{"id":"https://openalex.org/W4413144996","doi":"https://doi.org/10.1109/ipdpsw66978.2025.00173","title":"Understanding the Performance and Power of LLM Inferencing on Edge Accelerators","display_name":"Understanding the Performance and Power of LLM Inferencing on Edge Accelerators","publication_year":2025,"publication_date":"2025-06-03","ids":{"openalex":"https://openalex.org/W4413144996","doi":"https://doi.org/10.1109/ipdpsw66978.2025.00173"},"language":"en","primary_location":{"id":"doi:10.1109/ipdpsw66978.2025.00173","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ipdpsw66978.2025.00173","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","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/A5116482706","display_name":"Mayank Arya","orcid":null},"institutions":[{"id":"https://openalex.org/I59270414","display_name":"Indian Institute of Science Bangalore","ror":"https://ror.org/04dese585","country_code":"IN","type":"education","lineage":["https://openalex.org/I59270414"]}],"countries":["IN"],"is_corresponding":true,"raw_author_name":"Mayank Arya","raw_affiliation_strings":["Indian Institute of Science,Department of Computational and Data Sciences,Bangalore,India,560012"],"affiliations":[{"raw_affiliation_string":"Indian Institute of Science,Department of Computational and Data Sciences,Bangalore,India,560012","institution_ids":["https://openalex.org/I59270414"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5107958581","display_name":"Yogesh Simmhan","orcid":null},"institutions":[{"id":"https://openalex.org/I59270414","display_name":"Indian Institute of Science Bangalore","ror":"https://ror.org/04dese585","country_code":"IN","type":"education","lineage":["https://openalex.org/I59270414"]}],"countries":["IN"],"is_corresponding":false,"raw_author_name":"Yogesh Simmhan","raw_affiliation_strings":["Indian Institute of Science,Department of Computational and Data Sciences,Bangalore,India,560012"],"affiliations":[{"raw_affiliation_string":"Indian Institute of Science,Department of Computational and Data Sciences,Bangalore,India,560012","institution_ids":["https://openalex.org/I59270414"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":["https://openalex.org/A5116482706"],"corresponding_institution_ids":["https://openalex.org/I59270414"],"apc_list":null,"apc_paid":null,"fwci":1.319,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":{"value":0.8401521,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"1108","last_page":"1111"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9501000046730042,"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":0.9501000046730042,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.6118325591087341},{"id":"https://openalex.org/keywords/power","display_name":"Power (physics)","score":0.5534470677375793},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5300431251525879},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.33304423093795776},{"id":"https://openalex.org/keywords/nuclear-engineering","display_name":"Nuclear engineering","score":0.32760757207870483},{"id":"https://openalex.org/keywords/systems-engineering","display_name":"Systems engineering","score":0.3252413272857666},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.2151508331298828},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.2084801197052002},{"id":"https://openalex.org/keywords/physics","display_name":"Physics","score":0.1736554503440857}],"concepts":[{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.6118325591087341},{"id":"https://openalex.org/C163258240","wikidata":"https://www.wikidata.org/wiki/Q25342","display_name":"Power (physics)","level":2,"score":0.5534470677375793},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5300431251525879},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.33304423093795776},{"id":"https://openalex.org/C116915560","wikidata":"https://www.wikidata.org/wiki/Q83504","display_name":"Nuclear engineering","level":1,"score":0.32760757207870483},{"id":"https://openalex.org/C201995342","wikidata":"https://www.wikidata.org/wiki/Q682496","display_name":"Systems engineering","level":1,"score":0.3252413272857666},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2151508331298828},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2084801197052002},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.1736554503440857},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/ipdpsw66978.2025.00173","is_oa":false,"landing_page_url":"https://doi.org/10.1109/ipdpsw66978.2025.00173","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2748952813","https://openalex.org/W2052899165","https://openalex.org/W2008562718","https://openalex.org/W1495971547","https://openalex.org/W2067604523","https://openalex.org/W2602089430","https://openalex.org/W2154510413","https://openalex.org/W2386935432","https://openalex.org/W77590268","https://openalex.org/W2066556921"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"demonstrated":[5],"exceptional":[6],"benefits":[7],"to":[8,109,147,185],"a":[9,71,90,175],"wide":[10],"range":[11],"of":[12,43,66,79,93,122],"domains,":[13],"for":[14,54,198],"tasks":[15],"as":[16,18,59,113],"diverse":[17],"code":[19],"generation":[20],"and":[21,35,77,128,134,136,150,166,180],"robot":[22],"navigation.":[23],"While":[24],"LLMs":[25,184],"are":[26,70],"usually":[27],"served":[28],"from":[29,107],"cloud":[30],"data":[31],"centers,":[32],"mission":[33],"critical":[34],"privacy":[36],"sensitive":[37],"applications":[38],"may":[39],"require":[40],"local":[41],"hosting":[42],"open":[44],"LLM":[45,80,94,193],"models.":[46],"Given":[47],"the":[48,75,97,120,144,160,171],"large":[49],"GPU":[50],"memory":[51],"footprint":[52],"needed":[53],"LLMs,":[55],"edge":[56,83,196],"accelerators":[57,84,197],"such":[58,112],"Nvidia":[60],"Jetson":[61,99],"Orin":[62,100,145],"AGX":[63,146],"with":[64],"64GB":[65],"shared":[67],"GPU\u2013CPU":[68],"RAM":[69],"compelling":[72],"choice.":[73],"However,":[74],"feasibility":[76],"performance":[78],"inference":[81,95,164],"on":[82,96,102,131,143,159,195],"is":[85],"under-explored.":[86],"This":[87],"study":[88],"presents":[89],"detailed":[91],"evaluation":[92],"NVIDIA":[98],"AGX,":[101],"four":[103],"SOTA":[104],"models":[105],"ranging":[106],"2.7B":[108],"32.8B":[110],"parameters,":[111],"Meta":[114],"Llama3.1,":[115],"Microsoft-Phi2,":[116],"Deepseek-R1-Qwen.":[117],"We":[118],"investigate":[119],"impact":[121],"varying":[123],"batch":[124],"sizes,":[125],"sequence":[126,172],"lengths,":[127],"quantization":[129,181],"levels":[130],"latency,":[132],"throughput":[133,179],"perplexity,":[135],"also":[137],"explore":[138],"various":[139],"custom":[140],"power":[141,149],"modes":[142],"perform":[148],"energy":[151],"consumption":[152],"analysis.":[153],"Our":[154],"findings":[155],"offer":[156],"interesting":[157],"insights":[158],"trade-offs":[161],"between":[162],"efficiency,":[163],"speed":[165],"resource":[167],"use,":[168],"e.g.,":[169],"increasing":[170],"length":[173],"causes":[174,182],"decrease":[176],"in":[177],"token":[178],"smaller":[183],"be":[186],"slower.":[187],"These":[188],"results":[189],"can":[190],"help":[191],"optimize":[192],"serving":[194],"practical":[199],"applications.":[200]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2025-12-28T23:10:05.387466","created_date":"2025-10-10T00:00:00"}