{"id":"https://openalex.org/W7168040109","doi":"https://doi.org/10.1145/3806645.3807596","title":"Scaling Attention Beyond GPUs for LLM Inference","display_name":"Scaling Attention Beyond GPUs for LLM Inference","publication_year":2026,"publication_date":"2026-07-11","ids":{"openalex":"https://openalex.org/W7168040109","doi":"https://doi.org/10.1145/3806645.3807596"},"language":null,"primary_location":{"id":"doi:10.1145/3806645.3807596","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3806645.3807596","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 35th International Symposium on High-Performance Parallel and Distributed Computing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3806645.3807596","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078419196","display_name":"Weishu Deng","orcid":null},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Weishu Deng","raw_affiliation_strings":["The University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0000-6550-7484","affiliations":[{"raw_affiliation_string":"The University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140485105","display_name":"Yujie Yang","orcid":"https://orcid.org/0009-0000-9087-9693"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yujie Yang","raw_affiliation_strings":["The University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0000-9087-9693","affiliations":[{"raw_affiliation_string":"The University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113139833","display_name":"P. Le Du","orcid":"https://orcid.org/0009-0008-8731-3478"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Peiran Du","raw_affiliation_strings":["The University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0008-8731-3478","affiliations":[{"raw_affiliation_string":"The University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5140469754","display_name":"Lingfeng Xiang","orcid":"https://orcid.org/0009-0009-7303-3595"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lingfeng Xiang","raw_affiliation_strings":["University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0009-7303-3595","affiliations":[{"raw_affiliation_string":"University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101633004","display_name":"Zhen Lin","orcid":"https://orcid.org/0009-0004-3653-7173"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhen Lin","raw_affiliation_strings":["University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0004-3653-7173","affiliations":[{"raw_affiliation_string":"University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100430412","display_name":"Zhong Chen","orcid":"https://orcid.org/0000-0002-4098-6260"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen Zhong","raw_affiliation_strings":["University of texas at arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0000-0002-4098-6260","affiliations":[{"raw_affiliation_string":"University of texas at arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076280323","display_name":"Faraz Ahmed","orcid":"https://orcid.org/0000-0003-0688-6394"},"institutions":[{"id":"https://openalex.org/I1324840837","display_name":"Hewlett-Packard (United States)","ror":"https://ror.org/059rn9488","country_code":"US","type":"company","lineage":["https://openalex.org/I1324840837"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Faraz Ahmed","raw_affiliation_strings":["Hewlett Packard Labs, Milpitas, USA"],"raw_orcid":"https://orcid.org/0000-0003-0688-6394","affiliations":[{"raw_affiliation_string":"Hewlett Packard Labs, Milpitas, USA","institution_ids":["https://openalex.org/I1324840837"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082986219","display_name":"Lianjie Cao","orcid":"https://orcid.org/0000-0003-3408-9050"},"institutions":[{"id":"https://openalex.org/I4210122178","display_name":"Hewlett Packard Enterprise (United States)","ror":"https://ror.org/020x0c621","country_code":"US","type":"company","lineage":["https://openalex.org/I4210122178"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Lianjie Cao","raw_affiliation_strings":["Hewlett Packard Enterprise, Milpitas, USA"],"raw_orcid":"https://orcid.org/0000-0003-3408-9050","affiliations":[{"raw_affiliation_string":"Hewlett Packard Enterprise, Milpitas, USA","institution_ids":["https://openalex.org/I4210122178"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101915392","display_name":"Puneet Sharma","orcid":"https://orcid.org/0000-0003-4594-8164"},"institutions":[{"id":"https://openalex.org/I1324840837","display_name":"Hewlett-Packard (United States)","ror":"https://ror.org/059rn9488","country_code":"US","type":"company","lineage":["https://openalex.org/I1324840837"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Puneet Sharma","raw_affiliation_strings":["Hewlett Packard Labs, Milpitas, USA"],"raw_orcid":"https://orcid.org/0000-0003-4594-8164","affiliations":[{"raw_affiliation_string":"Hewlett Packard