{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:LEGVSBB7P6ECT3OCEHTXUUVQRI","short_pith_number":"pith:LEGVSBB7","canonical_record":{"source":{"id":"2505.06461","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-09T23:05:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ca7eff2a0a9c122d7c8cd9d3d84687d21d3ddf2de1d1a13fd5e20148c267cc04","abstract_canon_sha256":"da5c4c739f76395c112d76f302ba3d9dbe22d1eb75b4e4ab3da7375c0815a94b"},"schema_version":"1.0"},"canonical_sha256":"590d59043f7f8829edc221e77a52b08a1624f831956896a50a561f0783fe7c1d","source":{"kind":"arxiv","id":"2505.06461","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.06461","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"arxiv_version","alias_value":"2505.06461v1","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.06461","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_12","alias_value":"LEGVSBB7P6EC","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_16","alias_value":"LEGVSBB7P6ECT3OC","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_8","alias_value":"LEGVSBB7","created_at":"2026-07-05T11:01:09Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:LEGVSBB7P6ECT3OCEHTXUUVQRI","target":"record","payload":{"canonical_record":{"source":{"id":"2505.06461","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-09T23:05:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ca7eff2a0a9c122d7c8cd9d3d84687d21d3ddf2de1d1a13fd5e20148c267cc04","abstract_canon_sha256":"da5c4c739f76395c112d76f302ba3d9dbe22d1eb75b4e4ab3da7375c0815a94b"},"schema_version":"1.0"},"canonical_sha256":"590d59043f7f8829edc221e77a52b08a1624f831956896a50a561f0783fe7c1d","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:01:09.991583Z","signature_b64":"a8uMewDAGa2lpGC/Fzan10Yamx6WXdOsm2mCLZrHXS7VByGcbEv2ncohCb8JfqYtyZXWbqI6S7NAkUWjTLuWBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"590d59043f7f8829edc221e77a52b08a1624f831956896a50a561f0783fe7c1d","last_reissued_at":"2026-07-05T11:01:09.991104Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:01:09.991104Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2505.06461","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:01:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1moUx1d0E8BRmIjr3CwtOybq7834R2RKs1Qlg6ImYohqmEH9LGD/SMjFW94SXN2dv8Z46uJaMdxlgrSDw9k3Ag==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T22:52:13.779824Z"},"content_sha256":"c4bdaf2a369e8992dc8e50b0784e523a1d6f8f0b4ea78b7313d3d8ef2aac5ebe","schema_version":"1.0","event_id":"sha256:c4bdaf2a369e8992dc8e50b0784e523a1d6f8f0b4ea78b7313d3d8ef2aac5ebe"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:LEGVSBB7P6ECT3OCEHTXUUVQRI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Challenging GPU Dominance: When CPUs Outperform for On-Device LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Haolin Zhang, Jeff Huang","submitted_at":"2025-05-09T23:05:53Z","abstract_excerpt":"The common assumption in on-device AI is that GPUs, with their superior parallel processing, always provide the best performance for large language model (LLM) inference. In this work, we challenge this notion by empirically demonstrating that, under certain conditions, CPUs can outperform GPUs for LLM inference on mobile devices. Using a 1-billion-parameter LLM deployed via llama.cpp on the iPhone 15 Pro, we show that a CPU-only configuration (two threads, F16 precision) achieves 17 tokens per second, surpassing the 12.8 tokens per second obtained with GPU acceleration. We analyze the archite"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.06461","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/2505.06461/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:01:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"f+2xRJ/gfH8D+R3P8GDaBJErm3DPH3e0iddYd3I+KuPxSJNG7/n/qetd7j/Ocj1SSMmgBkDZrIqae+0dDVzpAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T22:52:13.780198Z"},"content_sha256":"224a80786a3484f9350a5213dd482fe70c5fde1291a34bdf592ed6e21830cfdd","schema_version":"1.0","event_id":"sha256:224a80786a3484f9350a5213dd482fe70c5fde1291a34bdf592ed6e21830cfdd"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/bundle.json","state_url