{"id":"https://openalex.org/W7127936796","doi":"https://doi.org/10.1145/3748173.3779551","title":"Striking the Balance: GEMM Performance Optimization Across Generations of Ryzen\u2122 AI NPUs","display_name":"Striking the Balance: GEMM Performance Optimization Across Generations of Ryzen\u2122 AI NPUs","publication_year":2026,"publication_date":"2026-02-05","ids":{"openalex":"https://openalex.org/W7127936796","doi":"https://doi.org/10.1145/3748173.3779551"},"language":null,"primary_location":{"id":"doi:10.1145/3748173.3779551","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779551","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 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3748173.3779551","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5084140879","display_name":"Endri Taka","orcid":null},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Endri Taka","raw_affiliation_strings":["The University of Texas at Austin, Austin, TX, USA"],"raw_orcid":"https://orcid.org/0009-0000-5136-7580","affiliations":[{"raw_affiliation_string":"The University of Texas at Austin, Austin, TX, USA","institution_ids":["https://openalex.org/I86519309"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120843086","display_name":"Andre Roesti","orcid":null},"institutions":[{"id":"https://openalex.org/I4210137977","display_name":"Advanced Micro Devices (United States)","ror":"https://ror.org/04kd6c783","country_code":"US","type":"company","lineage":["https://openalex.org/I4210137977"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Andre Roesti","raw_affiliation_strings":["Advanced Micro Devices, Inc., Longmont, CO, USA"],"raw_orcid":"https://orcid.org/0009-0000-5971-8584","affiliations":[{"raw_affiliation_string":"Advanced Micro Devices, Inc., Longmont, CO, USA","institution_ids":["https://openalex.org/I4210137977"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006148629","display_name":"Joseph Melber","orcid":"https://orcid.org/0000-0001-9519-0502"},"institutions":[{"id":"https://openalex.org/I4210137977","display_name":"Advanced Micro Devices (United States)","ror":"https://ror.org/04kd6c783","country_code":"US","type":"company","lineage":["https://openalex.org/I4210137977"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Joseph Melber","raw_affiliation_strings":["Advanced Micro Devices, Inc., Longmont, CO, USA"],"raw_orcid":"https://orcid.org/0000-0001-9519-0502","affiliations":[{"raw_affiliation_string":"Advanced Micro Devices, Inc., Longmont, CO, USA","institution_ids":["https://openalex.org/I4210137977"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123354892","display_name":"Pranathi Vasireddy","orcid":null},"institutions":[{"id":"https://openalex.org/I4210137977","display_name":"Advanced Micro Devices (United States)","ror":"https://ror.org/04kd6c783","country_code":"US","type":"company","lineage":["https://openalex.org/I4210137977"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Pranathi Vasireddy","raw_affiliation_strings":["Advanced Micro Devices, Inc., Longmont, CO, USA"],"raw_orcid":"https://orcid.org/0000-0003-4352-3816","affiliations":[{"raw_affiliation_string":"Advanced Micro Devices, Inc., Longmont, CO, USA","institution_ids":["https://openalex.org/I4210137977"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5035560671","display_name":"Kristof Denolf","orcid":"https://orcid.org/0000-0001-6668-4562"},"institutions":[{"id":"https://openalex.org/I4210137977","display_name":"Advanced Micro Devices (United States)","ror":"https://ror.org/04kd6c783","country_code":"US","type":"company","lineage":["https://openalex.org/I4210137977"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kristof Denolf","raw_affiliation_strings":["Advanced Micro Devices, Inc., Longmont, CO, USA"],"raw_orcid":"https://orcid.org/0000-0001-6668-4562","affiliations":[{"raw_affiliation_string":"Advanced Micro Devices, Inc., Longmont, CO, USA","institution_ids":["https://openalex.org/I4210137977"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5065985595","display_name":"Diana Marculescu","orcid":"https://orcid.org/0000-0002-5734-4221"},"institutions":[{"id":"https://openalex.org/I86519309","display_name":"The University of Texas at Austin","ror":"https://ror.org/00hj54h04","country_code":"US","type":"education","lineage":["https://openalex.org/I86519309"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Diana Marculescu","raw_affiliation_strings":["The University of Texas at Austin, Austin, TX, USA"],"raw_orcid":"https://orcid.org/0000-0002-5734-4221","affiliations":[{"raw_affiliation_string":"The University of Texas at Austin, Austin, TX, USA","institution_ids":["https://openalex.org/I86519309"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":6,"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":0.12170453,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"42","last_page":"42"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10054","display_name":"Parallel Computing and Optimization Techniques","score":0.3271999955177307,"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":0.3271999955177307,"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/T11697","display_name":"Numerical Methods and Algorithms","score":0.17659999430179596,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T12808","display_name":"Ferroelectric and Negative Capacitance Devices","score":0.06949999928474426,"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/implementation","display_name":"Implementation","score":0.682200014591217},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.633899986743927},{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.6319000124931335},{"id":"https://openalex.org/keywords/throughput","display_name":"Throughput","score":0.5074999928474426},{"id":"https://openalex.org/keywords/tops","display_name":"TOPS","score":0.5073999762535095},{"id":"https://openalex.org/keywords/matrix-multiplication","display_name":"Matrix