Labs, Milpitas, USA","institution_ids":["https://openalex.org/I1324840837"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101421445","display_name":"Song Jiang","orcid":"https://orcid.org/0000-0002-1681-9008"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Song Jiang","raw_affiliation_strings":["University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0000-0002-1681-9008","affiliations":[{"raw_affiliation_string":"University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101486501","display_name":"Hui Lu","orcid":"https://orcid.org/0009-0006-5062-4710"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hui Lu","raw_affiliation_strings":["The University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0009-0006-5062-4710","affiliations":[{"raw_affiliation_string":"The University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5073224556","display_name":"Jia Rao","orcid":"https://orcid.org/0000-0002-2133-4363"},"institutions":[{"id":"https://openalex.org/I189196454","display_name":"The University of Texas at Arlington","ror":"https://ror.org/019kgqr73","country_code":"US","type":"education","lineage":["https://openalex.org/I189196454"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jia Rao","raw_affiliation_strings":["The University of Texas at Arlington, Arlington, USA"],"raw_orcid":"https://orcid.org/0000-0002-2133-4363","affiliations":[{"raw_affiliation_string":"The University of Texas at Arlington, Arlington, USA","institution_ids":["https://openalex.org/I189196454"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"445","last_page":"456"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.2872999906539917,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.2872999906539917,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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/T10028","display_name":"Topic Modeling","score":0.10459999740123749,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"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":0.09610000252723694,"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/scalability","display_name":"Scalability","score":0.6777999997138977},{"id":"https://openalex.org/keywords/cache","display_name":"Cache","score":0.4964999854564667},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.48730000853538513},{"id":"https://openalex.org/keywords/pci-express","display_name":"PCI Express","score":0.48089998960494995},{"id":"https://openalex.org/keywords/central-processing-unit","display_name":"Central processing unit","score":0.4277999997138977},{"id":"https://openalex.org/keywords/decoding-methods","display_name":"Decoding methods","score":0.4047999978065491},{"id":"https://openalex.org/keywords/workflow","display_name":"Workflow","score":0.3977999985218048},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.39399999380111694},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.37119999527931213}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.864799976348877},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.6777999997138977},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.619700014591217},{"id":"https://openalex.org/C115537543","wikidata":"https://www.wikidata.org/wiki/Q165596","display_name":"Cache","level":2,"score":0.4964999854564667},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.48730000853538513},{"id":"https://openalex.org/C64270927","wikidata":"https://www.wikidata.org/wiki/Q206924","display_name":"PCI Express","level":3,"score":0.48089998960494995},{"id":"https://openalex.org/C49154492","wikidata":"https://www.wikidata.org/wiki/Q5300","display_name":"Central processing unit","level":2,"score":0.4277999997138977},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.4047999978065491},{"id":"https://openalex.org/C177212765","wikidata":"https://www.wikidata.org/wiki/Q627335","display_name":"Workflow","level":2,"score":0.3977999985218048},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.39399999380111694},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.37119999527931213},{"id":"https://openalex.org/C189783530","wikidata":"https://www.wikidata.org/wiki/Q352090","display_name":"CPU cache","level":3,"score":0.34610000252723694},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.3224000036716461},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.29980000853538513},{"id":"https://openalex.org/C111030470","wikidata":"https://www.wikidata.org/wiki/Q1430460","display_name":"Curse of dimensionality","level":2,"score":0.296999990940094},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.2897000014781952},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.288100004196167},{"id":"https://openalex.org/C126831891","wikidata":"https://www.wikidata.org/wiki/Q221673","display_name":"Host (biology)","level":2,"score":0.2766999900341034},{"id":"https://openalex.org/C186967261","wikidata":"https://www.wikidata.org/wiki/Q5082128","display_name":"Mobile