{"id":"https://openalex.org/W7225160122","doi":"https://doi.org/10.48550/arxiv.2611.03330","title":"The SIMPLE Law for Energy Cost of LLM Inference","display_name":"The SIMPLE Law for Energy Cost of LLM Inference","publication_year":2026,"publication_date":"2026-11-02","ids":{"openalex":"https://openalex.org/W7225160122","doi":"https://doi.org/10.48550/arxiv.2611.03330"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2611.03330","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03330","pdf_url":"https://arxiv.org/pdf/2611.03330","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2611.03330","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120514218","display_name":"Hiari Pizzini Cavagna","orcid":"https://orcid.org/0009-0005-2768-0418","observed_orcids":["https://orcid.org/0009-0005-2768-0418"]},"institutions":[{"id":"https://openalex.org/I9360294","display_name":"University of Bologna","ror":"https://ror.org/01111rn36","country_code":"IT","type":"education","lineage":["https://openalex.org/I9360294"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Cavagna, Hiari Pizzini","raw_affiliation_strings":["DEI -University of Bologna , Bologna , Italy"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DEI -University of Bologna , Bologna , Italy","institution_ids":["https://openalex.org/I9360294"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111397583","display_name":"Andrea Proia","orcid":"https://orcid.org/0009-0005-6367-762X","observed_orcids":["https://orcid.org/0009-0005-6367-762X"]},"institutions":[{"id":"https://openalex.org/I9360294","display_name":"University of Bologna","ror":"https://ror.org/01111rn36","country_code":"IT","type":"education","lineage":["https://openalex.org/I9360294"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Proia, Andrea","raw_affiliation_strings":["DISI -University of Bologna , Bologna , Italy"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DISI -University of Bologna , Bologna , Italy","institution_ids":["https://openalex.org/I9360294"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002332287","display_name":"Daniele Cesarini","orcid":"https://orcid.org/0000-0003-1294-372X","observed_orcids":["https://orcid.org/0000-0003-1294-372X"]},"institutions":[{"id":"https://openalex.org/I4210119786","display_name":"Cineca","ror":"https://ror.org/02f013h18","country_code":"IT","type":"other","lineage":["https://openalex.org/I4210119786"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Cesarini, Daniele","raw_affiliation_strings":["CINECA , Casalecchio di Reno , Italy"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CINECA , Casalecchio di Reno , Italy","institution_ids":["https://openalex.org/I4210119786"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033859818","display_name":"Zeynep Kiziltan","orcid":"https://orcid.org/0000-0003-0412-4396","observed_orcids":["https://orcid.org/0000-0003-0412-4396"]},"institutions":[{"id":"https://openalex.org/I9360294","display_name":"University of Bologna","ror":"https://ror.org/01111rn36","country_code":"IT","type":"education","lineage":["https://openalex.org/I9360294"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Kiziltan, Zeynep","raw_affiliation_strings":["DISI -University of Bologna , Bologna , Italy"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DISI -University of Bologna , Bologna , Italy","institution_ids":["https://openalex.org/I9360294"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5047906923","display_name":"Andrea Bartolini","orcid":"https://orcid.org/0000-0002-1148-2450","observed_orcids":["https://orcid.org/0000-0002-1148-2450"]},"institutions":[{"id":"https://openalex.org/I9360294","display_name":"University of Bologna","ror":"https://ror.org/01111rn36","country_code":"IT","type":"education","lineage":["https://openalex.org/I9360294"]}],"countries":["IT"],"is_corresponding":false,"raw_author_name":"Bartolini, Andrea","raw_affiliation_strings":["DEI -University of Bologna , Bologna , Italy"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"DEI -University of Bologna , Bologna , Italy","institution_ids":["https://openalex.org/I9360294"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T14098","display_name":"Energy Efficiency in Computing","score":0.2971000075340271,"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/T14098","display_name":"Energy Efficiency in Computing","score":0.2971000075340271,"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/T12238","display_name":"Green