multiplication","score":0.4661000072956085},{"id":"https://openalex.org/keywords/workload","display_name":"Workload","score":0.4156000018119812},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.3614000082015991}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8163999915122986},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.682200014591217},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.633899986743927},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.6319000124931335},{"id":"https://openalex.org/C173608175","wikidata":"https://www.wikidata.org/wiki/Q232661","display_name":"Parallel computing","level":1,"score":0.5375000238418579},{"id":"https://openalex.org/C157764524","wikidata":"https://www.wikidata.org/wiki/Q1383412","display_name":"Throughput","level":3,"score":0.5074999928474426},{"id":"https://openalex.org/C2777675136","wikidata":"https://www.wikidata.org/wiki/Q835642","display_name":"TOPS","level":3,"score":0.5073999762535095},{"id":"https://openalex.org/C17349429","wikidata":"https://www.wikidata.org/wiki/Q1049914","display_name":"Matrix multiplication","level":3,"score":0.4661000072956085},{"id":"https://openalex.org/C118524514","wikidata":"https://www.wikidata.org/wiki/Q173212","display_name":"Computer architecture","level":1,"score":0.4223000109195709},{"id":"https://openalex.org/C2778476105","wikidata":"https://www.wikidata.org/wiki/Q628539","display_name":"Workload","level":2,"score":0.4156000018119812},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4154999852180481},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.38109999895095825},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3614000082015991},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.36090001463890076},{"id":"https://openalex.org/C79974875","wikidata":"https://www.wikidata.org/wiki/Q483639","display_name":"Cloud computing","level":2,"score":0.3515999913215637},{"id":"https://openalex.org/C2780595030","wikidata":"https://www.wikidata.org/wiki/Q3860309","display_name":"Multiplication (music)","level":2,"score":0.3450999855995178},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.3359000086784363},{"id":"https://openalex.org/C97137487","wikidata":"https://www.wikidata.org/wiki/Q729138","display_name":"Integer (computer science)","level":2,"score":0.33410000801086426},{"id":"https://openalex.org/C67953723","wikidata":"https://www.wikidata.org/wiki/Q192525","display_name":"Workstation","level":2,"score":0.3151000142097473},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.30239999294281006},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2953000068664551},{"id":"https://openalex.org/C120314980","wikidata":"https://www.wikidata.org/wiki/Q180634","display_name":"Distributed computing","level":1,"score":0.2815999984741211},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2728999853134155},{"id":"https://openalex.org/C83283714","wikidata":"https://www.wikidata.org/wiki/Q121117","display_name":"Supercomputer","level":2,"score":0.27149999141693115},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.2630000114440918},{"id":"https://openalex.org/C107598950","wikidata":"https://www.wikidata.org/wiki/Q259864","display_name":"Microarchitecture","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C106515295","wikidata":"https://www.wikidata.org/wiki/Q26806595","display_name":"Parallel processing","level":2,"score":0.25619998574256897},{"id":"https://openalex.org/C75684735","wikidata":"https://www.wikidata.org/wiki/Q858810","display_name":"Big data","level":2,"score":0.2556999921798706},{"id":"https://openalex.org/C123657996","wikidata":"https://www.wikidata.org/wiki/Q12271","display_name":"Architecture","level":2,"score":0.2526000142097473}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3748173.3779551","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779551","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 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3748173.3779551","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3748173.3779551","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 2026 ACM/SIGDA International Symposium on Field Programmable Gate Arrays","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/9","score":0.4342416822910309,"display_name":"Industry, innovation and infrastructure"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"high":[1],"computational":[2],"and":[3,70,82,107,131],"memory":[4],"demands":[5],"of":[6,17,79,101,144],"modern":[7],"deep":[8],"learning":[9],"(DL)":[10],"workloads":[11,61,147],"have":[12],"led":[13],"to":[14,23,58,103,127],"the":[15,63,75,88],"development":[16],"specialized":[18],"hardware":[19],"devices":[20],"from":[21],"cloud":[22],"edge,":[24],"such":[25],"as":[26],"AMD's":[27,80],"Ryzen\u2122":[28],"AI":[29,150],"XDNA\u2122":[30],"NPUs.":[31,151],"Optimizing":[32],"general":[33],"matrix":[34],"multiplication":[35],"(GEMM)":[36],"algorithms":[37],"for":[38,43,111,117],"these":[39],"architectures":[40],"is":[41],"critical":[42],"improving":[44],"DL":[45],"workload":[46],"performance.":[47],"To":[48],"this":[49,51],"end,":[50],"paper":[52],"presents":[53],"a":[54],"common":[55],"systematic":[56],"methodology":[57],"optimize":[59],"GEMM":[60,96,123,146],"across":[62,94],"two":[64],"current":[65],"NPU":[66],"generations,":[67],"namely":[68],"XDNA":[69],"XDNA2.":[71],"Our":[72],"implementations":[73,124],"exploit":[74],"unique":[76],"architectural":[77],"features":[78],"NPUs":[81],"address":[83],"key":[84,141],"performance":[85,92,142],"bottlenecks":[86],"at":[87],"system":[89],"level.":[90],"End-to-end":[91],"evaluation":[93],"various":[95],"sizes":[97],"demonstrates":[98],"state-of-the-art":[99],"throughput":[100],"up":[102,126],"6.76":[104],"TOPS":[105,109,129,133],"(XDNA)":[106,130],"38.05":[108],"(XDNA2)":[110],"8-bit":[112],"integer":[113],"(int8)":[114],"precision.":[115],"Similarly,":[116],"brain":[118],"floating-point":[119],"(bf16)":[120],"precision,":[121],"our":[122],"attain":[125],"3.14":[128],"14.71":[132],"(XDNA2).":[134],"This":[135],"work":[136],"provides":[137],"significant":[138],"insights":[139],"into":[140],"aspects":[143],"optimizing":[145],"on":[148],"Ryzen":[149]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-02-07T00:00:00"}