F@H Daily Fleet Report (September 28, 2026 PDT)

1. Peak Performance Metrics (Top 10 Peaks)

2. Work Unit Completions (24-Hour Window: 08:00 to 08:00 PDT)

*Per-WU power coverage: 31/31 completions (100.0%). Rows without a measured draw show —; they are excluded from efficiency figures rather than back-filled with the cap.*

2.1 Silicon Lottery — All-Time Project-Normalized Rankings

Ranked over all recorded history, not this window, and normalized per Folding@Home project. Raw cumulative PPD/W is not comparable between two cards of the same model: the same silicon measures ~33k PPD/W on P15239 and ~57k PPD/W on P13015, so a card that happened to be assigned more of the high-scoring projects "wins" on assignment luck alone. A project therefore enters a model family's comparison pool only when every ranked card has 10+ completions on it, and each card is scored as the WU-weighted mean of its per-project median PPD/W measured against that project's class median. The median (not the mean) is used per project because F@H logs a resumed checkpoint as one WU at a post-resume burst rate — a single real row reads 782,723 PPD/W, 18x physical reality. Cards must still clear a lifetime sample gate of 25% of the class's median WU count — counting live cards only — floored at 10 and capped at 200. 🏆 requires both ≥ +1.5% overall and beating the class median on ≥ 67% of the shared projects; a leader inside that band is marked *within noise*. Only WUs carrying both a physical GPU binding and a measured power draw are counted; unresolved WUs are dropped, never guessed at.

