{"id":"https://openalex.org/W4401211883","doi":"https://doi.org/10.1109/isca59077.2024.00064","title":"MAD-Max Beyond Single-Node: Enabling Large Machine Learning Model Acceleration on Distributed Systems","display_name":"MAD-Max Beyond Single-Node: Enabling Large Machine Learning Model Acceleration on Distributed Systems","publication_year":2024,"publication_date":"2024-06-29","ids":{"openalex":"https://openalex.org/W4401211883","doi":"https://doi.org/10.1109/isca59077.2024.00064"},"language":"en","primary_location":{"id":"doi:10.1109/isca59077.2024.00064","doi":"https://doi.org/10.1109/isca59077.2024.00064","is_oa":false,"landing_page_url":"https://doi.org/10.1109/isca59077.2024.00064","pdf_url":null,"source":{"id":"https://openalex.org/S4393916863","display_name":"Proceedings - International Symposium on Computer Architecture/Proceedings","issn_l":"1063-6897","issn":["1063-6897"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2024 ACM/IEEE 51st Annual International Symposium on Computer Architecture (ISCA)","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2310.02784","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5047381366","display_name":"Samuel Hsia","orcid":"https://orcid.org/0000-0003-4053-2100","observed_orcids":["https://orcid.org/0000-0003-4053-2100"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Samuel Hsia","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104179257","display_name":"Alicia Golden","orcid":"https://orcid.org/0009-0009-5246-859X","observed_orcids":["https://orcid.org/0009-0009-5246-859X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alicia Golden","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005070359","display_name":"Bilge Acun","orcid":"https://orcid.org/0000-0002-9233-3018","observed_orcids":["https://orcid.org/0000-0002-9233-3018"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bilge Acun","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5031627270","display_name":"Newsha Ardalani","orcid":"https://orcid.org/0000-0002-9975-4819","observed_orcids":["https://orcid.org/0000-0002-9975-4819"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Newsha Ardalani","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008830900","display_name":"Zachary DeVito","orcid":"https://orcid.org/0009-0002-8863-1503","observed_orcids":["https://orcid.org/0009-0002-8863-1503"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zachary DeVito","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043327132","display_name":"Gu-Yeon Wei","orcid":"https://orcid.org/0000-0001-5730-9904","observed_orcids":["https://orcid.org/0000-0001-5730-9904"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gu-Yeon Wei","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5026496503","display_name":"David M. Brooks","orcid":"https://orcid.org/0000-0002-0662-7889","observed_orcids":["https://orcid.org/0000-0002-0662-7889"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"David Brooks","raw_affiliation_strings":["Harvard University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard University","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5028220093","display_name":"Carole-Jean Wu","orcid":"https://orcid.org/0000-0002-9032-7239","observed_orcids":["https://orcid.org/0000-0002-9032-7239"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Carole-Jean Wu","raw_affiliation_strings":["FAIR at Meta"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"FAIR at Meta","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":2.4239,"has_fulltext":true,"cited_by_count":8,"citation_normalized_percentile":{"value":0.90834924,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"818","last_page":"833"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.98580002784729,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.98580002784729,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10101","display_name":"Cloud Computing and Resource Management","score":0.005799999926239252,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10715","display_name":"Distributed and Parallel Computing Systems","score":0.002099999925121665,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/mad-max","display_name":"Mad Max","score":0.8930000066757202},{"id":"https://openalex.org/keywords/large-language-model-training","display_name":"large language model training","score":0.7910000085830688},{"id":"https://openalex.org/keywords/distributed-training","display_name":"distributed training","score":0.5770000219345093},{"id":"https://openalex.org/keywords/gpu-clusters","display_name":"GPU clusters","score":0.6949999928474426},{"id":"https://openalex.org/keywords/communication-delay","display_name":"communication delay","score":0.718999981880188},{"id":"https://openalex.org/keywords/performance-modeling","display_name":"performance modeling","score":0.8040000200271606},{"id":"https://openalex.org/keywords/parallelization-strategies","display_name":"parallelization strategies","score":0.7630000114440918},{"id":"https://openalex.org/keywords/hardware-software-co-design","display_name":"hardware-software co-design","score":0.9929999709129333},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.5490000247955322}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7423093914985657},{"id":"https://openalex.org/C117896860","wikidata":"https://www.wikidata.org/wiki/Q11376","display_name":"Acceleration","level":2,"score":0.7278921604156494},{"id":"https://openalex.org/C62611344","wikidata":"https://www.wikidata.org/wiki/Q1062658","display_name":"Node (physics)","level":2,"score":0.5895571708679199},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.48620954155921936},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.09368002414703369},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C74650414","wikidata":"https://www.wikidata.org/wiki/Q11397","display_name":"Classical mechanics","level":1,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.1109/isca59077.2024.00064","doi":"https://doi.org/10.1109/isca59077.2024.00064","is_oa":false,"landing_page_url":"https://doi.org/10.1109/isca59077.2024.00064","pdf_url":null,"source":{"id":"https://openalex.org/S4393916863","display_name":"Proceedings - International Symposium on Computer Architecture/Proceedings","issn_l":"1063-6897","issn":["1063-6897"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2024 ACM/IEEE 51st Annual International Symposium on Computer Architecture (ISCA)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2310.02784","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.02784","pdf_url":"https://arxiv.org/pdf/2310.02784","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2310.02784","doi":"https://doi.org/10.48550/arxiv.2310.02784","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2310.02784","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2310.02784","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.02784","pdf_url":"https://arxiv.org/pdf/2310.02784","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4401211883.pdf","grobid_xml":"https://content.openalex.org/works/W4401211883.grobid-xml"},"referenced_works_count":62,"referenced_works":["https://openalex.org/W95608104","https://openalex.org/W1603598191","https://openalex.org/W2113235308","https://openalex.org/W2475334473","https://openalex.org/W2605350416","https://openalex.org/W2606722458","https://openalex.org/W2626721309","https://openalex.org/W2626778328","https://openalex.org/W2723293840","https://openalex.org/W2735029112","https://openalex.org/W2794670651","https://openalex.org/W2884700152","https://openalex.org/W2910100551","https://openalex.org/W2947737663","https://openalex.org/W2962745591","https://openalex.org/W2964182926","https://openalex.org/W2973171206","https://openalex.org/W2973172293","https://openalex.org/W2973727699","https://openalex.org/W2979719709","https://openalex.org/W2986124642","https://openalex.org/W2994850640","https://openalex.org/W3016842236","https://openalex.org/W3030163527","https://openalex.org/W3036878841","https://openalex.org/W3040573126","https://openalex.org/W3043023836","https://openalex.org/W3043433718","https://openalex.org/W3092558320","https://openalex.org/W3093945404","https://openalex.org/W3096425133","https://openalex.org/W3129831491","https://openalex.org/W3155243801","https://openalex.org/W3158146252","https://openalex.org/W3163465001","https://openalex.org/W3190062760","https://openalex.org/W3197720002","https://openalex.org/W4200634402","https://openalex.org/W4214658871","https://openalex.org/W4226343359","https://openalex.org/W4226479682","https://openalex.org/W4229005866","https://openalex.org/W4235645210","https://openalex.org/W4281790033","https://openalex.org/W4295680057","https://openalex.org/W4300973219","https://openalex.org/W4308760184","https://openalex.org/W4310557597","https://openalex.org/W4322718191","https://openalex.org/W4327930478","https://openalex.org/W4380874786","https://openalex.org/W4380881128","https://openalex.org/W4380881154","https://openalex.org/W4381886086","https://openalex.org/W4384918448","https://openalex.org/W4385567093","https://openalex.org/W4386768656","https://openalex.org/W4387951242","https://openalex.org/W4390051523","https://openalex.org/W4399809828","https://openalex.org/W7106552862","https://openalex.org/W7133225605"],"related_works":["https://openalex.org/W7203703145","https://openalex.org/W3153553004","https://openalex.org/W4404648787","https://openalex.org/W7139041841","https://openalex.org/W3187255235","https://openalex.org/W7162461522","https://openalex.org/W4415308595","https://openalex.org/W4412626539","https://openalex.org/W4229054011","https://openalex.org/W7131389093"],"abstract_inverted_index":{"Training":[0],"and":[1,14,51,72,106],"deploying":[2],"large-scale":[3,87],"machine":[4],"learning":[5],"models":[6,89],"is":[7,66],"time-consuming,":[8],"requires":[9],"significant":[10],"distributed":[11],"computing":[12],"infrastructures,":[13,29],"incurs":[15],"high":[16],"operational":[17],"costs.":[18],"Our":[19],"analysis,":[20],"grounded":[21],"in":[22],"real-world":[23,86],"large":[24],"model":[25],"training":[26],"on":[27,39,90],"datacenter-scale":[28],"reveals":[30],"that":[31],"14~32%":[32],"of":[33,80,85,99],"all":[34],"GPU":[35,92],"hours":[36],"are":[37],"spent":[38],"communication":[40,49],"with":[41],"no":[42],"overlapping":[43],"computation.":[44],"To":[45],"minimize":[46],"this":[47],"outstanding":[48],"latency":[50],"other":[52],"inherent":[53],"at-scale":[54],"inefficiencies,":[55],"we":[56,94],"introduce":[57],"an":[58],"agile":[59],"performance":[60],"modeling":[61],"framework,":[62],"MAD-Max.":[63],"This":[64],"framework":[65],"designed":[67],"to":[68,82,101,108],"optimize":[69],"parallelization":[70],"strategies":[71],"facilitate":[73],"hardware-software":[74],"co-design":[75],"opportunities.":[76],"Through":[77],"the":[78],"application":[79],"MAD-Max":[81],"a":[83],"suite":[84],"ML":[88],"state-of-the-art":[91],"clusters,":[93],"showcase":[95],"potential":[96],"throughput":[97],"enhancements":[98],"up":[100,107],"2.24":[102],"\u00d7":[103,110],"for":[104,111],"pretraining":[105],"5.27":[109],"inference":[112],"scenarios,":[113],"respectively.":[114]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":5}],"updated_date":"2026-10-30T12:39:25.321706","created_date":"2024-08-02T00:00:00"}