{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:T3BKNKPD76N7BO6Z6MFDBVE4HQ","short_pith_number":"pith:T3BKNKPD","schema_version":"1.0","canonical_sha256":"9ec2a6a9e3ff9bf0bbd9f30a30d49c3c0b557a5f20734d050059ca42bce39227","source":{"kind":"arxiv","id":"2508.14318","version":2},"attestation_state":"computed","paper":{"title":"Power Stabilization for AI Training Datacenters","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.AR","authors_text":"Aaron Hurst, Alex Wetmore, Alok Gautam Kumbhare, Andy Sullivan, Apoorv Gupta, April Zhao, Brian Harry, Brijesh Warrier, Caroline Lichtenberger, Charles Frederick, Daniel Jurnove, David Lukofsky, Divya Ramakrishnan, Esha Choukse, Gene Oden, Girish Bhatia, Gurpreet Sahota, Hassan Ali Khan, Hien Thai, Hirofumi Kobayashi, Ian Buck, Ileana Incorvaia, Jeff Liott, John Siegler, Jonah Alben, Justin Harp, Kelsey Wildstone, Kushal Datta, Laurentiu Olariu, Lawrence Marwong, Luz Belmont, Matthew Kappel, Melur Raghuraman, Michael Petrov, Mike Kendrick, Miro Enev, Mohammed Ayna, Nilesh Dattani, Nithish Mahalingam, Prabhat Ram, Praneeth Gottumukkala, Pratikkumar Patel, Rebecca Naughton, Reid Carper, Reza Zamani, Ricardo Bianchini, Rob Nertney, Rory Carmichael, Russell J. Hewett, Satyajit Desai, Scot Heath, Srini Varkala, Tom Li, Venkatesh Ramamurthy, Xin Li, Yu Pei, Zaid Kahn","submitted_at":"2025-08-20T00:04:06Z","abstract_excerpt":"Large Artificial Intelligence (AI) training workloads spanning several tens of thousands of GPUs present unique power management challenges. These arise due to the high variability in power consumption during the training. Given the synchronous nature of these jobs, during every iteration there is a computation-heavy phase, where each GPU works on the local data, and a communication-heavy phase where all the GPUs synchronize on the data. Because compute-heavy phases require much more power than communication phases, large power swings occur. The amplitude of these power swings is ever increasi"},"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":"2508.14318","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AR","submitted_at":"2025-08-20T00:04:06Z","cross_cats_sorted":["cs.AI","cs.DC"],"title_canon_sha256":"81a113b563d780167ff1f97c902fb4ae0b03a30cdf3be732cfef57ae3b94904c","abstract_canon_sha256":"b44ad7ce7a891d98b706988e0571af09929178cbee4789137d5ad15fb9678747"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:57:02.789456Z","signature_b64":"OIuUolLj3SkRzH8weACC1rQ1bWa7V7GP393OVPXYf89wwICGxc0s3vQOmq4TehWEu57EoIFwgjbfBjlzcdz6Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9ec2a6a9e3ff9bf0bbd9f30a30d49c3c0b557a5f20734d050059ca42bce39227","last_reissued_at":"2026-07-05T11:57:02.788934Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:57:02.788934Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Power Stabilization for AI Training Datacenters","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.AR","authors_text":"Aaron Hurst, Alex Wetmore, Alok Gautam Kumbhare, Andy Sullivan, Apoorv Gupta, April Zhao, Brian Harry, Brijesh Warrier, Caroline Lichtenberger, Charles Frederick, Daniel Jurnove, David Lukofsky, Divya Ramakrishnan, Esha Choukse, Gene Oden, Girish Bhatia, Gurpreet Sahota, Hassan Ali Khan, Hien Thai, Hirofumi Kobayashi, Ian Buck, Ileana Incorvaia, Jeff Liott, John Siegler, Jonah Alben, Justin Harp, Kelsey Wildstone, Kushal Datta, Laurentiu Olariu, Lawrence Marwong, Luz Belmont, Matthew Kappel, Melur Raghuraman, Michael Petrov, Mike Kendrick, Miro Enev, Mohammed Ayna, Nilesh Dattani, Nithish Mahalingam, Prabhat Ram, Praneeth Gottumukkala, Pratikkumar Patel, Rebecca Naughton, Reid Carper, Reza Zamani, Ricardo Bianchini, Rob Nertney, Rory Carmichael, Russell J. Hewett, Satyajit Desai, Scot Heath, Srini Varkala, Tom Li, Venkatesh Ramamurthy, Xin Li, Yu Pei, Zaid Kahn","submitted_at":"2025-08-20T00:04:06Z","abstract_excerpt":"Large Artificial Intelligence (AI) training workloads spanning several tens of thousands of GPUs present unique power management challenges. These arise due to the high variability in power consumption during the training. Given the synchronous nature of these jobs, during every iteration there is a computation-heavy phase, where each GPU works on the local data, and a communication-heavy phase where all the GPUs synchronize on the data. Because compute-heavy phases require much more power than communication phases, large power swings occur. The amplitude of these power swings is ever increasi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.14318","kind":"arxiv","version":2},"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/2508.14318/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":"2508.14318","created_at":"2026-07-05T11:57:02.788999+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.14318v2","created_at":"2026-07-05T11:57:02.788999+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.14318","created_at":"2026-07-05T11:57:02.788999+00:00"},{"alias_kind":"pith_short_12","alias_value":"T3BKNKPD76N7","created_at":"2026-07-05T11:57:02.788999+00:00"},{"alias_kind":"pith_short_16","alias_value":"T3BKNKPD76N7BO6Z","created_at":"2026-07-05T11:57:02.788999+00:00"},{"alias_kind":"pith_short_8","alias_value":"T3BKNKPD","created_at":"2026-07-05T11:57:02.788999+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25095","citing_title":"Toward