量弹性微体系结构。这些新一代实时嵌入式GPU提供更多的计算能力，使未来的嵌入式自主系统变得更加安全和节能。计算能力的激增还可能给社会带来无数影响，例如体积更小、功能更强的改进型便携式医疗设备、精度和安全性更高的汽车碰撞检测系统，以及物联网甚至更大规模的基于GPU的计算机系统的更低功耗。该项目将为本科生提供研究机会，并扩大少数族裔的参与，为新一代计算机工程师提供培训，他们将迎接未来嵌入式系统的设计挑战。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

KRISP: Enabling Kernel-wise RIght-sizing for Spatial Partitioned GPU Inference Servers

• DOI: 10.1109/hpca56546.2023.10071121
• 发表时间: 2023-02
• 期刊: 2023 IEEE International Symposium on High-Performance Computer Architecture (HPCA)
• 作者: M. Chow;Ali Jahanshahi;Daniel Wong

JetStream: Graph Analytics on Streaming Data with Event-Driven Hardware Accelerator

• DOI: 10.1145/3466752.3480126
• 发表时间: 2021-10
• 期刊: MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture
• 作者: Shafiur Rahman;Mahbod Afarin;N. Abu-Ghazaleh;Rajiv Gupta

BlockMaestro: Enabling Programmer-Transparent Task-based Execution in GPU Systems

• DOI: 10.1109/isca52012.2021.00034
• 发表时间: 2021-06
• 期刊: 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA)
• 作者: AmirAli Abdolrashidi;Hodjat Asghari Esfeden;A. Jahanshahi;Kaustubh Singh;N. Abu-Ghazaleh;Daniel Wong

ScaleServe: a scalable multi-GPU machine learning inference system and benchmarking suite

ScaleServe：可扩展的多 GPU 机器学习推理系统和基准测试套件

• DOI: 10.1145/3530390.3532735
• 发表时间: 2022
• 期刊: Proceedings of the 14th Workshop on General Purpose Processing Using GPU
• 作者: Jahanshahi, Ali;Chow, Marcus;Wong, Daniel
• 通讯作者: Wong, Daniel

Balancing Energy Efficiency and Real-Time Performance in GPU Scheduling

• DOI: 10.1109/rtss52674.2021.00021
• 发表时间: 2021-12
• 期刊: 2021 IEEE Real-Time Systems Symposium (RTSS)
• 作者: Yidi Wang;Mohsen Karimi;Yecheng Xiang;Hyoseung Kim