{"id":"https://openalex.org/W7116773616","doi":"https://doi.org/10.1109/lca.2025.3646622","title":"Enabling Cost-Efficient LLM Inference on Mid-Tier GPUs With NMP DIMMs","display_name":"Enabling Cost-Efficient LLM Inference on Mid-Tier GPUs With NMP DIMMs","publication_year":2025,"publication_date":"2025-12-22","ids":{"openalex":"https://openalex.org/W7116773616","doi":"https://doi.org/10.1109/lca.2025.3646622"},"language":null,"primary_location":{"id":"doi:10.1109/lca.2025.3646622","is_oa":false,"landing_page_url":"https://doi.org/10.1109/lca.2025.3646622","pdf_url":null,"source":{"id":"https://openalex.org/S17643076","display_name":"IEEE Computer Architecture Letters","issn_l":"1556-6056","issn":["1556-6056","1556-6064","2473-2575"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Computer Architecture Letters","raw_type":"journal-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/A5121011353","display_name":"Heewoo Kim","orcid":null},"institutions":[{"id":"https://openalex.org/I188538660","display_name":"University of Colorado Boulder","ror":"https://ror.org/02ttsq026","country_code":"US","type":"education","lineage":["https://openalex.org/I188538660"]}],"countries":["US"],"is_corresponding":true,"raw_author_name":"Heewoo Kim","raw_affiliation_strings":["University of Colorado Boulder, Boulder, CO, USA"],"raw_orcid":"https://orcid.org/0000-0001-6748-2890","affiliations":[{"raw_affiliation_string":"University of Colorado Boulder, Boulder, CO, USA","institution_ids":["https://openalex.org/I188538660"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5121000941","display_name":"Alan La","orcid":null},"institutions":[{"id":"https://openalex.org/I188538660","display_name":"University of Colorado Boulder","ror":"https://ror.org/02ttsq026","country_code":"US","type":"education","lineage":["https://openalex.org/I188538660"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Alan La","raw_affiliation_strings":["University of Colorado Boulder, Boulder, CO, USA"],"raw_orcid":"https://orcid.org/0009-0001-1697-1632","affiliations":[{"raw_affiliation_string":"University of Colorado Boulder, Boulder, CO, USA","institution_ids":["https://openalex.org/I188538660"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5061335683","display_name":"Joseph Izraelevitz","orcid":"https://orcid.org/0009-0002-1267-5024"},"institutions":[{"id":"https://openalex.org/I188538660","display_name":"University of Colorado Boulder","ror":"https://ror.org/02ttsq026","country_code":"US","type":"education","lineage":["https://openalex.org/I188538660"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Joseph Izraelevitz","raw_affiliation_strings":["University of Colorado Boulder, Boulder, CO, USA"],"raw_orcid":"https://orcid.org/0009-0002-1267-5024","affiliations":[{"raw_affiliation_string":"University of Colorado Boulder, Boulder, CO, USA","institution_ids":["https://openalex.org/I188538660"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5121011353"],"corresponding_institution_ids":["https://openalex.org/I188538660"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.60681669,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"25","issue":"1","first_page":"21","last_page":"24"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.39590001106262207,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10036","display_name":"Advanced Neural Network Applications","score":0.39590001106262207,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.07959999889135361,"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":0.05790000036358833,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.7950999736785889},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.5242000222206116},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4893999993801117},{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.4041999876499176},{"id":"https://openalex.org/keywords/approximate-inference","display_name":"Approximate inference","score":0.3693999946117401},{"id":"https://openalex.org/keywords/matrix-multiplication","display_name":"Matrix