{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UL66UQ3UF6AU7PTZJXTXGD7JJV","short_pith_number":"pith:UL66UQ3U","schema_version":"1.0","canonical_sha256":"a2fdea43742f814fbe794de7730fe94d6dea8f59a1c14127c8fc3517e786a04b","source":{"kind":"arxiv","id":"2502.00722","version":2},"attestation_state":"computed","paper":{"title":"Demystifying Cost-Efficiency in LLM Serving over Heterogeneous GPUs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Ana Klimovic, Bin Cui, Binhang Yuan, Eiko Yoneki, Fangcheng Fu, Guoliang He, Xiaozhe Yao, Xupeng Miao, Youhe Jiang","submitted_at":"2025-02-02T08:44:43Z","abstract_excerpt":"Recent advancements in Large Language Models (LLMs) have led to increasingly diverse requests, accompanied with varying resource (compute and memory) demands to serve them. However, this in turn degrades the cost-efficiency of LLM serving as common practices primarily rely on homogeneous GPU resources. In response to this problem, this work conducts a thorough study about serving LLMs over heterogeneous GPU resources on cloud platforms. The rationale is that different GPU types exhibit distinct compute and memory characteristics, aligning well with the divergent resource demands of diverse req"},"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":"2502.00722","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2025-02-02T08:44:43Z","cross_cats_sorted":[],"title_canon_sha256":"45247125abb06b3be4460e9e1ae267b1924a2c092f603cfcd423acdbe25368fc","abstract_canon_sha256":"d933e37301f592738aafd4a52d5d915bd849de351ce1af96f77a4a361d603f94"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:05.757502Z","signature_b64":"yqFL3x9go+HmGEB3cY9i2td+efV6RcMg0lSLmyVQ+qgAhYynHhJqf0I6efMC4NkPo4Scg6wGlQIjKc/uk26sDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a2fdea43742f814fbe794de7730fe94d6dea8f59a1c14127c8fc3517e786a04b","last_reissued_at":"2026-07-05T11:16:05.756970Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:05.756970Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Demystifying Cost-Efficiency in LLM Serving over Heterogeneous GPUs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Ana Klimovic, Bin Cui, Binhang Yuan, Eiko Yoneki, Fangcheng Fu, Guoliang He, Xiaozhe Yao, Xupeng Miao, Youhe Jiang","submitted_at":"2025-02-02T08:44:43Z","abstract_excerpt":"Recent advancements in Large Language Models (LLMs) have led to increasingly diverse requests, accompanied with varying resource (compute and memory) demands to serve them. However, this in turn degrades the cost-efficiency of LLM serving as common practices primarily rely on homogeneous GPU resources. In response to this problem, this work conducts a thorough study about serving LLMs over heterogeneous GPU resources on cloud platforms. The rationale is that different GPU types exhibit distinct compute and memory characteristics, aligning well with the divergent resource demands of diverse req"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.00722","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/2502.00722/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":"2502.00722","created_at":"2026-07-05T11:16:05.757038+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.00722v2","created_at":"2026-07-05T11:16:05.757038+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.00722","created_at":"2026-07-05T11:16:05.757038+00:00"},{"alias_kind":"pith_short_12","alias_value":"UL66UQ3UF6AU","created_at":"2026-07-05T11:16:05.757038+00:00"},{"alias_kind":"pith_short_16","alias_value":"UL66UQ3UF6AU7PTZ","created_at":"2026-07-05T11:16:05.757038+00:00"},{"alias_kind":"pith_short_8","alias_value":"UL66UQ3U","created_at":"2026-07-05T11:16:05.757038+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19271","citing_title":"TurboServe: Serving Streaming Video Generation Efficiently and Economically","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2409.01143","citing_title":"HexiScale: Facilitating Large Language Model Training over Heterogeneous Hardware","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16637","citing_title":"HexAGenT: Efficient Agentic LLM Serving via Workflow- and Heterogeneity-Aware Scheduling","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16867","citing_title":"GoodServe: Towards High-Goodput Serving of Agentic LLM Inferences over Heterogeneous Resources","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00831","citing_title":"GhostServe: A Lightweight Checkpointing System in the Shadow for Fault-Tolerant LLM Serving","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11335","citing_title":"ChunkFlow: Communication-Aware Chunked Prefetching for Layerwise Offloading in Distributed Diffusion Transformer Inference","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04357","citing_title":"Coral: Cost-Efficient Multi-LLM Serving over Heterogeneous Cloud GPUs","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07144","citing_title":"Autopoiesis: A Self-Evolving System Paradigm for LLM Serving Under Runtime Dynamics","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07569","citing_title":"HexiSeq: Accommodating Long Context Training of LLMs over Heterogeneous Hardware","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05227","citing_title":"Rethinking Data Curation in LLM Training: Online Reweighting Offers Better Generalization than Offline Methods","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV","json":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV.json","graph_json":"https://pith.science/api/pith-number/UL66UQ3UF6AU7PTZJXTXGD7JJV/graph.json","events_json":"https://pith.science/api/pith-number/UL66UQ3UF6AU7PTZJXTXGD7JJV/events.json","paper":"https://pith.science/paper/UL66UQ3U"},"agent_actions":{"view_html":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV","download_json":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV.json","view_paper":"https://pith.science/paper/UL66UQ3U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.00722&json=true","fetch_graph":"https://pith.science/api/pith-number/UL66UQ3UF6AU7PTZJXTXGD7JJV/graph.json","fetch_events":"https://pith.science/api/pith-number/UL66UQ3UF6AU7PTZJXTXGD7JJV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV/action/storage_attestation","attest_author":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV/action/author_attestation","sign_citation":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV/action/citation_signature","submit_replication":"https://pith.science/pith/UL66UQ3UF6AU7PTZJXTXGD7JJV/action/replication_record"}},"created_at":"2026-07-05T11:16:05.757038+00:00","updated_at":"2026-07-05T11:16:05.757038+00:00"}