{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:NKLQ7RPBZ3534B73BNJWYCBQU2","short_pith_number":"pith:NKLQ7RPB","schema_version":"1.0","canonical_sha256":"6a970fc5e1cefbbe07fb0b536c0830a6aabadc3451acecb5d34159e641204035","source":{"kind":"arxiv","id":"1912.03413","version":1},"attestation_state":"computed","paper":{"title":"Dissecting the Graphcore IPU Architecture via Microbenchmarking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AR","cs.PF"],"primary_cat":"cs.DC","authors_text":"Blake Tillman, Daniele Paolo Scarpazza, Marco Maggioni, Zhe Jia","submitted_at":"2019-12-07T02:10:19Z","abstract_excerpt":"This report focuses on the architecture and performance of the Intelligence Processing Unit (IPU), a novel, massively parallel platform recently introduced by Graphcore and aimed at Artificial Intelligence/Machine Learning (AI/ML) workloads. We dissect the IPU's performance behavior using microbenchmarks that we crafted for the purpose. We study the IPU's memory organization and performance. We study the latency and bandwidth that the on-chip and off-chip interconnects offer, both in point-to-point transfers and in a spectrum of collective operations, under diverse loads. We evaluate the IPU's"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"1912.03413","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2019-12-07T02:10:19Z","cross_cats_sorted":["cs.AR","cs.PF"],"title_canon_sha256":"3cb21947e2caa67a74c9f7bd386d17522106cf415f5083b7bb42b3dde22487b1","abstract_canon_sha256":"8324189a67473b4db6ef3c163fa5738f3fea131db1d1766e8794a13f1136f883"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:24:45.226626Z","signature_b64":"t3kPgZ6ja/oxx5TMDWd/qd3VCt2U3ueH8CgOtnq6K2lCDKI47Gs2+qQHNRZxGsgHpHiX7jl57QwiL11QCQGVBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a970fc5e1cefbbe07fb0b536c0830a6aabadc3451acecb5d34159e641204035","last_reissued_at":"2026-07-05T00:24:45.226132Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:24:45.226132Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dissecting the Graphcore IPU Architecture via Microbenchmarking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AR","cs.PF"],"primary_cat":"cs.DC","authors_text":"Blake Tillman, Daniele Paolo Scarpazza, Marco Maggioni, Zhe Jia","submitted_at":"2019-12-07T02:10:19Z","abstract_excerpt":"This report focuses on the architecture and performance of the Intelligence Processing Unit (IPU), a novel, massively parallel platform recently introduced by Graphcore and aimed at Artificial Intelligence/Machine Learning (AI/ML) workloads. We dissect the IPU's performance behavior using microbenchmarks that we crafted for the purpose. We study the IPU's memory organization and performance. We study the latency and bandwidth that the on-chip and off-chip interconnects offer, both in point-to-point transfers and in a spectrum of collective operations, under diverse loads. We evaluate the IPU's"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1912.03413","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/1912.03413/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"1912.03413","created_at":"2026-07-05T00:24:45.226191+00:00"},{"alias_kind":"arxiv_version","alias_value":"1912.03413v1","created_at":"2026-07-05T00:24:45.226191+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1912.03413","created_at":"2026-07-05T00:24:45.226191+00:00"},{"alias_kind":"pith_short_12","alias_value":"NKLQ7RPBZ353","created_at":"2026-07-05T00:24:45.226191+00:00"},{"alias_kind":"pith_short_16","alias_value":"NKLQ7RPBZ3534B73","created_at":"2026-07-05T00:24:45.226191+00:00"},{"alias_kind":"pith_short_8","alias_value":"NKLQ7RPB","created_at":"2026-07-05T00:24:45.226191+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09735","citing_title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2512.22168","citing_title":"TileLoom: Automatic Dataflow Planning for Tile-Based Languages on Spatial Dataflow Accelerators","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2205.14135","citing_title":"FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09735","citing_title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12891","citing_title":"TCL: Enabling Fast and Efficient Cross-Hardware Tensor Program Optimization via Continual Learning","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07881","citing_title":"AccelSync: Verifying Synchronization Coverage in Accelerator Pipeline Programs","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2","json":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2.json","graph_json":"https://pith.science/api/pith-number/NKLQ7RPBZ3534B73BNJWYCBQU2/graph.json","events_json":"https://pith.science/api/pith-number/NKLQ7RPBZ3534B73BNJWYCBQU2/events.json","paper":"https://pith.science/paper/NKLQ7RPB"},"agent_actions":{"view_html":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2","download_json":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2.json","view_paper":"https://pith.science/paper/NKLQ7RPB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1912.03413&json=true","fetch_graph":"https://pith.science/api/pith-number/NKLQ7RPBZ3534B73BNJWYCBQU2/graph.json","fetch_events":"https://pith.science/api/pith-number/NKLQ7RPBZ3534B73BNJWYCBQU2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2/action/storage_attestation","attest_author":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2/action/author_attestation","sign_citation":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2/action/citation_signature","submit_replication":"https://pith.science/pith/NKLQ7RPBZ3534B73BNJWYCBQU2/action/replication_record"}},"created_at":"2026-07-05T00:24:45.226191+00:00","updated_at":"2026-07-05T00:24:45.226191+00:00"}