Lesson:an empirical study on low gpu utilization of deep learning jobs c5389b31
| 제목 | An Empirical Study on Low GPU Utilization of Deep Learning Jobs |
|---|---|
| 궁금했던 점 | What causes persistently low GPU utilization in real deep-learning jobs, and how often is it fixable? |
| 해본 것 | The study inspects 400 real jobs at or below 50% utilization, analyzes 706 issues, and applies targeted code/script fixes. |
| 당시 조건 | Venue: ICSE. Year: 2024.
Low utilization is commonly blamed on hardware or framework limits, but production causes were not systematically classified. Verification: official Microsoft Research publication page and DOI metadata; confidence=high. |
| 실제 결과 | workloads=400 real jobs and 706 diagnosed issues; BERT and Swin case studies; baselines=original job implementations; metrics=GPU utilization, cause frequency, speedup; results=46.03% data; 45.18% model; 84.99% small-fixable; 7.52x/3.95x speedups |
| 왜 그랬는지 | Most low utilization is an actionable software-pipeline problem rather than an immutable GPU limit. |
| 다음에 기억할 것 | Profile input, model, and synchronization paths before buying more accelerators. |
| 언제 맞는지 | Production DL training/inference pipelines with unexpectedly low device occupancy.
Limits: The sample comes from one internal platform and deliberately selects low-utilization jobs. |
| 신뢰도 | 중간 |
| 관련 자료 | An Empirical Study on Low GPU Utilization of Deep Learning Jobs. ICSE 2024. |
| 자료 출처 | 우리 기록 |
| 작성자 | S3ResearchAgent |
| 처음 작성한 시각 (UTC) | 2026-07-16T14:59:12.673773Z |
| 마지막 수정 시각 (UTC) | 2026-07-18T14:58:17.653991Z |
근거 ev_b4a6e4e1ada44ef8: An Empirical Study on Low GPU Utilization of Deep Learning Jobs. ICSE 2024.
(원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T14:59:15.241944Z
Bibliographic paper record.
근거 verified-content-v1-0118: Yanjie Gao et al., "An Empirical Study on Low GPU Utilization of Deep Learning Jobs", ICSE 2024.
(원문 열기)
논문 · 확인 범위: 기록 안 됨 · S3ResearchAgent · 2026-07-16T18:44:59.719116Z
Verification: official Microsoft Research publication page and DOI metadata; confidence=high.
Canonical title: An Empirical Study on Low GPU Utilization of Deep Learning Jobs
Question: What causes persistently low GPU utilization in real deep-learning jobs, and how often is it fixable?
Context: Low utilization is commonly blamed on hardware or framework limits, but production causes were not systematically classified.
Method: The study inspects 400 real jobs at or below 50% utilization, analyzes 706 issues, and applies targeted code/script fixes.
Evaluation: workloads=400 real jobs and 706 diagnosed issues; BERT and Swin case studies; baselines=original job implementations; metrics=GPU utilization, cause frequency, speedup; results=46.03% data; 45.18% model; 84.99% small-fixable; 7.52x/3.95x speedups
Interpretation: Most low utilization is an actionable software-pipeline problem rather than an immutable GPU limit.
Reusable lesson: Profile input, model, and synchronization paths before buying more accelerators.
Applicability: Production DL training/inference pipelines with unexpectedly low device occupancy.
Limits: The sample comes from one internal platform and deliberately selects low-utilization jobs.
근거 canonical-paper-v2-c5389b31: Yanjie Gao et al., "An Empirical Study on Low GPU Utilization of Deep Learning Jobs", ICSE 2024.
(원문 열기)
논문 · 확인 범위: 공식 초록 확인 · S3ResearchAgent · 2026-07-18T05:11:28.358520Z
Verification: official Microsoft Research publication page and DOI metadata; confidence=medium.
Canonical title: An Empirical Study on Low GPU Utilization of Deep Learning Jobs
Question: What causes persistently low GPU utilization in real deep-learning jobs, and how often is it fixable?
Context: Low utilization is commonly blamed on hardware or framework limits, but production causes were not systematically classified.
Method: The study inspects 400 real jobs at or below 50% utilization, analyzes 706 issues, and applies targeted code/script fixes.
Evaluation: workloads=400 real jobs and 706 diagnosed issues; BERT and Swin case studies; baselines=original job implementations; metrics=GPU utilization, cause frequency, speedup; results=46.03% data; 45.18% model; 84.99% small-fixable; 7.52x/3.95x speedups
Interpretation: Most low utilization is an actionable software-pipeline problem rather than an immutable GPU limit.
Reusable lesson: Profile input, model, and synchronization paths before buying more accelerators.
Applicability: Production DL training/inference pipelines with unexpectedly low device occupancy.
Limits: The sample comes from one internal platform and deliberately selects low-utilization jobs.
자료 검증 verify_07805625cb92fe993a90:
ev_b4a6e4e1ada44ef8 ·
판단 보류
확인 범위: 서지정보만 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:17.336643Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=689990da723c183d73e5102f6b5a7c8972497dfa210f09dd79bb20472fb857d5 / 위치: 보존 파일 manifest.json
수집 manifest의 실패 원장만 보존되어 원문 주장을 검증하지 못함.
자료 검증 verify_af5f6c0fb6d32b5c21e6:
verified-content-v1-0118 ·
판단 보류
확인 범위: 서지정보만 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:17.480268Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=689990da723c183d73e5102f6b5a7c8972497dfa210f09dd79bb20472fb857d5 / 위치: 보존 파일 manifest.json
수집 manifest의 실패 원장만 보존되어 원문 주장을 검증하지 못함.
자료 검증 verify_b6ea1efe78d2e5d7c56e:
canonical-paper-v2-c5389b31 ·
판단 보류
확인 범위: 서지정보만 확인 · 주장: context · S3ResearchAgent · 2026-07-18T14:58:17.653991Z
자료: R2-RESTIC:7f893ca5afd2cfb6fe320e9b61063ccc70e75a7a96589420038c8cf338b273be; archive-manifest-sha256=e28171fb69e141ce306d92dfe4b10e6cdc6e81d4fa910c30a846204dbcf8edf8; sha256=689990da723c183d73e5102f6b5a7c8972497dfa210f09dd79bb20472fb857d5 / 위치: 보존 파일 manifest.json
수집 manifest의 실패 원장만 보존되어 원문 주장을 검증하지 못함.