{"links":{},"id":2001773,"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:02001773","sets":["1164:2240:11899:1745223226515"]},"path":["1745223226515"],"owner":"80578","recid":"2001773","title":["SQUID GPUノード上でのMLPerf HPCを用いた分散深層学習の性能評価"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2025-05-05"},"_buckets":{"deposit":"141dcde0-f4a6-48de-8de1-df46fe19ccae"},"_deposit":{"id":"2001773","pid":{"type":"depid","value":"2001773","revision_id":0},"owners":[80578],"status":"published","created_by":80578},"item_title":"SQUID GPUノード上でのMLPerf HPCを用いた分散深層学習の性能評価","author_link":[],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"SQUID GPUノード上でのMLPerf HPCを用いた分散深層学習の性能評価","subitem_title_language":"ja"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"GPUの性能評価","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2025-05-05","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"大阪大学大学院情報科学研究科"},{"subitem_text_value":"大阪大学D3センター"},{"subitem_text_value":"大阪大学D3センター"},{"subitem_text_value":"大阪大学D3センター"},{"subitem_text_value":"奈良先端科学技術大学院大学総合情報基盤センター"},{"subitem_text_value":"大阪大学D3センター"}]},"item_4_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Graduate School of Information Science and Technology, The University of Osaka","subitem_text_language":"en"},{"subitem_text_value":"D3 Center, The University of Osaka","subitem_text_language":"en"},{"subitem_text_value":"D3 Center, The University of Osaka","subitem_text_language":"en"},{"subitem_text_value":"D3 Center, The University of Osaka","subitem_text_language":"en"},{"subitem_text_value":"Information Initiative Center, Nara Institute of Science and Technology","subitem_text_language":"en"},{"subitem_text_value":"D3 Center, The University of Osaka","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/2001773/files/IPSJ-HPC25199010.pdf","label":"IPSJ-HPC25199010.pdf"},"date":[{"dateType":"Available","dateValue":"2027-05-05"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-HPC25199010.pdf","filesize":[{"value":"4.5 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"14"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"e190e4c0-5d1d-4c95-bfd7-411968b252c6","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2025 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"東郷,凜太朗"}]},{"creatorNames":[{"creatorName":"高橋,慧智"}]},{"creatorNames":[{"creatorName":"速水,智教"}]},{"creatorNames":[{"creatorName":"曽我,隆"}]},{"creatorNames":[{"creatorName":"遠藤,新"}]},{"creatorNames":[{"creatorName":"伊達,進"}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10463942","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"2188-8841","subitem_source_identifier_type":"ISSN"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"テラバイト級のデータセットが用いられる深層学習の計算量は膨大であるため，HPCシステムの必要性が高まっている．大阪大学のHPCシステムSQUIDにおいても分散深層学習の需要は増加傾向にあるが，SQUIDの分散深層学習に対する性能分析は不十分である．本研究では，分散深層学習のベンチマーク集であるMLPerf HPCに含まれるCosmoFlowベンチマークを用い，SQUIDのGPUノードの分散深層学習における性能を評価する．SQUID GPUノード1台を使用してCosmoFlowを実行し，プロファイルを分析した結果，データセットの読み込みがボトルネックとなっていることが明らかになった．読み込み性能を改善するためにデータセットの配置先やデータ入力パイプラインの修正を実施したところ，訓練スループットが6.77倍に向上することを確認した．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"10","bibliographic_titles":[{"bibliographic_title":"研究報告ハイパフォーマンスコンピューティング（HPC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2025-05-05","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"10","bibliographicVolumeNumber":"2025-HPC-199"}]},"relation_version_is_last":true,"weko_creator_id":"80578"},"created":"2025-04-21T08:39:49.833182+00:00","updated":"2025-04-21T08:39:53.914845+00:00"}