{"id":"https://openalex.org/W7133565664","doi":"https://doi.org/10.1109/hpca68181.2026.11408539","title":"AUM: Unleashing the Efficiency Potential of Shared Processors with Accelerator Units for LLM Serving","display_name":"AUM: Unleashing the Efficiency Potential of Shared Processors with Accelerator Units for LLM Serving","publication_year":2026,"publication_date":"2026-01-31","ids":{"openalex":"https://openalex.org/W7133565664","doi":"https://doi.org/10.1109/hpca68181.2026.11408539"},"language":null,"primary_location":{"id":"doi:10.1109/hpca68181.2026.11408539","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca68181.2026.11408539","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE International Symposium on High Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5077593888","display_name":"Xinkai Wang","orcid":"https://orcid.org/0009-0000-8447-4411"},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Xinkai Wang","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5007581833","display_name":"Chang Li","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chao Li","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115637304","display_name":"Yiming Zhuansun","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yiming Zhuansun","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128069215","display_name":"Jinyang Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jinyang Guo","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128099519","display_name":"Xiaofeng Hou","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaofeng Hou","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128114242","display_name":"Jing Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jing Wang","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123575537","display_name":"Luping Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Luping Wang","raw_affiliation_strings":["Alibaba Group"],"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028116620","display_name":"Weigao Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Weigao Chen","raw_affiliation_strings":["Alibaba Group"],"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5128062528","display_name":"Cheng Huang","orcid":null},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Cheng Huang","raw_affiliation_strings":["Alibaba Group"],"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073989304","display_name":"Guodong Yang","orcid":"https://orcid.org/0000-0003-1908-071X"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Guodong Yang","raw_affiliation_strings":["Alibaba Group"],"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5030649184","display_name":"L. Y. Zhang","orcid":"https://orcid.org/0009-0007-2876-3623"},"institutions":[{"id":"https://openalex.org/I4210095624","display_name":"Alibaba Group (United States)","ror":"https://ror.org/00rn0m335","country_code":"US","type":"company","lineage":["https://openalex.org/I4210095624","https://openalex.org/I45928872"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liping Zhang","raw_affiliation_strings":["Alibaba Group"],"affiliations":[{"raw_affiliation_string":"Alibaba Group","institution_ids":["https://openalex.org/I4210095624"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5120913770","display_name":"Minyi Guo","orcid":null},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Minyi Guo","raw_affiliation_strings":["School of Computer Science, Shanghai Jiao Tong University"],"affiliations":[{"raw_affiliation_string":"School of Computer Science, Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":12,"corresponding_author_ids":["https://openalex.org/A5077593888"],"corresponding_institution_ids":["https://openalex.org/I183067930"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.49174862,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"15"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.34279999136924744,"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"}},"topics":[{"id":"https://openalex.org/T11181","display_name":"Advanced Data Storage Technologies","score":0.34279999136924744,"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"}},{"id":"https://openalex.org/T10772","display_name":"Distributed systems and fault tolerance","score":0.16339999437332153,"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"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.05950000137090683,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.31470000743865967},{"id":"https://openalex.org/keywords/troubleshooting","display_name":"Troubleshooting","score":0.2939999997615814},{"id":"https://openalex.org/keywords/power","display_name":"Power (physics)","score":0.27250000834465027},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.2624000012874603},{"id":"https://openalex.org/keywords/software","display_name":"Software","score":0.2547999918460846}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.554099977016449},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3515999913215637},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.33000001311302185},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C147494362","wikidata":"https://www.wikidata.org/wiki/Q2078905","display_name":"Troubleshooting","level":2,"score":0.2939999997615814},{"id":"https://openalex.org/C163258240","wikidata":"https://www.wikidata.org/wiki/Q25342","display_name":"Power (physics)","level":2,"score":0.27250000834465027},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.2547999918460846},{"id":"https://openalex.org/C111919701","wikidata":"https://www.wikidata.org/wiki/Q9135","display_name":"Operating