서브메뉴
검색
Accelerating Deep Learning in the Post-Moore's Era of Computing
Accelerating Deep Learning in the Post-Moore's Era of Computing
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260209102839
- ISBN
- 9798314843642
- DDC
- 004
- 저자명
- Hegde, Kartik.
- 서명/저자
- Accelerating Deep Learning in the Post-Moores Era of Computing
- 발행사항
- [Sl] : University of Illinois at Urbana-Champaign, 2023
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2023
- 형태사항
- 152 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
- 주기사항
- Advisor: Fletcher, Christopher W.
- 학위논문주기
- Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2023.
- 초록/해제
- 요약Rapid growth in the availability of cheap and fast computing power-made possible by Moore's scaling-has been a major tailwind behind the rise of modern deep learning. However, over the last decade, the performance deficit between compute requirements of deep learning workloads and the peak compute capability offered by modern processors has been rapidly increasing due to two opposing trends. On the one hand, improving the accuracy of deep learning models requires larger models and datasets, which in turn requires more compute resources. On the other hand, Moore's scaling has been slowing down, leading to a slowdown in the rate of increase in peak compute throughput. While specialization has been seen as a way to address the increasing performance deficit, there is mounting evidence that specialization has only afforded us a "one-time" boost in performance. To make matters worse, the increasing gap between logic and memory, often called the memory wall, makes it harder to utilize all the compute on-chip, let alone add more. Therefore, the key question facing deep learning architects is, how do we continue to accelerate deep learning workloads in the post-Moore's era of computing?This thesis presents a set of techniques to accelerate deep learning workloads without scaling the number of transistors on-chip. At a high level, we classify them into two thrusts. First, techniques that reduce the total compute required for deep learning workloads without sacrificing accuracy and utilize that to improve performance. Second, techniques that increase the compute-per-transistor by improving the resource utilization of deep learning accelerators. Notably, we show that the proposed techniques are not "zero-sum", i.e., the gained performance comes at negligible area/power overheads. We begin by describing a baseline deep learning accelerator (DLA) tailored for deep learning workloads and represents state-of-the-art deep learning accelerators. We then describe concrete ways to evolve the baseline architecture to incorporate our proposed techniques. Finally, we provide a detailed analysis of the area overheads of the proposed techniques and demonstrate that they are minimal while the performance gains are significant.Overall, this thesis throws light on different directions that computer architects can take to continue to accelerate deep learning workloads without scaling the number of transistors on-chip. We believe that expanding the scope of specialization beyond hardware to include other layers of the deep learning stack and careful co-design will enable us to continue to accelerate deep learning workloads in the post-Moore's era of computing.
- 일반주제명
- Computer science
- 일반주제명
- Engineering
- 일반주제명
- Information technology
- 키워드
- Deep learning
- 키워드
- Moore's law
- 기타저자
- University of Illinois at Urbana-Champaign Computer Science
- 기본자료저록
- Dissertations Abstracts International. 86-11B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260203s2023 us c eng d■001000017365854
■00520260209102839
■006m o d
■007cr#unu||||||||
■020 ▼a9798314843642
■035 ▼a(MiAaPQ)AAI32063794
■035 ▼a(MiAaPQ)httphdlhandlenet2142120125
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aHegde, Kartik.
■24510▼aAccelerating Deep Learning in the Post-Moore's Era of Computing
■260 ▼a[Sl]▼bUniversity of Illinois at Urbana-Champaign▼c2023
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2023
■300 ▼a152 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-11, Section: B.
■500 ▼aAdvisor: Fletcher, Christopher W.
■5021 ▼aThesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2023.
■520 ▼aRapid growth in the availability of cheap and fast computing power-made possible by Moore's scaling-has been a major tailwind behind the rise of modern deep learning. However, over the last decade, the performance deficit between compute requirements of deep learning workloads and the peak compute capability offered by modern processors has been rapidly increasing due to two opposing trends. On the one hand, improving the accuracy of deep learning models requires larger models and datasets, which in turn requires more compute resources. On the other hand, Moore's scaling has been slowing down, leading to a slowdown in the rate of increase in peak compute throughput. While specialization has been seen as a way to address the increasing performance deficit, there is mounting evidence that specialization has only afforded us a "one-time" boost in performance. To make matters worse, the increasing gap between logic and memory, often called the memory wall, makes it harder to utilize all the compute on-chip, let alone add more. Therefore, the key question facing deep learning architects is, how do we continue to accelerate deep learning workloads in the post-Moore's era of computing?This thesis presents a set of techniques to accelerate deep learning workloads without scaling the number of transistors on-chip. At a high level, we classify them into two thrusts. First, techniques that reduce the total compute required for deep learning workloads without sacrificing accuracy and utilize that to improve performance. Second, techniques that increase the compute-per-transistor by improving the resource utilization of deep learning accelerators. Notably, we show that the proposed techniques are not "zero-sum", i.e., the gained performance comes at negligible area/power overheads. We begin by describing a baseline deep learning accelerator (DLA) tailored for deep learning workloads and represents state-of-the-art deep learning accelerators. We then describe concrete ways to evolve the baseline architecture to incorporate our proposed techniques. Finally, we provide a detailed analysis of the area overheads of the proposed techniques and demonstrate that they are minimal while the performance gains are significant.Overall, this thesis throws light on different directions that computer architects can take to continue to accelerate deep learning workloads without scaling the number of transistors on-chip. We believe that expanding the scope of specialization beyond hardware to include other layers of the deep learning stack and careful co-design will enable us to continue to accelerate deep learning workloads in the post-Moore's era of computing.
■590 ▼aSchool code: 0090.
■650 4▼aComputer science
■650 4▼aEngineering
■650 4▼aInformation technology
■653 ▼aDeep learning
■653 ▼aComputer architecture
■653 ▼aMoore's law
■653 ▼aSpecialized accelerators
■653 ▼aDeep neural networks
■653 ▼aProgrammable accelerators
■690 ▼a0984
■690 ▼a0489
■690 ▼a0800
■690 ▼a0537
■71020▼aUniversity of Illinois at Urbana-Champaign▼bComputer Science.
■7730 ▼tDissertations Abstracts International▼g86-11B.
■790 ▼a0090
■791 ▼aPh.D.
■792 ▼a2023
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17365854▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


