{"id":"https://openalex.org/W7167048275","doi":"https://doi.org/10.1145/3797905.3807839","title":"Closing the Efficiency Gap: AI Datacenter Co-design Roadmap for Scalable Training of LLMs","display_name":"Closing the Efficiency Gap: AI Datacenter Co-design Roadmap for Scalable Training of LLMs","publication_year":2026,"publication_date":"2026-07-02","ids":{"openalex":"https://openalex.org/W7167048275","doi":"https://doi.org/10.1145/3797905.3807839"},"language":null,"primary_location":{"id":"doi:10.1145/3797905.3807839","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3807839","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3797905.3807839","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5026593207","display_name":"Jesmin Jahan Tithi","orcid":"https://orcid.org/0000-0002-2861-8393"},"institutions":[{"id":"https://openalex.org/I1343180700","display_name":"Intel (United States)","ror":"https://ror.org/01ek73717","country_code":"US","type":"company","lineage":["https://openalex.org/I1343180700"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jesmin Jahan Tithi","raw_affiliation_strings":["Intel, Santa Clara, California, USA"],"raw_orcid":"https://orcid.org/0000-0002-2861-8393","affiliations":[{"raw_affiliation_string":"Intel, Santa Clara, California, USA","institution_ids":["https://openalex.org/I1343180700"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102596877","display_name":"Hanjiang Wu","orcid":"https://orcid.org/0009-0003-3718-5272"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hanjiang Wu","raw_affiliation_strings":["Georgia Institute of Technology, Atlanta, Georgia, USA"],"raw_orcid":"https://orcid.org/0009-0003-3718-5272","affiliations":[{"raw_affiliation_string":"Georgia Institute of Technology, Atlanta, Georgia, USA","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042175526","display_name":"Joongun Park","orcid":"https://orcid.org/0009-0000-8406-0817"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Joongun Park","raw_affiliation_strings":["Georgia Institute of Technology, Atlanta, Georgia, USA"],"raw_orcid":"https://orcid.org/0009-0000-8406-0817","affiliations":[{"raw_affiliation_string":"Georgia Institute of Technology, Atlanta, Georgia, USA","institution_ids":["https://openalex.org/I130701444"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5028182668","display_name":"Avishaii Abuhatzera","orcid":null},"institutions":[{"id":"https://openalex.org/I4210104622","display_name":"Intel (Israel)","ror":"https://ror.org/027t2s119","country_code":"IL","type":"company","lineage":["https://openalex.org/I1343180700","https://openalex.org/I4210104622"]}],"countries":["IL"],"is_corresponding":false,"raw_author_name":"Avishaii Abuhatzera","raw_affiliation_strings":["Ex-Intel, Haifa, Israel"],"raw_orcid":"https://orcid.org/0009-0000-4000-3132","affiliations":[{"raw_affiliation_string":"Ex-Intel, Haifa, Israel","institution_ids":["https://openalex.org/I4210104622"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066195314","display_name":"Fabrizio Petrini","orcid":"https://orcid.org/0000-0002-4977-7107"},"institutions":[{"id":"https://openalex.org/I1343180700","display_name":"Intel (United States)","ror":"https://ror.org/01ek73717","country_code":"US","type":"company","lineage":["https://openalex.org/I1343180700"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Fabrizio Petrini","raw_affiliation_strings":["Intel, Santa Clara, California, USA"],"raw_orcid":"https://orcid.org/0000-0002-4977-7107","affiliations":[{"raw_affiliation_string":"Intel, Santa Clara, California, USA","institution_ids":["https://openalex.org/I1343180700"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5034089074","display_name":"Tushar Krishna","orcid":"https://orcid.org/0000-0001-5738-6942"},"institutions":[{"id":"https://openalex.org/I130701444","display_name":"Georgia Institute of Technology","ror":"https://ror.org/01zkghx44","country_code":"US","type":"education","lineage":["https://openalex.org/I130701444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Tushar Krishna","raw_affiliation_strings":["Georgia Institute of Technology, Atlanta, Georgia, USA"],"raw_orcid":"https://orcid.org/0000-0001-5738-6942","affiliations":[{"raw_affiliation_string":"Georgia