multiplication","score":0.3538999855518341}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8623999953269958},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7950999736785889},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.5242000222206116},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4893999993801117},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.4417000114917755},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.4041999876499176},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.39890000224113464},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.37070000171661377},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.3693999946117401},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.3538999855518341},{"id":"https://openalex.org/C854659","wikidata":"https://www.wikidata.org/wiki/Q1859284","display_name":"Message passing","level":2,"score":0.29899999499320984},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.29100000858306885},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.28450000286102295},{"id":"https://openalex.org/C2777115002","wikidata":"https://www.wikidata.org/wiki/Q7168246","display_name":"Performance prediction","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.28110000491142273},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.27140000462532043},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.26440000534057617}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/lca.2025.3646622","is_oa":false,"landing_page_url":"https://doi.org/10.1109/lca.2025.3646622","pdf_url":null,"source":{"id":"https://openalex.org/S17643076","display_name":"IEEE Computer Architecture Letters","issn_l":"1556-6056","issn":["1556-6056","1556-6064","2473-2575"],"is_oa":false,"is_in_doaj":false,"is_core":true,"host_organization":"https://openalex.org/P4310319808","host_organization_name":"Institute of Electrical and Electronics Engineers","host_organization_lineage":["https://openalex.org/P4310319808"],"host_organization_lineage_names":["Institute of Electrical and Electronics Engineers"],"type":"journal"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"IEEE Computer Architecture Letters","raw_type":"journal-article"}],"best_oa_location":null,"sustainable_development_goals":[{"display_name":"Industry, innovation and infrastructure","score":0.4131578803062439,"id":"https://metadata.un.org/sdg/9"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":9,"referenced_works":["https://openalex.org/W4394998968","https://openalex.org/W4401211642","https://openalex.org/W4401211704","https://openalex.org/W4404955001","https://openalex.org/W4407425079","https://openalex.org/W4408925909","https://openalex.org/W4409248734","https://openalex.org/W4409248815","https://openalex.org/W4411486182"],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"require":[4],"substantial":[5],"computational":[6],"resources,":[7],"making":[8],"cost-efficient":[9,116],"inference":[10,48,118],"challenging.":[11],"Scaling":[12],"out":[13],"with":[14,85],"mid-tier":[15,120],"GPUs":[16,36],"(e.g.,":[17,37],"NVIDIA":[18],"A10)":[19],"appears":[20],"attractive":[21],"for":[22,68,115],"LLMs,":[23],"but":[24],"our":[25,113],"characterization":[26],"shows":[27,81],"that":[28,82],"communication":[29,77,90],"bottlenecks":[30],"prevent":[31],"them":[32],"from":[33],"matching":[34],"high-end":[35],"4\u00d7A100).":[38],"Using":[39],"16\u00d7A10":[40,84],"GPUs,":[41],"we":[42],"find":[43],"the":[44,110],"decode":[45],"stage\u2014dominant":[46],"in":[47,51,55],"runtime\u2014is":[49],"memory-bandwidth-bound":[50],"matrix":[52],"multiplications,":[53],"I/O-bandwidth-bound":[54],"<italic":[56],"xmlns:mml=\"http://www.w3.org/1998/Math/MathML\"":[57],"xmlns:xlink=\"http://www.w3.org/1999/xlink\">AllReduce</i>,":[58],"and":[59,75,88,100],"underutilizes":[60],"compute":[61],"resources.":[62],"These":[63,107],"traits":[64],"make":[65],"it":[66,103],"well-suited":[67],"both":[69],"DIMM-based":[70],"Near-Memory":[71],"Processing":[72],"(NMP)":[73],"offloading":[74],"also":[76],"quantization.":[78],"Analytical":[79],"modeling":[80],"a":[83],"NMP":[86],"DIMMs":[87],"INT8":[89],"quantization":[91],"can":[92],"match":[93],"4\u00d7A100":[94],"performance":[95],"at":[96],"30%":[97],"lower":[98],"cost":[99],"even":[101],"surpass":[102],"under":[104],"equal":[105],"cost.":[106],"results":[108],"demonstrate":[109],"potential":[111],"of":[112],"approach":[114],"LLM":[117],"on":[119],"GPUs.":[121]},"counts_by_year":[],"updated_date":"2026-01-11T23:08:45.486102","created_date":"2025-12-22T00:00:00"}