":"https://pith.science/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-09T22:52:13Z","links":{"resolver":"https://pith.science/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI","bundle":"https://pith.science/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/bundle.json","state":"https://pith.science/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/LEGVSBB7P6ECT3OCEHTXUUVQRI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:LEGVSBB7P6ECT3OCEHTXUUVQRI","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"da5c4c739f76395c112d76f302ba3d9dbe22d1eb75b4e4ab3da7375c0815a94b","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-09T23:05:53Z","title_canon_sha256":"ca7eff2a0a9c122d7c8cd9d3d84687d21d3ddf2de1d1a13fd5e20148c267cc04"},"schema_version":"1.0","source":{"id":"2505.06461","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.06461","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"arxiv_version","alias_value":"2505.06461v1","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.06461","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_12","alias_value":"LEGVSBB7P6EC","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_16","alias_value":"LEGVSBB7P6ECT3OC","created_at":"2026-07-05T11:01:09Z"},{"alias_kind":"pith_short_8","alias_value":"LEGVSBB7","created_at":"2026-07-05T11:01:09Z"}],"graph_snapshots":[{"event_id":"sha256:224a80786a3484f9350a5213dd482fe70c5fde1291a34bdf592ed6e21830cfdd","target":"graph","created_at":"2026-07-05T11:01:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2505.06461/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"The common assumption in on-device AI is that GPUs, with their superior parallel processing, always provide the best performance for large language model (LLM) inference. In this work, we challenge this notion by empirically demonstrating that, under certain conditions, CPUs can outperform GPUs for LLM inference on mobile devices. Using a 1-billion-parameter LLM deployed via llama.cpp on the iPhone 15 Pro, we show that a CPU-only configuration (two threads, F16 precision) achieves 17 tokens per second, surpassing the 12.8 tokens per second obtained with GPU acceleration. We analyze the archite","authors_text":"Haolin Zhang, Jeff Huang","cross_cats":["cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-09T23:05:53Z","title":"Challenging GPU Dominance: When CPUs Outperform for On-Device LLM Inference"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.06461","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c4bdaf2a369e8992dc8e50b0784e523a1d6f8f0b4ea78b7313d3d8ef2aac5ebe","target":"record","created_at":"2026-07-05T11:01:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"da5c4c739f76395c112d76f302ba3d9dbe22d1eb75b4e4ab3da7375c0815a94b","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2025-05-09T23:05:53Z","title_canon_sha256":"ca7eff2a0a9c122d7c8cd9d3d84687d21d3ddf2de1d1a13fd5e20148c267cc04"},"schema_version":"1.0","source":{"id":"2505.06461","kind":"arxiv","version":1}},"canonical_sha256":"590d59043f7f8829edc221e77a52b08a1624f831956896a50a561f0783fe7c1d","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"590d59043f7f8829edc221e77a52b08a1624f831956896a50a561f0783fe7c1d","first_computed_at":"2026-07-05T11:01:09.991104Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:01:09.991104Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"a8uMewDAGa2lpGC/Fzan10Yamx6WXdOsm2mCLZrHXS7VByGcbEv2ncohCb8JfqYtyZXWbqI6S7NAkUWjTLuWBA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:01:09.991583Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.06461","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c4bdaf2a369e8992dc8e50b0784e523a1d6f8f0b4ea78b7313d3d8ef2aac5ebe","sha256:224a80786a3484f9350a5213dd482fe70c5fde1291a34bdf592ed6e21830cfdd"],"state_sha256":"0a641b1d216ffdc20b9a5276d4f3278c4c6cd3778c7172753a98c38f51c45238"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"cRIZoU9Rm95SXivZe6enSsdxfh3UcjnDBdSf9hjVr2gHs3WD+6rRngY1Xc02759ocQwCDEA1FnBsgQO4wJMICQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-09T22:52:13.782091Z","bundle_sha256":"8058947b7d80c6d2656b43700f9f4cf4db335e31e03d406e374ebac3381221ce"}}