IT and Sustainability","score":0.2800000011920929,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.19760000705718994,"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/energy-modeling","display_name":"energy modeling","score":0.44600000977516174},{"id":"https://openalex.org/keywords/kv-cache","display_name":"KV cache","score":0.47999998927116394},{"id":"https://openalex.org/keywords/multi-head-attention","display_name":"multi-head attention","score":0.7400000095367432},{"id":"https://openalex.org/keywords/analytical-cost-model","display_name":"analytical cost model","score":0.414000004529953},{"id":"https://openalex.org/keywords/prefix-caching","display_name":"prefix caching","score":0.9520000219345093},{"id":"https://openalex.org/keywords/nvidia-h100","display_name":"NVIDIA H100","score":0.8029999732971191}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7547000050544739},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6826000213623047},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.6459000110626221},{"id":"https://openalex.org/C2779058145","wikidata":"https://www.wikidata.org/wiki/Q6294583","display_name":"Joule (programming language)","level":3,"score":0.6291999816894531},{"id":"https://openalex.org/C186370098","wikidata":"https://www.wikidata.org/wiki/Q442787","display_name":"Energy (signal processing)","level":2,"score":0.5968999862670898},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5273000001907349},{"id":"https://openalex.org/C144237770","wikidata":"https://www.wikidata.org/wiki/Q747534","display_name":"Mathematical economics","level":1,"score":0.4041000008583069},{"id":"https://openalex.org/C2742236","wikidata":"https://www.wikidata.org/wiki/Q924713","display_name":"Efficient energy use","level":2,"score":0.3684999942779541},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.34790000319480896},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.3330000042915344},{"id":"https://openalex.org/C128849468","wikidata":"https://www.wikidata.org/wiki/Q1341258","display_name":"Energy budget","level":2,"score":0.3303000032901764},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.32679998874664307},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3091000020503998},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.30390000343322754},{"id":"https://openalex.org/C2780331096","wikidata":"https://www.wikidata.org/wiki/Q24965464","display_name":"Energy modeling","level":3,"score":0.3021000027656555},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.29420000314712524},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.2930000126361847},{"id":"https://openalex.org/C2776650193","wikidata":"https://www.wikidata.org/wiki/Q264661","display_name":"Obstacle","level":2,"score":0.28929999470710754},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.28049999475479126},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2773999869823456},{"id":"https://openalex.org/C2777172336","wikidata":"https://www.wikidata.org/wiki/Q2700433","display_name":"Energy security","level":3,"score":0.2759999930858612},{"id":"https://openalex.org/C199539241","wikidata":"https://www.wikidata.org/wiki/Q7748","display_name":"Law","level":1,"score":0.2709999978542328},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.27000001072883606},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.27000001072883606},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.2689000070095062},{"id":"https://openalex.org/C155076860","wikidata":"https://www.wikidata.org/wiki/Q1932004","display_name":"Token economy","level":3,"score":0.2687999904155731},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.26809999346733093},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.267300009727478},{"id":"https://openalex.org/C172790937","wikidata":"https://www.wikidata.org/wiki/Q18343","display_name":"Dark energy","level":3,"score":0.257999986410141},{"id":"https://openalex.org/C175444787","wikidata":"https://www.wikidata.org/wiki/Q39072","display_name":"Microeconomics","level":1,"score":0.2515000104904175}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2611.03330","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03330","pdf_url":"https://arxiv.org/pdf/2611.03330","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2611.03330","doi":"https://doi.org/10.48550/arxiv.2611.03330","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.03330","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2611.03330","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03330","pdf_url":"https://arxiv.org/pdf/2611.03330","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"display_name":"Affordable