Next-Generation AI Data Centers: Power Delivery Architecture Shifts, Emerging Technologies, and Challenges","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22096","citing_title":"A Pre-Dispatch Resonance Safety Criterion for AI Training Clusters","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21833","citing_title":"Inference as Flexibility: Ramp Management for Transmission-Connected AI Data Centres","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21536","citing_title":"Stability Enhancement of Centralized UPS Data Center Systems Under Weak-Grid Conditions","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09163","citing_title":"EnclaveScale: Hardware-Assisted Edge-DP for Secure Data Centre Power Telemetry","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04869","citing_title":"Source Side Mitigation of AI Datacenter Power Fluctuations with a Hybrid Energy Storage System and Residual Differentiable Predictive Control","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26384","citing_title":"GridPilot: Real-Time Grid-Responsive Control for AI Supercomputers","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04869","citing_title":"Source Side Mitigation of AI Datacenter Power Fluctuations with a Hybrid Energy Storage System and Residual Differentiable Predictive Control","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24326","citing_title":"ScaleAcross Explorer: Exploring Communication Optimization for Scale-Across AI Model Training","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24461","citing_title":"Provisioning to Runtime Optimization of a 100 MW-Scale AI Cluster","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17190","citing_title":"Replicating Real-World 23-Hz Oscillations Caused by Large Electronic Loads","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2602.10900","citing_title":"AI Infrastructure Sovereignty","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2603.00415","citing_title":"Grid Integration of AI Data Centers: A Critical Review of Energy Storage Solutions","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09593","citing_title":"Benchmarking Compound AI Applications for Hardware-Software Co-Design","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2603.20897","citing_title":"The data heat island effect: quantifying the impact of AI data centers in a warming world","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03090","citing_title":"From Barrier to Bridge: The Case for AI Data Center/Power Grid Co-Design","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00769","citing_title":"Voltage Ride-Through in Large Loads- A Dual PQ Approach","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07345","citing_title":"Measurement of Generative AI Workload Power Profiles for Whole-Facility Data Center Infrastructure Planning","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10769","citing_title":"Workload composition smooths aggregate power demand while sustaining short-horizon ramps in AI data centers","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15522","citing_title":"EasyRider: Mitigating Power Transients in Datacenter-Scale Training Workloads","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ","json":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ.json","graph_json":"https://pith.science/api/pith-number/T3BKNKPD76N7BO6Z6MFDBVE4HQ/graph.json","events_json":"https://pith.science/api/pith-number/T3BKNKPD76N7BO6Z6MFDBVE4HQ/events.json","paper":"https://pith.science/paper/T3BKNKPD"},"agent_actions":{"view_html":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ","download_json":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ.json","view_paper":"https://pith.science/paper/T3BKNKPD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.14318&json=true","fetch_graph":"https://pith.science/api/pith-number/T3BKNKPD76N7BO6Z6MFDBVE4HQ/graph.json","fetch_events":"https://pith.science/api/pith-number/T3BKNKPD76N7BO6Z6MFDBVE4HQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ/action/storage_attestation","attest_author":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ/action/author_attestation","sign_citation":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ/action/citation_signature","submit_replication":"https://pith.science/pith/T3BKNKPD76N7BO6Z6MFDBVE4HQ/action/replication_record"}},"created_at":"2026-07-05T11:57:02.788999+00:00","updated_at":"2026-07-05T11:57:02.788999+00:00"}