{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:6ISI6BNWF24OLWH5GSVL6KZ3IM","short_pith_number":"pith:6ISI6BNW","schema_version":"1.0","canonical_sha256":"f2248f05b62eb8e5d8fd34aabf2b3b4323ddd19cd0181fe5038e4ef8e6f4a689","source":{"kind":"arxiv","id":"1907.10701","version":4},"attestation_state":"computed","paper":{"title":"Benchmarking TPU, GPU, and CPU Platforms for Deep Learning","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.PF","stat.ML"],"primary_cat":"cs.LG","authors_text":"David Brooks, Gu-Yeon Wei, Yu Emma Wang","submitted_at":"2019-07-24T20:18:28Z","abstract_excerpt":"Training deep learning models is compute-intensive and there is an industry-wide trend towards hardware specialization to improve performance. To systematically benchmark deep learning platforms, we introduce ParaDnn, a parameterized benchmark suite for deep learning that generates end-to-end models for fully connected (FC), convolutional (CNN), and recurrent (RNN) neural networks. Along with six real-world models, we benchmark Google's Cloud TPU v2/v3, NVIDIA's V100 GPU, and an Intel Skylake CPU platform. We take a deep dive into TPU architecture, reveal its bottlenecks, and highlight valuabl"},"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":"1907.10701","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2019-07-24T20:18:28Z","cross_cats_sorted":["cs.PF","stat.ML"],"title_canon_sha256":"0ea3690756adea8194792dfa69db08268dc291f16724c98eef9b8c306908e564","abstract_canon_sha256":"298c27ea4299f879e8d7631615e7937dc98ab51f58b146d7c9ea269897cab6e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:13:51.864377Z","signature_b64":"DA/H9VB1Mf8gzyb/fXSGGFpobZ4GjxVkafCGYAp7cDMjKikWaVzfnmtw+09GEpdbv8xqfBw46ZyY+jHejmO/Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f2248f05b62eb8e5d8fd34aabf2b3b4323ddd19cd0181fe5038e4ef8e6f4a689","last_reissued_at":"2026-07-05T00:13:51.864027Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:13:51.864027Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Benchmarking TPU, GPU, and CPU Platforms for Deep Learning","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.PF","stat.ML"],"primary_cat":"cs.LG","authors_text":"David Brooks, Gu-Yeon Wei, Yu Emma Wang","submitted_at":"2019-07-24T20:18:28Z","abstract_excerpt":"Training deep learning models is compute-intensive and there is an industry-wide trend towards hardware specialization to improve performance. To systematically benchmark deep learning platforms, we introduce ParaDnn, a parameterized benchmark suite for deep learning that generates end-to-end models for fully connected (FC), convolutional (CNN), and recurrent (RNN) neural networks. Along with six real-world models, we benchmark Google's Cloud TPU v2/v3, NVIDIA's V100 GPU, and an Intel Skylake CPU platform. We take a deep dive into TPU architecture, reveal its bottlenecks, and highlight valuabl"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1907.10701","kind":"arxiv","version":4},"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/1907.10701/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":"1907.10701","created_at":"2026-07-05T00:13:51.864082+00:00"},{"alias_kind":"arxiv_version","alias_value":"1907.10701v4","created_at":"2026-07-05T00:13:51.864082+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1907.10701","created_at":"2026-07-05T00:13:51.864082+00:00"},{"alias_kind":"pith_short_12","alias_value":"6ISI6BNWF24O","created_at":"2026-07-05T00:13:51.864082+00:00"},{"alias_kind":"pith_short_16","alias_value":"6ISI6BNWF24OLWH5","created_at":"2026-07-05T00:13:51.864082+00:00"},{"alias_kind":"pith_short_8","alias_value":"6ISI6BNW","created_at":"2026-07-05T00:13:51.864082+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23833","citing_title":"DORA: Dataflow-Instruction Orchestration Architecture for DNN Acceleration","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2409.07609","citing_title":"Survival of the Cheapest: Cost-Aware Hardware Adaptation for Adversarial Robustness","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2503.08223","citing_title":"Will LLMs Scaling Hit the Wall? Breaking Barriers via Distributed Resources on Massive Edge Devices","ref_index":134,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07523","citing_title":"FILCO: Flexible Composing Architecture with Real-Time Reconfigurability for DNN Acceleration","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17476","citing_title":"Privatar: Scalable Privacy-preserving Multi-user VR via Secure Offloading","ref_index":79,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM","json":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM.json","graph_json":"https://pith.science/api/pith-number/6ISI6BNWF24OLWH5GSVL6KZ3IM/graph.json","events_json":"https://pith.science/api/pith-number/6ISI6BNWF24OLWH5GSVL6KZ3IM/events.json","paper":"https://pith.science/paper/6ISI6BNW"},"agent_actions":{"view_html":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM","download_json":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM.json","view_paper":"https://pith.science/paper/6ISI6BNW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1907.10701&json=true","fetch_graph":"https://pith.science/api/pith-number/6ISI6BNWF24OLWH5GSVL6KZ3IM/graph.json","fetch_events":"https://pith.science/api/pith-number/6ISI6BNWF24OLWH5GSVL6KZ3IM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM/action/storage_attestation","attest_author":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM/action/author_attestation","sign_citation":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM/action/citation_signature","submit_replication":"https://pith.science/pith/6ISI6BNWF24OLWH5GSVL6KZ3IM/action/replication_record"}},"created_at":"2026-07-05T00:13:51.864082+00:00","updated_at":"2026-07-05T00:13:51.864082+00:00"}