Institute of Technology, Atlanta, Georgia, USA","institution_ids":["https://openalex.org/I130701444"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"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":0.87043072,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"314","last_page":"327"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10101","display_name":"Cloud Computing and Resource Management","score":0.819100022315979,"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"}},"topics":[{"id":"https://openalex.org/T10101","display_name":"Cloud Computing and Resource Management","score":0.819100022315979,"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"}},{"id":"https://openalex.org/T10714","display_name":"Software-Defined Networks and 5G","score":0.07739999890327454,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T10847","display_name":"Advanced Optical Network Technologies","score":0.021299999207258224,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/scalability","display_name":"Scalability","score":0.7339000105857849},{"id":"https://openalex.org/keywords/agile-software-development","display_name":"Agile software development","score":0.6638000011444092},{"id":"https://openalex.org/keywords/suite","display_name":"Suite","score":0.6342999935150146},{"id":"https://openalex.org/keywords/traverse","display_name":"Traverse","score":0.6180999875068665},{"id":"https://openalex.org/keywords/bandwidth","display_name":"Bandwidth (computing)","score":0.5127000212669373},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.47929999232292175},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.4474000036716461},{"id":"https://openalex.org/keywords/network-topology","display_name":"Network topology","score":0.42559999227523804}],"concepts":[{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7339000105857849},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6880000233650208},{"id":"https://openalex.org/C14185376","wikidata":"https://www.wikidata.org/wiki/Q30232","display_name":"Agile software development","level":2,"score":0.6638000011444092},{"id":"https://openalex.org/C79581498","wikidata":"https://www.wikidata.org/wiki/Q1367530","display_name":"Suite","level":2,"score":0.6342999935150146},{"id":"https://openalex.org/C176809094","wikidata":"https://www.wikidata.org/wiki/Q15401496","display_name":"Traverse","level":2,"score":0.6180999875068665},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.5338000059127808},{"id":"https://openalex.org/C2776257435","wikidata":"https://www.wikidata.org/wiki/Q1576430","display_name":"Bandwidth (computing)","level":2,"score":0.5127000212669373},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.47929999232292175},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.4474000036716461},{"id":"https://openalex.org/C199845137","wikidata":"https://www.wikidata.org/wiki/Q145490","display_name":"Network topology","level":2,"score":0.42559999227523804},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.41819998621940613},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.3952000141143799},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.3781000077724457},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.3386000096797943},{"id":"https://openalex.org/C31258907","wikidata":"https://www.wikidata.org/wiki/Q1301371","display_name":"Computer network","level":1,"score":0.3197000026702881},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.3158000111579895},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.3124000132083893},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2865999937057495},{"id":"https://openalex.org/C2778775528","wikidata":"https://www.wikidata.org/wiki/Q5135432","display_name":"Closing (real estate)","level":2,"score":0.2840000092983246},{"id":"https://openalex.org/C10597312","wikidata":"https://www.wikidata.org/wiki/Q5473302","display_name":"Forwarding plane","level":3,"score":0.28189998865127563},{"id":"https://openalex.org/C114563136","wikidata":"https://www.wikidata.org/wiki/Q19725982","display_name":"Network planning and