{"id":"https://openalex.org/W4417403290","doi":"https://doi.org/10.1109/pact65351.2025.00013","title":"Scalable Processing-Near-Memory for 1M-Token LLM Inference: CXL-Enabled KV-Cache Management Beyond GPU Limits","display_name":"Scalable Processing-Near-Memory for 1M-Token LLM Inference: CXL-Enabled KV-Cache Management Beyond GPU Limits","publication_year":2025,"publication_date":"2025-11-03","ids":{"openalex":"https://openalex.org/W4417403290","doi":"https://doi.org/10.1109/pact65351.2025.00013"},"language":null,"primary_location":{"id":"doi:10.1109/pact65351.2025.00013","is_oa":false,"landing_page_url":"https://doi.org/10.1109/pact65351.2025.00013","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 34th International Conference on Parallel Architectures and Compilation Techniques (PACT)","raw_type":"proceedings-article"},"type":"conference-paper","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/A5101642625","display_name":"Dowon Kim","orcid":"https://orcid.org/0009-0007-0870-5418"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Dowon Kim","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100449728","display_name":"Minjae Lee","orcid":"https://orcid.org/0000-0002-1600-5557"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"MinJae Lee","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110910238","display_name":"J.Y. Kim","orcid":null},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Janghyeon Kim","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5056678587","display_name":"Hyuck M. Kwon","orcid":"https://orcid.org/0000-0002-4077-142X"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"HyuckSung Kwon","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037600962","display_name":"Hyeonggyu Jeong","orcid":null},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Hyeonggyu Jeong","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101727465","display_name":"Sangsoo Park","orcid":"https://orcid.org/0000-0002-9617-6312"},"institutions":[{"id":"https://openalex.org/I2250650973","display_name":"Samsung (South Korea)","ror":"https://ror.org/04w3jy968","country_code":"KR","type":"company","lineage":["https://openalex.org/I2250650973"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Sang-Soo Park","raw_affiliation_strings":["Samsung Electronics,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Electronics,Republic of Korea","institution_ids":["https://openalex.org/I2250650973"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5064813707","display_name":"Minyong Yoon","orcid":"https://orcid.org/0009-0003-0529-8184"},"institutions":[{"id":"https://openalex.org/I2250650973","display_name":"Samsung (South Korea)","ror":"https://ror.org/04w3jy968","country_code":"KR","type":"company","lineage":["https://openalex.org/I2250650973"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Minyong Yoon","raw_affiliation_strings":["Samsung Electronics,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Electronics,Republic of Korea","institution_ids":["https://openalex.org/I2250650973"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062044366","display_name":"Si-Dong Roh","orcid":"https://orcid.org/0000-0001-5961-948X"},"institutions":[{"id":"https://openalex.org/I2250650973","display_name":"Samsung (South Korea)","ror":"https://ror.org/04w3jy968","country_code":"KR","type":"company","lineage":["https://openalex.org/I2250650973"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Si-Dong Roh","raw_affiliation_strings":["Samsung Electronics,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Electronics,Republic of Korea","institution_ids":["https://openalex.org/I2250650973"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5089034192","display_name":"Yongsuk Kwon","orcid":"https://orcid.org/0000-0002-1956-4629"},"institutions":[{"id":"https://openalex.org/I2250650973","display_name":"Samsung (South Korea)","ror":"https://ror.org/04w3jy968","country_code":"KR","type":"company","lineage":["https://openalex.org/I2250650973"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Yongsuk Kwon","raw_affiliation_strings":["Samsung Electronics,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Electronics,Republic of Korea","institution_ids":["https://openalex.org/I2250650973"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081625166","display_name":"Jinin So","orcid":"https://orcid.org/0000-0002-7569-3505"},"institutions":[{"id":"https://openalex.org/I2250650973","display_name":"Samsung (South Korea)","ror":"https://ror.org/04w3jy968","country_code":"KR","type":"company","lineage":["https://openalex.org/I2250650973"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jinin So","raw_affiliation_strings":["Samsung Electronics,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Samsung Electronics,Republic of Korea","institution_ids":["https://openalex.org/I2250650973"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101935712","display_name":"Jungwook Choi","orcid":"https://orcid.org/0000-0002-5691-4771"},"institutions":[{"id":"https://openalex.org/I4575257","display_name":"Hanyang University","ror":"https://ror.org/046865y68","country_code":"KR","type":"education","lineage":["https://openalex.org/I4575257"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Jungwook Choi","raw_affiliation_strings":["Hanyang University,Republic of Korea"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanyang University,Republic of Korea","institution_ids":["https://openalex.org/I4575257"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.37284873,"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":"13"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.2727000117301941,"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":0.2727000117301941,"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":0.2603999972343445,"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/T14347","display_name":"Big Data and Digital Economy","score":0.11050000041723251,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.8266000151634216},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6255999803543091},{"id":"https://openalex.org/keywords/efficient-energy-use","display_name":"Efficient energy use","score":0.5347999930381775},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.49959999322891235},{"id":"https://openalex.org/keywords/memory-management","display_name":"Memory management","score":0.487199991941452},{"id":"https://openalex.org/keywords/context-switch","display_name":"Context switch","score":0.4724999964237213},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.47099998593330383},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.447299987077713}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.871399998664856},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.8266000151634216},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6255999803543091},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.5347999930381775},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5083000063896179},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.49959999322891235},{"id":"https://openalex.org/C176649486","wikidata":"https://www.wikidata.org/wiki/Q2308807","display_name":"Memory management","level":3,"score":0.487199991941452},{"id":"https://openalex.org/C53833338","wikidata":"https://www.wikidata.org/wiki/Q1061424","display_name":"Context switch","level":2,"score":0.4724999964237213},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.47099998593330383},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.447299987077713},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.44690001010894775},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.39399999380111694},{"id":"https://openalex.org/C2780165032","wikidata":"https://www.wikidata.org/wiki/Q16869822","display_name":"Energy consumption","level":2,"score":0.36809998750686646},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3677000105381012},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.35839998722076416},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.3521000146865845},{"id":"https://openalex.org/C50630238","wikidata":"https://www.wikidata.org/wiki/Q971505","display_name":"General-purpose computing on graphics processing units","level":3,"score":0.31450000405311584},{"id":"https://openalex.org/C82687282","wikidata":"https://www.wikidata.org/wiki/Q66221","display_name":"Auxiliary memory","level":2,"score":0.3075999915599823},{"id":"https://openalex.org/C47487241","wikidata":"https://www.wikidata.org/wiki/Q5227230","display_name":"Data access","level":2,"score":0.30720001459121704},{"id":"https://openalex.org/C2775973920","wikidata":"https://www.wikidata.org/wiki/Q3252726","display_name":"Selection algorithm","level":3,"score":0.27619999647140503},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.2687000036239624},{"id":"https://openalex.org/C2780385302","wikidata":"https://www.wikidata.org/wiki/Q367158","display_name":"Protocol (science)","level":3,"score":0.2639999985694885},{"id":"https://openalex.org/C2781335571","wikidata":"https://www.wikidata.org/wiki/Q2633544","display_name":"GPU cluster","level":3,"score":0.25839999318122864},{"id":"https://openalex.org/C2989134064","wikidata":"https://www.wikidata.org/wiki/Q288510","display_name":"Execution time","level":2,"score":0.25360000133514404},{"id":"https://openalex.org/C2984984529","wikidata":"https://www.wikidata.org/wiki/Q7619925","display_name":"Storage management","level":2,"score":0.25060001015663147}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/pact65351.2025.00013","is_oa":false,"landing_page_url":"https://doi.org/10.1109/pact65351.2025.00013","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 34th International Conference on Parallel Architectures and Compilation Techniques (PACT)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320671","display_name":"National Research Foundation","ror":"https://ror.org/05s0g1g46"}],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":20,"referenced_works":["https://openalex.org/W3171639395","https://openalex.org/W4308536694","https://openalex.org/W4312402905","https://openalex.org/W4316660797","https://openalex.org/W4317038433","https://openalex.org/W4318541517","https://openalex.org/W4390636178","https://openalex.org/W4393407316","https://openalex.org/W4400081222","https://openalex.org/W4401211590","https://openalex.org/W4402671766","https://openalex.org/W4402671799","https://openalex.org/W4404782591","https://openalex.org/W4404954500","https://openalex.org/W4404954870","https://openalex.org/W4407425079","https://openalex.org/W4411487104","https://openalex.org/W4412889578","https://openalex.org/W4414197699","https://openalex.org/W4415795655"],"related_works":[],"abstract_inverted_index":{"The":[0],"expansion":[1],"of":[2],"context":[3,62],"windows":[4],"in":[5,20],"large":[6],"language":[7],"models":[8],"(LLMs)":[9],"to":[10,40,57,94,122,144,164,170,179],"multi-million":[11],"tokens":[12,56],"introduces":[13],"severe":[14],"memory":[15,60,82],"and":[16,83,104,117,126,147,155,177],"compute":[17,124],"bottlenecks,":[18],"particularly":[19],"managing":[21],"the":[22,37,187],"growing":[23],"Key-Value":[24],"(KV)":[25],"cache.":[26],"While":[27],"Compute":[28],"Express":[29],"Link":[30],"(CXL)":[31],"enables":[32],"non-eviction":[33],"frameworks":[34,45],"that":[35,80,190],"offload":[36],"full":[38],"KV-cache":[39],"scalable":[41,68,198],"external":[42],"memory,":[43,100],"these":[44],"still":[46],"suffer":[47],"from":[48],"costly":[49,102],"data":[50],"transfers":[51],"when":[52],"recalling":[53],"non-resident":[54],"KV":[55],"limited":[58],"GPU":[59,86,107],"as":[61,196],"lengths":[63],"increase.":[64],"This":[65],"work":[66],"proposes":[67],"Processing-NearMemory":[69],"(PNM)":[70],"for":[71,140,200],"1M-Token":[72],"LLM":[73,203],"Inference,":[74],"a":[75,95,113,118,130,197],"CXL-enabled":[76,191],"KVcache":[77],"management":[78],"system":[79],"coordinates":[81],"computation":[84],"beyond":[85],"limits.":[87],"Our":[88,150],"design":[89],"offloads":[90],"token":[91],"page":[92],"selection":[93,120],"PNM":[96],"accelerator":[97],"within":[98],"CXL":[99],"eliminating":[101],"recalls":[103],"enabling":[105],"larger":[106],"batch":[108],"sizes.":[109],"We":[110],"further":[111],"introduce":[112],"hybrid":[114,157],"parallelization":[115],"strategy":[116],"steady-token":[119,159],"mechanism":[121],"enhance":[123],"efficiency":[125,185],"scalability.":[127],"Implemented":[128],"atop":[129],"state-of-the-art":[131],"CXL-PNM":[132],"system,":[133],"our":[134],"solution":[135],"delivers":[136],"consistent":[137],"performance":[138],"gains":[139],"LLMs":[141],"with":[142,158],"up":[143,163,169,178],"405B":[145],"parameters":[146],"1Mtoken":[148],"contexts.":[149],"PNM-only":[151],"offloading":[152],"scheme":[153],"(PNM-KV)":[154],"GPU-PNM":[156],"execution":[160],"(PnG-KV)":[161],"achieve":[162],"$21.9":[165],"\\times$":[166,172,181],"throughput":[167],"improvement,":[168],"$60":[171],"lower":[173],"energy":[174],"per":[175],"token,":[176],"$7.3":[180],"better":[182],"total":[183],"cost":[184],"than":[186],"baseline,":[188],"demonstrating":[189],"multi-PNM":[192],"architectures":[193],"can":[194],"serve":[195],"backbone":[199],"future":[201],"long-context":[202],"inference.":[204]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-12-16T00:00:00"}