{"paper":{"title":"Ascend-RaBitQ: Heterogeneous NPU-CPU Acceleration of Billion-Scale Similarity Search with 1-bit Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Decoupling NPU coarse ranking on 1-bit vectors from CPU fine re-ranking accelerates billion-scale vector similarity search by up to 100 times.","cross_cats":[],"primary_cat":"cs.IR","authors_text":"Baolong Cui, Chao Zhan, Chuyue Ye, Fujun He, Hao Yi, Huaxiang Cai, Jie Xiang, Pengfei Zheng, Wenru Yan, Xiabing Li, Yuhang Gai, Yunfei Du, Zetao Lv, Zigang Zhang, Ziyang Zhang","submitted_at":"2026-05-15T14:37:18Z","abstract_excerpt":"Vector similarity search is a critical component of modern AI systems, but traditional CPU-based implementations face fundamental scalability bottlenecks for billion-scale corpora due to prohibitive computational overhead and memory bandwidth limitations. While Neural Processing Units (NPUs) offer orders-of-magnitude higher compute density, existing CPU/GPU-optimized 1-bit RaBitQ quantization implementations cannot be directly ported to NPU architectures due to fundamental hardware mismatches, and homogeneous design paradigms struggle to simultaneously balance accuracy, memory footprint, and p"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Ascend-RaBitQ achieves 3.0* to 62.8* faster index construction than the CPU baseline, up to 4.6* throughput improvement over the fastest CPU IVF-RaBitQ implementation, and over 100* over the mathematically equivalent CPU baseline, while demonstrating encouraging scalability on distributed multi-NPU systems.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the three-stage heterogeneous pipeline (NPU coarse ranking on 1-bit vectors, on-device AI CPU Top-k, host CPU fine re-ranking) preserves accuracy without post-hoc adjustments while the four NPU-native optimizations (fused AIC-AIV operators, computation restructuring, block-level load balancing, intra-NPU pipeline) deliver the reported speedups on real hardware.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Ascend-RaBitQ is the first heterogeneous NPU-CPU optimized IVF-RaBitQ system for billion-scale vector search that decouples coarse ranking on NPU from fine ranking on CPU to leverage optimal hardware per stage.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Decoupling NPU coarse ranking on 1-bit vectors from CPU fine re-ranking accelerates billion-scale vector similarity search by up to 100 times.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b6bcd0503e28f950b85d9b7a4141b775a59bc2073a36e519615cfecd6fb46fdb"},"source":{"id":"2605.16007","kind":"arxiv","version":1},"verdict":{"id":"eb4be89f-042f-4f10-977c-c3982b507bc3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T22:07:24.045111Z","strongest_claim":"Ascend-RaBitQ achieves 3.0* to 62.8* faster index construction than the CPU baseline, up to 4.6* throughput improvement over the fastest CPU IVF-RaBitQ implementation, and over 100* over the mathematically equivalent CPU baseline, while demonstrating encouraging scalability on distributed multi-NPU systems.","one_line_summary":"Ascend-RaBitQ is the first heterogeneous NPU-CPU optimized IVF-RaBitQ system for billion-scale vector search that decouples coarse ranking on NPU from fine ranking on CPU to leverage optimal hardware per stage.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the three-stage heterogeneous pipeline (NPU coarse ranking on 1-bit vectors, on-device AI CPU Top-k, host CPU fine re-ranking) preserves accuracy without post-hoc adjustments while the four NPU-native optimizations (fused AIC-AIV operators, computation restructuring, block-level load balancing, intra-NPU pipeline) deliver the reported speedups on real hardware.","pith_extraction_headline":"Decoupling NPU coarse ranking on 1-bit vectors from CPU fine re-ranking accelerates billion-scale vector similarity search by up to 100 times."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16007/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T22:31:19.639528Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T22:21:49.698988Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T17:33:42.169688Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T17:01:55.650407Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"dd7a0c392cf7e15bec0243e2e26d2d97bc438ee91630c855bd2aaba70a1826fb"},"references":{"count":49,"sample":[{"doi":"","year":2025,"title":"Philip Adams, Menghao Li, Shi Zhang, Li Tan, Qi Chen, Mingqin Li, Zengzhong Li, Knut Risvik, and Harsha Vardhan Simhadri. 2025. Distributedann: Efficient scaling of a single diskann graph across thous","work_id":"01324ea7-dad2-4eae-852f-6d5befeec0fd","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Fabien André, Anne-Marie Kermarrec, and Nicolas Le Scouarnec. 2016. Cache locality is not enough: High-performance nearest neighbor search with product quantization fast scan. In42nd International Con","work_id":"6e0ab3b1-c24e-4284-959d-5c6b30f23f11","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Fabien André, Anne-Marie Kermarrec, and Nicolas Le Scouarnec. 2017. Acceler- ated nearest neighbor search with quick adc. InProceedings of the 2017 ACM on International Conference on Multimedia Retrie","work_id":"222fb1af-85e4-4038-909a-1b8e5f0e7a2e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Artem Babenko and Victor Lempitsky. 2014. The inverted multi-index.IEEE transactions on pattern analysis and machine intelligence37, 6 (2014), 1247–1260","work_id":"7d34e2a5-27d5-4d1e-a6ae-8de3c72269b7","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Oren Barkan and Noam Koenigstein. 2016. Item2vec: neural item embedding for collaborative filtering. In2016 IEEE 26th international workshop on machine learning for signal processing (MLSP). IEEE, 1–6","work_id":"b0f01f13-a02a-42a8-91ff-e85f526dfcfd","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":49,"snapshot_sha256":"33142013c7909c66f7a7da786cd3160a7c80e9aa004d15f4049d87bb394210df","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}