Comparative evaluation of deep learning workloads for leadership-class systems
Deep learning (DL) workloads and their performance at scale are becoming important factors to consider as we design, develop and deploy next-generation high-performance computing systems. Since DL applications rely heavily on DL frameworks and underlying compute (CPU/GPU) stacks, it is essential to...
Main Authors: | , , , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
KeAi Communications Co. Ltd.
2021-10-01
|
Series: | BenchCouncil Transactions on Benchmarks, Standards and Evaluations |
Subjects: | |
Online Access: | http://www.sciencedirect.com/science/article/pii/S2772485921000053 |
_version_ | 1811292596926939136 |
---|---|
author | Junqi Yin Aristeidis Tsaris Sajal Dash Ross Miller Feiyi Wang Mallikarjun (Arjun) Shankar |
author_facet | Junqi Yin Aristeidis Tsaris Sajal Dash Ross Miller Feiyi Wang Mallikarjun (Arjun) Shankar |
author_sort | Junqi Yin |
collection | DOAJ |
description | Deep learning (DL) workloads and their performance at scale are becoming important factors to consider as we design, develop and deploy next-generation high-performance computing systems. Since DL applications rely heavily on DL frameworks and underlying compute (CPU/GPU) stacks, it is essential to gain a holistic understanding from compute kernels, models, and frameworks of popular DL stacks, and to assess their impact on science-driven, mission-critical applications. At Oak Ridge Leadership Computing Facility (OLCF), we employ a set of micro and macro DL benchmarks established through the Collaboration of Oak Ridge, Argonne, and Livermore (CORAL) to evaluate the AI readiness of our next-generation supercomputers. In this paper, we present our early observations and performance benchmark comparisons between the Nvidia V100 based Summit system with its CUDA stack and an AMD MI100 based testbed system with its ROCm stack. We take a layered perspective on DL benchmarking and point to opportunities for future optimizations in the technologies that we consider. |
first_indexed | 2024-04-13T04:47:49Z |
format | Article |
id | doaj.art-523170481d1b483892fdbf719a600830 |
institution | Directory Open Access Journal |
issn | 2772-4859 |
language | English |
last_indexed | 2024-04-13T04:47:49Z |
publishDate | 2021-10-01 |
publisher | KeAi Communications Co. Ltd. |
record_format | Article |
series | BenchCouncil Transactions on Benchmarks, Standards and Evaluations |
spelling | doaj.art-523170481d1b483892fdbf719a6008302022-12-22T03:01:47ZengKeAi Communications Co. Ltd.BenchCouncil Transactions on Benchmarks, Standards and Evaluations2772-48592021-10-0111100005Comparative evaluation of deep learning workloads for leadership-class systemsJunqi Yin0Aristeidis Tsaris1Sajal Dash2Ross Miller3Feiyi Wang4Mallikarjun (Arjun) Shankar5Corresponding author.; Oak Ridge National Laboratory, United States of AmericaOak Ridge National Laboratory, United States of AmericaOak Ridge National Laboratory, United States of AmericaOak Ridge National Laboratory, United States of AmericaOak Ridge National Laboratory, United States of AmericaOak Ridge National Laboratory, United States of AmericaDeep learning (DL) workloads and their performance at scale are becoming important factors to consider as we design, develop and deploy next-generation high-performance computing systems. Since DL applications rely heavily on DL frameworks and underlying compute (CPU/GPU) stacks, it is essential to gain a holistic understanding from compute kernels, models, and frameworks of popular DL stacks, and to assess their impact on science-driven, mission-critical applications. At Oak Ridge Leadership Computing Facility (OLCF), we employ a set of micro and macro DL benchmarks established through the Collaboration of Oak Ridge, Argonne, and Livermore (CORAL) to evaluate the AI readiness of our next-generation supercomputers. In this paper, we present our early observations and performance benchmark comparisons between the Nvidia V100 based Summit system with its CUDA stack and an AMD MI100 based testbed system with its ROCm stack. We take a layered perspective on DL benchmarking and point to opportunities for future optimizations in the technologies that we consider.http://www.sciencedirect.com/science/article/pii/S2772485921000053CORAL benchmarkDeep learning stackROCm |
spellingShingle | Junqi Yin Aristeidis Tsaris Sajal Dash Ross Miller Feiyi Wang Mallikarjun (Arjun) Shankar Comparative evaluation of deep learning workloads for leadership-class systems BenchCouncil Transactions on Benchmarks, Standards and Evaluations CORAL benchmark Deep learning stack ROCm |
title | Comparative evaluation of deep learning workloads for leadership-class systems |
title_full | Comparative evaluation of deep learning workloads for leadership-class systems |
title_fullStr | Comparative evaluation of deep learning workloads for leadership-class systems |
title_full_unstemmed | Comparative evaluation of deep learning workloads for leadership-class systems |
title_short | Comparative evaluation of deep learning workloads for leadership-class systems |
title_sort | comparative evaluation of deep learning workloads for leadership class systems |
topic | CORAL benchmark Deep learning stack ROCm |
url | http://www.sciencedirect.com/science/article/pii/S2772485921000053 |
work_keys_str_mv | AT junqiyin comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems AT aristeidistsaris comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems AT sajaldash comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems AT rossmiller comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems AT feiyiwang comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems AT mallikarjunarjunshankar comparativeevaluationofdeeplearningworkloadsforleadershipclasssystems |