design","level":2,"score":0.28130000829696655},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C2778464652","wikidata":"https://www.wikidata.org/wiki/Q309849","display_name":"Open research","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C44210515","wikidata":"https://www.wikidata.org/wiki/Q16968978","display_name":"Bespoke","level":2,"score":0.2587999999523163}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3797905.3807839","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3807839","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3797905.3807839","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3797905.3807839","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 40th ACM International Conference on Supercomputing","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.5395806431770325,"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":12,"referenced_works":["https://openalex.org/W2019250072","https://openalex.org/W2986148674","https://openalex.org/W3037182822","https://openalex.org/W3095319910","https://openalex.org/W4297097347","https://openalex.org/W4378895279","https://openalex.org/W4381886086","https://openalex.org/W4381894573","https://openalex.org/W4388031315","https://openalex.org/W4399347275","https://openalex.org/W4401211883","https://openalex.org/W4401567198"],"related_works":[],"abstract_inverted_index":{"The":[0],"massive":[1],"compute,":[2],"memory,":[3],"and":[4,21,43,59,64,89,103,163,170,180,188,192,202,210,216],"networking":[5],"needs":[6],"for":[7,31,34,139],"LLM":[8],"training":[9],"necessitate":[10],"a":[11,41,81,129],"fundamental":[12],"rethinking":[13],"of":[14,27,54,100,122,131,150,177,200],"datacenter":[15],"architectures":[16],"to":[17,72,134,147],"ensure":[18],"scalability,":[19],"efficiency,":[20],"cost-effectiveness.":[22],"In":[23],"particular,":[24],"the":[25,28,52,55,74,98,144,148,175,185],"design":[26,209],"network":[29,92,152],"fabric":[30],"AI":[32],"datacenters":[33],"emerging":[35],"LLMs":[36],"(such":[37],"as":[38],"MoEs)":[39],"remains":[40],"crucial":[42],"challenging":[44],"open":[45],"question,":[46],"spanning":[47],"technology":[48],"choices":[49,212],"(that":[50],"determine":[51],"size":[53,99],"high-bandwidth":[56],"domain),":[57],"topology,":[58],"software":[60],"optimizations":[61],"(collective":[62],"algorithms":[63],"overlap":[65],"strategies).":[66],"This":[67,77],"necessitates":[68],"an":[69,136],"agile":[70],"framework":[71],"traverse":[73],"co-design":[75],"space.":[76],"work":[78],"introduces":[79],"Calculon-MoE,":[80,126],"tool":[82],"that":[83,114],"jointly":[84],"explores":[85],"FLOPS,":[86],"HBM":[87],"bandwidth":[88,191],"capacity,":[90],"multiple":[91],"topologies":[93],"(Two-tiered":[94],"vs.":[95],"FullFlat":[96],"optical),":[97],"scale-up":[101,186],"domain,":[102,187],"popular":[104],"parallelism/optimization":[105],"strategies":[106],"used":[107],"in":[108,219],"LLMs.":[109],"Our":[110,194],"validation":[111],"studies":[112,133],"demonstrate":[113,164],"our":[115],"LLM/MoE":[116],"runtime":[117],"predictions":[118],"are":[119],"within":[120],"10%":[121],"real-world":[123],"measurements.":[124],"Using":[125],"we":[127],"conduct":[128],"suite":[130],"case":[132],"develop":[135],"actionable":[137],"roadmap":[138],"data":[140],"centers.":[141],"For":[142],"example,":[143],"results":[145],"point":[146],"promise":[149],"Fullflat":[151],"architectures,":[153],"which":[154],"provide":[155],"uniform":[156],"high-bandwidth,":[157],"low-latency":[158],"connectivity":[159],"between":[160],"all":[161],"nodes":[162],"their":[165],"positive":[166],"impacts":[167],"on":[168],"performance":[169],"scalability.":[171],"We":[172],"also":[173],"quantify":[174],"benefits":[176],"overlapping":[178],"compute":[179],"communication,":[181],"hardware-accelerated":[182],"collectives,":[183],"widening":[184],"higher":[189],"memory":[190],"capacity.":[193],"study":[195],"spans":[196],"both":[197,220],"sparse":[198],"(mixture":[199],"experts)":[201],"dense":[203],"transformer-based":[204],"LLMs,":[205],"revealing":[206],"how":[207],"system":[208,214],"optimization":[211],"affect":[213],"efficiency":[215],"overall":[217],"throughput":[218],"cases.":[221]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-03T00:00:00"}