{"id":"https://openalex.org/W4411486231","doi":"https://doi.org/10.1145/3695053.3731412","title":"Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures","display_name":"Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures","publication_year":2025,"publication_date":"2025-06-20","ids":{"openalex":"https://openalex.org/W4411486231","doi":"https://doi.org/10.1145/3695053.3731412"},"language":"en","primary_location":{"id":"doi:10.1145/3695053.3731412","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3695053.3731412","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3695053.3731412","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 52nd Annual International Symposium on Computer Architecture","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://dl.acm.org/doi/pdf/10.1145/3695053.3731412","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5079470006","display_name":"Chenggang Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chenggang Zhao","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0005-2297-0790","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5084839131","display_name":"Chengqi Deng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chengqi Deng","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0006-4777-2448","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5016017027","display_name":"Chong Ruan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chong Ruan","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0000-7896-2558","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020456783","display_name":"Damai Dai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Damai Dai","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0004-9714-7902","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102602497","display_name":"Huazuo Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huazuo Gao","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-8238-440X","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104127081","display_name":"Jiashi Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiashi Li","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0007-0709-7692","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5076005594","display_name":"Liyue Zhang","orcid":"https://orcid.org/0009-0002-8250-1249"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liyue Zhang","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0002-8250-1249","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5053679681","display_name":"Panpan Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Panpan Huang","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0008-2875-0184","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035849215","display_name":"Shangyan Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shangyan Zhou","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0002-9696-1801","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109920796","display_name":"S. MA","orcid":"https://orcid.org/0009-0008-1686-407X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shirong Ma","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0008-1686-407X","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5110193145","display_name":"Wenfeng Liang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenfeng Liang","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0009-2177-6864","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Ying He","orcid":"https://orcid.org/0009-0003-2906-7751"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ying He","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0003-2906-7751","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Yuqing Wang","orcid":"https://orcid.org/0009-0004-9282-0408"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuqing Wang","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0004-9282-0408","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040821260","display_name":"Yuxuan Liu","orcid":"https://orcid.org/0009-0008-2770-7926"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yuxuan Liu","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0008-2770-7926","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5039873833","display_name":"Yen-Hsin Wei","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Y.X. Wei","raw_affiliation_strings":["DeepSeek-AI, Beijing, China"],"raw_orcid":"https://orcid.org/0009-0001-3926-923X","affiliations":[{"raw_affiliation_string":"DeepSeek-AI, Beijing, China","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":44.979,"has_fulltext":true,"cited_by_count":33,"citation_normalized_percentile":{"value":1.0,"is_in_top_1_percent":true,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":99,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"1731","last_page":"1745"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9995999932289124,"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/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.9995999932289124,"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/T10829","display_name":"Interconnection Networks and Systems","score":0.9983999729156494,"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/T10904","display_name":"Embedded Systems Design Techniques","score":0.9980000257492065,"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/computer-science","display_name":"Computer science","score":0.7253686785697937},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.6077454686164856},{"id":"https://openalex.org/keywords/computer-architecture","display_name":"Computer architecture","score":0.5505592226982117},{"id":"https://openalex.org/keywords/computer-hardware","display_name":"Computer hardware","score":0.3397718071937561},{"id":"https://openalex.org/keywords/embedded-system","display_name":"Embedded system","score":0.3326607346534729}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7253686785697937},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.6077454686164856},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.5505592226982117},{"id":"https://openalex.org/C9390403","wikidata":"https://www.wikidata.org/wiki/Q3966","display_name":"Computer hardware","level":1,"score":0.3397718071937561},{"id":"https://openalex.org/C149635348","wikidata":"https://www.wikidata.org/wiki/Q193040","display_name":"Embedded system","level":1,"score":0.3326607346534729},{"id":"https://openalex.org/C2524010","wikidata":"https://www.wikidata.org/wiki/Q8087","display_name":"Geometry","level":1,"score":0.0},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3695053.3731412","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3695053.3731412","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3695053.3731412","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 52nd Annual International Symposium on Computer Architecture","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.09343","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.09343","pdf_url":"https://arxiv.org/pdf/2505.09343","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"}],"best_oa_location":{"id":"doi:10.1145/3695053.3731412","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3695053.3731412","pdf_url":"https://dl.acm.org/doi/pdf/10.1145/3695053.3731412","source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 52nd Annual International Symposium on Computer Architecture","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4411486231.pdf","grobid_xml":"https://content.openalex.org/works/W4411486231.grobid-xml"},"referenced_works_count":19,"referenced_works":["https://openalex.org/W2128043278","https://openalex.org/W2730016893","https://openalex.org/W2778540992","https://openalex.org/W2987265314","https://openalex.org/W2990138404","https://openalex.org/W3129251660","https://openalex.org/W3147460469","https://openalex.org/W4297097318","https://openalex.org/W4380874786","https://openalex.org/W4385192563","https://openalex.org/W4389518760","https://openalex.org/W4390690464","https://openalex.org/W4390833687","https://openalex.org/W4393145114","https://openalex.org/W4396815229","https://openalex.org/W4399836663","https://openalex.org/W4401176521","https://openalex.org/W4401176799","https://openalex.org/W4405756264"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"The":[0],"rapid":[1],"scaling":[2],"of":[3,54,77,93],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"has":[8],"unveiled":[9],"critical":[10],"limitations":[11],"in":[12,18],"current":[13],"hardware":[14,94],"architectures,":[15],"including":[16],"constraints":[17],"memory":[19,74],"capacity,":[20],"computational":[21],"efficiency,":[22,75],"and":[23,45,59,96],"interconnection":[24],"bandwidth.DeepSeek-V3,":[25],"trained":[26],"on":[27],"2,048":[28],"NVIDIA":[29],"H800":[30],"GPUs,":[31],"demonstrates":[32],"how":[33],"hardware-aware":[34],"model":[35,57],"co-design":[36],"can":[37],"effectively":[38],"address":[39],"these":[40],"challenges,":[41],"enabling":[42],"cost-efficient":[43],"training":[44,87],"inference":[46],"at":[47],"scale.This":[48],"paper":[49],"presents":[50],"an":[51],"in-depth":[52],"analysis":[53],"the":[55,90],"DeepSeek-V3/R1":[56],"architecture":[58],"its":[60],"AI":[61],"infrastructure,":[62],"highlighting":[63],"key":[64],"innovations":[65],"such":[66],"as":[67],"Multi-head":[68],"Latent":[69],"Attention":[70],"(MLA)":[71],"for":[72,81],"enhanced":[73],"Mixture":[76],"Experts":[78],"(MoE)":[79],"architectures":[80],"optimized":[82],"computation-communication":[83],"trade-offs,":[84],"FP8":[85],"mixed-precision":[86],"to":[88,101],"unlock":[89],"full":[91],"potential":[92],"capabilities,":[95],"a":[97],"Multi-Plane":[98],"Network":[99],"Topology":[100],"minimize":[102]},"counts_by_year":[{"year":2026,"cited_by_count":21},{"year":2025,"cited_by_count":12}],"updated_date":"2026-08-11T07:18:39.950985","created_date":"2025-10-10T00:00:00"}