{"id":"https://openalex.org/W7163878524","doi":"https://doi.org/10.48550/arxiv.2606.06527","title":"Characterizing the Impact of NVFP4 Quantization for Low-Power Edge AI Deployment","display_name":"Characterizing the Impact of NVFP4 Quantization for Low-Power Edge AI Deployment","publication_year":2026,"publication_date":"2026-06-03","ids":{"openalex":"https://openalex.org/W7163878524","doi":"https://doi.org/10.48550/arxiv.2606.06527"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.06527","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06527","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.06527","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5085799102","display_name":"Ovishake Sen","orcid":"https://orcid.org/0000-0001-7727-9503"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sen, Ovishake","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113337322","display_name":"V. Kamineni","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kamineni, Venkata Nithin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138113062","display_name":"Daniel Lobo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lobo, Daniel","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138149718","display_name":"Swarup Bhunia","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bhunia, Swarup","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014892277","display_name":"Rickard Ewetz","orcid":"https://orcid.org/0000-0002-4183-6926"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ewetz, Rickard","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5039890044","display_name":"Baibhab Chatterjee","orcid":"https://orcid.org/0000-0002-2688-281X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chatterjee, Baibhab","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"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":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10363","display_name":"Low-power high-performance VLSI design","score":0.4542999863624573,"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"}},"topics":[{"id":"https://openalex.org/T10363","display_name":"Low-power high-performance VLSI design","score":0.4542999863624573,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.1039000004529953,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.07119999825954437,"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/inference","display_name":"Inference","score":0.546500027179718},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4569999873638153},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.4562000036239624},{"id":"https://openalex.org/keywords/multiplication","display_name":"Multiplication (music)","score":0.4196000099182129},{"id":"https://openalex.org/keywords/enhanced-data-rates-for-gsm-evolution","display_name":"Enhanced Data Rates for GSM Evolution","score":0.3837999999523163},{"id":"https://openalex.org/keywords/block","display_name":"Block (permutation group theory)","score":0.3765999972820282},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.3560999929904938}],"concepts":[{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5759999752044678},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.546500027179718},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4569999873638153},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.4562000036239624},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4323999881744385},{"id":"https://openalex.org/C2780595030","wikidata":"https://www.wikidata.org/wiki/Q3860309","display_name":"Multiplication (music)","level":2,"score":0.4196000099182129},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3991999924182892},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.3837999999523163},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.3822999894618988},{"id":"https://openalex.org/C2777210771","wikidata":"https://www.wikidata.org/wiki/Q4927124","display_name":"Block (permutation group theory)","level":2,"score":0.3765999972820282},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3560999929904938},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3402999937534332},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C2780388253","wikidata":"https://www.wikidata.org/wiki/Q5421508","display_name":"Exponent","level":2,"score":0.33730000257492065},{"id":"https://openalex.org/C39927690","wikidata":"https://www.wikidata.org/wiki/Q11197","display_name":"Logarithm","level":2,"score":0.33709999918937683},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.32280001044273376},{"id":"https://openalex.org/C138236772","wikidata":"https://www.wikidata.org/wiki/Q25098575","display_name":"Edge device","level":3,"score":0.2847999930381775},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.26759999990463257},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.2632000148296356},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.26100000739097595}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.06527","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06527","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,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.06527","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06527","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,"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"},"sustainable_development_goals":[{"score":0.9037237763404846,"display_name":"Affordable and clean energy","id":"https://metadata.un.org/sdg/7"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Energy-efficient":[0],"neural-network":[1],"inference":[2,70,157,161,172,234],"at":[3],"the":[4,37,124,138,143,146,206,210,214],"edge":[5,233],"requires":[6],"reducing":[7],"arithmetic":[8],"cost,":[9],"memory":[10],"traffic,":[11],"computation":[12],"energy,":[13],"and":[14,47,62,113,133,158,163,196,253],"storage":[15],"overhead":[16],"while":[17,71,179],"maintaining":[18],"acceptable":[19],"accuracy.":[20,49],"This":[21],"paper":[22],"presents":[23],"an":[24,58,63],"ablation-focused":[25],"study":[26],"of":[27,137,145,216,230,239],"NVFP4":[28,50,126,147,159,180,204],"quantization":[29,132],"for":[30,101,175,227],"edge-efficient":[31,81],"neural":[32],"networks,":[33],"with":[34,162,202],"emphasis":[35],"on":[36],"relationship":[38],"between":[39],"activation":[40,73,127,131,156,160,189],"precision,":[41,43],"weight":[42,106],"block-size":[44,77],"scaling,":[45],"retraining,":[46,203],"model":[48],"activations":[51],"are":[52],"represented":[53],"using":[54],"4-bit":[55],"FP4":[56,121,155,171,218],"data,":[57],"FP8":[59,112,193],"block":[60,85,194],"scale,":[61,66],"FP32":[64,197],"tensor":[65,198],"enabling":[67],"ultra-low":[68],"precision":[69,107],"preserving":[72],"dynamic":[74,190],"range.":[75],"A":[76,105],"ablation":[78,108],"over":[79,120],"six":[80],"models":[82],"shows":[83,110],"that":[84,111,130,169],"size":[86],"B":[87],"=":[88,103],"16":[89],"provides":[90],"a":[91,236],"practical":[92],"accuracy/storage":[93],"trade-off,":[94],"requiring":[95],"only":[96,117],"4.5078":[97],"bits":[98],"per":[99],"input":[100],"N":[102],"4096.":[104],"further":[109],"FP16":[114],"weights":[115,122],"provide":[116,223],"modest":[118],"gains":[119],"under":[123],"same":[125],"path,":[128],"suggesting":[129],"scaling":[134,195],"dominate":[135],"much":[136],"accuracy":[139,174,186,208],"behavior.":[140],"To":[141],"isolate":[142],"benefit":[144],"data":[148],"type,":[149],"this":[150],"work":[151],"compares":[152],"conventional":[153,170],"unscaled":[154],"without":[164,181],"retraining.":[165],"The":[166],"results":[167],"show":[168],"collapses":[173],"most":[176],"compact":[177],"models,":[178,212],"retraining":[182],"already":[183],"recovers":[184],"substantial":[185],"by":[187],"restoring":[188],"range":[191,238],"through":[192],"scaling.":[199],"When":[200],"combined":[201],"achieves":[205],"best":[207],"across":[209,235],"evaluated":[211],"demonstrating":[213],"effectiveness":[215],"scaling-aware":[217],"(NVFP4)":[219],"inference.":[220],"These":[221],"findings":[222],"general":[224],"design":[225],"guidance":[226],"hardware-software":[228],"co-design":[229],"low":[231],"power":[232],"broad":[237],"accelerator":[240],"platforms,":[241],"including":[242],"GPUs,":[243],"Tensor":[244],"Cores,":[245],"FPGAs,":[246],"domain-specific":[247],"AI":[248],"accelerators,":[249],"near-memory":[250],"computing":[251],"systems,":[252],"emerging":[254],"edge-computing":[255],"architectures.":[256]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-09T00:00:00"}