and clean energy","score":0.5690000057220459,"id":"https://metadata.un.org/sdg/7"}],"study_designs":[],"awards":[],"funders":[{"id":"https://openalex.org/F1100377486","display_name":"European High Performance Computing Joint Undertaking","ror":"https://ror.org/04dvh0v63"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7225160122.pdf","grobid_xml":"https://content.openalex.org/works/W7225160122.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7204823529","https://openalex.org/W4390189088","https://openalex.org/W7171813800","https://openalex.org/W7127953631","https://openalex.org/W7208820377","https://openalex.org/W4416156613","https://openalex.org/W7164308339","https://openalex.org/W7140185737","https://openalex.org/W7224878815","https://openalex.org/W7127147162"],"abstract_inverted_index":{"Generative":[0],"AI":[1],"is":[2,50,283],"billed":[3],"per":[4,225,247],"token,":[5],"yet":[6],"tokens":[7,33],"are":[8],"not":[9,53],"equally":[10],"expensive.":[11],"Each":[12],"token":[13,42,199],"enters":[14],"the":[15,19,45,70,101,149,167,185,197,205,218,257,261,264,276,280],"KV-cache":[16,121,265,281],"and":[17,44,72,85,87,118,123,144,178,232,263],"raises":[18,245],"cost":[20,122,278],"of":[21,69,75,80,91,94,128,259,275],"those":[22],"that":[23,113],"follow,":[24],"so":[25,170,223,285],"energy":[26,163,187,208,212,220,246],"grows":[27,164,216,221],"non-linearly":[28],"along":[29],"a":[30,41,51,54,89,110,239,287],"sequence":[31],"whose":[32],"look":[34],"identical.":[35],"No":[36],"analytical":[37,92],"model":[38,49],"states":[39],"what":[40],"costs,":[43],"most":[46,174,274],"accurate":[47],"SOTA":[48],"fit,":[52],"law:":[55],"it":[56,139],"absorbs":[57],"architecture":[58],"into":[59,120],"fitted":[60],"coefficients,":[61],"requiring":[62],"per-LLM":[63,152],"fitting.":[64],"We":[65],"derive":[66],"closed-form":[67],"expressions":[68],"computational":[71],"memory-access":[73],"complexity":[74],"LLM":[76],"inference":[77],"in":[78],"terms":[79,266],"architectural":[81],"hyperparameters,":[82],"covering":[83],"MHA":[84],"GQA,":[86],"compare":[88],"family":[90],"models":[93],"increasing":[95],"granularity.":[96],"From":[97],"these,":[98],"we":[99],"obtain":[100],"SIMPLE":[102,155],"Inference":[103],"Model":[104],"to":[105,190,193,236,252],"Predict":[106],"LLMs":[107,131,172],"Energy":[108],"(SIMPLE),":[109],"four-parameter":[111],"law":[112,156],"separates":[114],"prefill":[115,162,262,277],"from":[116,132,175],"decode,":[117],"decode":[119],"next-token":[124],"computation.":[125],"Over":[126],"traces":[127],"13":[129],"dense":[130],"six":[133],"families":[134],"on":[135],"NVIDIA":[136],"H100":[137],"GPUs,":[138],"attains":[140],"R2":[141],"=":[142],"0.969":[143],"11.97%":[145],"MAPE,":[146],"3.1-3.5x":[147],"below":[148],"SOTA,":[150],"without":[151],"calibration.":[153],"The":[154],"explains":[157],"as":[158,160],"well":[159],"predicts:":[161],"quadratically":[165],"with":[166,217,267],"hidden":[168],"dimension,":[169],"larger":[171],"suffer":[173],"long":[176],"inputs;":[177],"more":[179],"aggressive":[180],"GQA":[181],"KV":[182],"grouping":[183],"reduces":[184],"per-token":[186],"by":[188,234,249,256],"up":[189,235],"3x":[191],"relative":[192,251],"standard":[194],"MHA.":[195],"Increasing":[196],"reasoning":[198],"budget":[200,231],"pushes":[201],"output":[202],"length":[203],"beyond":[204],"analytically":[206],"derived":[207],"minimum,":[209],"driving":[210],"super-linear":[211],"growth;":[213],"although":[214],"accuracy":[215],"budget,":[219],"faster,":[222],"Intelligence":[224],"Joule":[226],"peaks":[227],"at":[228],"an":[229],"intermediate":[230],"falls":[233],"3.14x":[237],"across":[238],"4.0x":[240],"reasoning-budget":[241],"range.":[242],"Agentic":[243],"serving":[244],"request":[248,290],"13.0x":[250],"non-agentic":[253],"ones,":[254],"driven":[255],"growth":[258],"both":[260],"input":[268],"length;":[269],"while":[270],"prefix":[271],"caching":[272],"removes":[273],"(-93%),":[279],"term":[282],"unaffected,":[284],"even":[286],"cached":[288],"agentic":[289],"costs":[291],"4.8x":[292],"more.":[293]},"counts_by_year":[],"updated_date":"2026-11-06T12:05:46.655746","created_date":"2026-11-04T00:00:00"}