device","level":2,"score":0.27399998903274536},{"id":"https://openalex.org/C78766204","wikidata":"https://www.wikidata.org/wiki/Q555032","display_name":"Multi-core processor","level":2,"score":0.2727000117301941},{"id":"https://openalex.org/C2781335571","wikidata":"https://www.wikidata.org/wiki/Q2633544","display_name":"GPU cluster","level":3,"score":0.2685999870300293},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.26829999685287476},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.2667999863624573},{"id":"https://openalex.org/C1665295","wikidata":"https://www.wikidata.org/wiki/Q6887219","display_name":"Mobile processor","level":5,"score":0.26669999957084656},{"id":"https://openalex.org/C2778119891","wikidata":"https://www.wikidata.org/wiki/Q477690","display_name":"CUDA","level":2,"score":0.2639000117778778},{"id":"https://openalex.org/C141917322","wikidata":"https://www.wikidata.org/wiki/Q1025017","display_name":"Cache coherence","level":5,"score":0.25450000166893005},{"id":"https://openalex.org/C177950962","wikidata":"https://www.wikidata.org/wiki/Q10997658","display_name":"Non-volatile memory","level":2,"score":0.25279998779296875}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3806645.3807596","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3806645.3807596","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 35th International Symposium on High-Performance Parallel and Distributed Computing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3806645.3807596","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3806645.3807596","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 35th International Symposium on High-Performance Parallel and Distributed Computing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":10,"referenced_works":["https://openalex.org/W3098136301","https://openalex.org/W4285225959","https://openalex.org/W4286001027","https://openalex.org/W4387321091","https://openalex.org/W4400909494","https://openalex.org/W4414198736","https://openalex.org/W7133198832","https://openalex.org/W7133224090","https://openalex.org/W7133224126","https://openalex.org/W7133227460"],"related_works":[],"abstract_inverted_index":{"Scaling":[0],"inference":[1],"for":[2,23,74,118],"large":[3],"language":[4],"models":[5,185],"is":[6],"increasingly":[7],"constrained":[8],"by":[9,54,168],"limited":[10],"GPU":[11,131,173,213],"memory,":[12,38],"primarily":[13],"due":[14],"to":[15,43,66,99],"the":[16,30,34,55,158,169],"expanding":[17],"intermediate":[18],"states":[19],"(KV":[20],"caches)":[21],"required":[22],"long-context":[24],"generation":[25],"and":[26,46,69,82,102,172,186,196,200,209],"multi-user":[27],"workloads.":[28],"Once":[29],"KV":[31,64,105,127,142,163],"cache":[32],"exceeds":[33],"capacity":[35],"of":[36],"high-bandwidth":[37],"it":[39],"must":[40],"be":[41],"offloaded":[42],"host":[44],"memory":[45,68,132,174],"reloaded":[47],"on":[48,139,211],"demand,":[49],"a":[50,90,95,113,154],"workflow":[51],"severely":[52],"bottlenecked":[53],"CPU\u2013GPU":[56,115],"interconnect,":[57],"typically":[58],"PCIe.":[59],"Existing":[60],"approaches":[61],"exploiting":[62],"offload":[63],"caches":[65,164],"CPU":[67,79,146,171],"selectively":[70,100],"reload":[71],"partial":[72],"segments":[73],"attention":[75,116,124,138,204],"computation":[76],"often":[77],"underutilize":[78],"compute":[80],"resources":[81],"suffer":[83],"from":[84],"accuracy":[85],"degradation.":[86],"We":[87],"present":[88],"Beyond,":[89],"drop-in":[91],"runtime":[92],"that":[93,189],"integrates":[94],"smart":[96],"offloading":[97],"scheme":[98],"identify":[101],"retain":[103],"salient":[104,140],"entries":[106,128,143],"across":[107,183],"continuous":[108],"decoding":[109,160],"sessions,":[110],"together":[111],"with":[112,176],"hybrid":[114],"mechanism":[117],"scalable":[119],"inference.":[120],"Beyond":[121,190],"executes":[122],"dense":[123],"over":[125],"recent":[126],"stored":[129],"in":[130,145,206],"while":[133],"performing":[134],"parallel,":[135],"per-head":[136],"sparse":[137,203],"contextual":[141],"residing":[144],"memory.":[147],"The":[148],"outputs":[149],"are":[150,165],"fused":[151],"efficiently":[152],"through":[153],"log-sum-exp":[155],"scheme.":[156],"During":[157],"bandwidth-constrained":[159],"phase,":[161],"oversized":[162],"processed":[166],"cooperatively":[167],"aggregated":[170],"bandwidth,":[175],"only":[177],"minimal":[178],"PCIe":[179],"data":[180],"movement.":[181],"Experiments":[182],"diverse":[184],"workloads":[187],"demonstrate":[188],"improves":[191],"scalability,":[192],"supports":[193],"longer":[194],"sequences":[195],"larger":[197],"batch":[198],"sizes,":[199],"outperforms":[201],"existing":[202],"baselines":[205],"both":[207],"efficiency":[208],"accuracy\u2014all":[210],"commodity":[212],"hardware.":[214]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-07-12T00:00:00"}