system","level":1,"score":0.25029999017715454},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.24899999797344208}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/hpca68181.2026.11408539","is_oa":false,"landing_page_url":"https://doi.org/10.1109/hpca68181.2026.11408539","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2026 IEEE International Symposium on High Performance Computer Architecture (HPCA)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Affordable and clean energy","score":0.830897331237793,"id":"https://metadata.un.org/sdg/7"}],"awards":[{"id":"https://openalex.org/G6615303026","display_name":null,"funder_award_id":"U23A6007,U24A20234","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":44,"referenced_works":["https://openalex.org/W1864199185","https://openalex.org/W1983096442","https://openalex.org/W2013062050","https://openalex.org/W2017004353","https://openalex.org/W2022820501","https://openalex.org/W2036895660","https://openalex.org/W2038387802","https://openalex.org/W2126364467","https://openalex.org/W2156077332","https://openalex.org/W2166713816","https://openalex.org/W2279541644","https://openalex.org/W2613264175","https://openalex.org/W2796645376","https://openalex.org/W2926143647","https://openalex.org/W2930660349","https://openalex.org/W2932629768","https://openalex.org/W3010640493","https://openalex.org/W3099214871","https://openalex.org/W3158545849","https://openalex.org/W4214512541","https://openalex.org/W4220980777","https://openalex.org/W4244254628","https://openalex.org/W4251038806","https://openalex.org/W4283380581","https://openalex.org/W4285345683","https://openalex.org/W4285609622","https://openalex.org/W4288658754","https://openalex.org/W4387321091","https://openalex.org/W4388080974","https://openalex.org/W4390712964","https://openalex.org/W4394998510","https://openalex.org/W4395106452","https://openalex.org/W4398774961","https://openalex.org/W4401211672","https://openalex.org/W4401211704","https://openalex.org/W4404954356","https://openalex.org/W4406014788","https://openalex.org/W4406266041","https://openalex.org/W4406658017","https://openalex.org/W4408893327","https://openalex.org/W4411486062","https://openalex.org/W4411486391","https://openalex.org/W4414736068","https://openalex.org/W4415792994"],"related_works":[],"abstract_inverted_index":{"Generative":[0],"AI,":[1],"especially":[2],"LLM,":[3],"is":[4,78],"driving":[5],"a":[6,140,178],"fundamental":[7],"shift":[8],"in":[9,37,122,132],"software":[10],"paradigms,":[11],"prompting":[12],"cloud":[13],"providers":[14],"to":[15,40,58,62,80,115,129,146,185],"build":[16],"more":[17],"efficient":[18],"serving":[19,60],"infrastructures.":[20],"To":[21,69],"meet":[22],"the":[23,38,151,171,182],"computational":[24],"demands":[25],"of":[26,127,153],"emerging":[27],"software,":[28],"modern":[29],"CPU":[30,56,74,209],"processors":[31],"are":[32,113],"integrating":[33],"Accelerator":[34,117],"Units":[35],"(AU)":[36],"pipeline":[39],"accelerate":[41],"key":[42],"operations,":[43],"such":[44],"as":[45],"Intel":[46],"AMX":[47],"for":[48,164],"matrix":[49],"multiplication.":[50],"Current":[51],"practices":[52],"that":[53,206],"dedicate":[54],"AU-enabled":[55,73,202],"exclusively":[57],"LLM":[59],"lead":[61],"significant":[63],"resource":[64,108,111,143,175,195,236],"waste":[65],"and":[66,84,106,124,149,174,192],"inferior":[67],"efficiency.":[68],"this":[70,136],"end,":[71],"sharing":[72,90],"with":[75,161,234],"general":[76],"workloads":[77],"necessary":[79],"harvest":[81],"redundant":[82],"resources":[83],"improve":[85],"platform":[86],"performance-per-watt.":[87],"However,":[88],"perfectly":[89],"AU":[91,219],"can":[92],"be":[93],"challenging":[94],"since":[95],"they":[96],"introduce":[97],"three-dimensional":[98,165],"variations:":[99],"variable":[100],"usage":[101],"patterns,":[102],"compulsory":[103],"frequency":[104],"interferences,":[105],"dissimilar":[107],"bounds.":[109],"Existing":[110],"managers":[112],"oblivious":[114],"complex":[116],"Unit":[118],"Variations":[119],"(AUV),":[120],"resulting":[121],"performance":[123],"efficiency":[125,152,210],"degradations":[126],"up":[128],"50":[130],"%":[131],"shared":[133,154],"environments.":[134],"Therefore,":[135],"paper":[137],"introduces":[138],"AUM,":[139],"novel":[141],"AU-aware":[142],"manager":[144],"designed":[145],"handle":[147],"AUV":[148],"maximize":[150],"processors.":[155],"AUM":[156,207],"has":[157],"two":[158],"cooperative":[159],"components":[160],"three":[162],"stages":[163],"AUV.":[166],"The":[167],"background":[168],"profiler":[169],"characterizes":[170],"usage,":[172],"frequency,":[173],"information":[176],"into":[177],"discrete":[179],"model,":[180],"guiding":[181],"runtime":[183],"controller":[184],"analyze":[186],"usage-aware":[187],"requirements,":[188],"select":[189],"frequency-aware":[190],"divisions,":[191],"make":[193],"bound-aware":[194],"decisions.":[196],"Through":[197],"extensive":[198],"evaluations":[199],"on":[200],"production":[201],"CPUs,":[203],"we":[204],"show":[205],"improves":[208],"by":[211,221,225],"<tex":[212,226],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[213,227],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$4.7-8.8":[214],"\\%$</tex>":[215],"while":[216],"maintaining":[217],"high-performance":[218],"applications":[220],"reducing":[222],"SLO":[223],"violations":[224],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">$\\mathbf{7":[228],"-":[229],"1":[230,231],"\\%}$</tex>":[232],"compared":[233],"state-of-the-art":[235],"managers.":[237]},"counts_by_year":[],"updated_date":"2026-04-09T08:11:56.329763","created_date":"2026-03-05T00:00:00"}