• RTX 2060 — 1 card(s), 204 lifetime WUs with measured power:
• *Only one card of this model has ever folded here; there is nothing to rank it against.*
• scraps (535ba65c): 21,339 PPD/W across 204 WUs *(unranked)*
• RTX 2060 SUPER — 1 card(s), 228 lifetime WUs with measured power:
• *Only one card of this model has ever folded here; there is nothing to rank it against.*
• scraps (8990101d): 22,481 PPD/W across 228 WUs *(unranked)*
• RTX 3060 Ti — 3 card(s), 445 lifetime WUs with measured power:
• *Project-normalized baseline across 4 shared project(s) (P12138, P15240, P18246, P18247), 308 like-for-like WUs: class median 18,505 PPD/W. Each card is scored as the WU-weighted mean of its per-project median PPD/W measured against that project's class median, so a card cannot win on assignment luck. Sample gate 37 lifetime WUs, 3/3 cards ranked; a 🏆 requires ≥ +1.5% overall AND beating the class median on ≥ 3/4 shared projects.*
• scraps (d1ea526f): +0.48% normalized (18,593 PPD/W over 106 shared WUs, won 3/4 projects) — lifetime 18,770 PPD/W raw across 156 WUs 🏆 *(Class Leader — within noise)*
• scraps (a76d84a0): +0.05% normalized (18,515 PPD/W over 102 shared WUs, won 1/4 projects) — lifetime 18,313 PPD/W raw across 147 WUs
• scraps (1ed25c7d): -2.58% normalized (18,028 PPD/W over 100 shared WUs, won 0/4 projects) — lifetime 17,986 PPD/W raw across 142 WUs 🐌 *(Silicon Laggard)*
• *Leadership unchanged since 2026-07-28: scraps a76d84a0 at +0.31% normalized (19,718 PPD/W).*
• *Title held by a76d84a0: 🏆 scraps d1ea526f now tops the standings at +0.48% vs +0.05%, but has not cleared the ±1.5% debounce margin, so the lead is not recorded as having changed hands.*
• RTX 4070 — 1 card(s), 245 lifetime WUs with measured power:
• *Only one card of this model has ever folded here; there is nothing to rank it against.*
• smallpieces (d65f3411): 42,625 PPD/W across 245 WUs *(unranked)*
• RTX 4090 — 1 card(s), 321 lifetime WUs with measured power:
• *Only one card of this model has ever folded here; there is nothing to rank it against.*
• itztli (212cc188): 88,289 PPD/W across 321 WUs *(unranked)*
• RTX 5060 Ti — 4 card(s), 908 lifetime WUs with measured power:
• *Project-normalized baseline across 5 shared project(s) (P13015, P15239, P18246, P18247, P18264), 808 like-for-like WUs: class median 43,859 PPD/W. Each card is scored as the WU-weighted mean of its per-project median PPD/W measured against that project's class median, so a card cannot win on assignment luck. Sample gate 57 lifetime WUs, 4/4 cards ranked; a 🏆 requires ≥ +1.5% overall AND beating the class median on ≥ 3/5 shared projects.*
• smallpieces (8854baa9): +0.96% normalized (44,279 PPD/W over 191 shared WUs, won 3/5 projects) — lifetime 41,749 PPD/W raw across 219 WUs 🏆 *(Class Leader — within noise)*
• smallpieces (56d10701): +0.79% normalized (44,206 PPD/W over 211 shared WUs, won 4/5 projects) — lifetime 45,982 PPD/W raw across 232 WUs
• smallpieces (2e4a1d96): -0.19% normalized (43,774 PPD/W over 200 shared WUs, won 3/5 projects) — lifetime 42,973 PPD/W raw across 228 WUs
• smallpieces (a83e55f9): -4.14% normalized (42,044 PPD/W over 206 shared WUs, won 0/5 projects) — lifetime 41,885 PPD/W raw across 229 WUs 🐌 *(Silicon Laggard)*
• *Leadership history (most recent 3, newest first):*
• 2026-07-23 — 🏆 smallpieces 8854baa9 took the lead at +0.82% normalized (51,405 PPD/W) (from a83e55f9 at -1.60% / 50,171 PPD/W) — *new silicon installed*
• 2026-07-12 — smallpieces a83e55f9 took the lead at +7.79% normalized (56,964 PPD/W) (from 56d10701 at +0.00% / 52,849 PPD/W) — *new silicon installed*
• 2026-07-11 — smallpieces 56d10701 took the lead at +5.14% normalized (57,999 PPD/W) — *first card to clear the class sample gate*
• RTX 5070 — 3 card(s), 1,044 lifetime WUs with measured power:
• *Project-normalized baseline across 4 shared project(s) (P13015, P18246, P18247, P18264), 951 like-for-like WUs: class median 70,669 PPD/W. Each card is scored as the WU-weighted mean of its per-project median PPD/W measured against that project's class median, so a card cannot win on assignment luck. Sample gate 91 lifetime WUs, 3/3 cards ranked; a 🏆 requires ≥ +1.5% overall AND beating the class median on ≥ 3/4 shared projects.*
• origami (dd40d876): +2.18% normalized (72,213 PPD/W over 401 shared WUs, won 3/4 projects) — lifetime 67,827 PPD/W raw across 436 WUs 🏆 *(Silicon Winner)*
• origami (18de45ee) *(retired)*: +0.26% normalized (70,854 PPD/W over 311 shared WUs, won 1/4 projects) — lifetime 67,294 PPD/W raw across 315 WUs
• smallpieces (d87d48a0): -16.64% normalized (58,909 PPD/W over 239 shared WUs, won 0/4 projects) — lifetime 56,537 PPD/W raw across 293 WUs 🐌 *(Silicon Laggard)*
• *Leadership unchanged since 2026-07-10: 🏆 origami dd40d876 at +0.75% normalized (77,394 PPD/W).*
• *Cross-host architecture note:* The smallpieces RTX 5070 (d87d48a0) comparison against origami (EPYC 7443P / ASRock Rack ROMED8-2T, PCIe Gen 4 x16) is not apples-to-apples because smallpieces v2 (Core i7-7700 4C/8T) is currently CPU-starved feeding 6 GPUs. Project normalization removes the *science* variable but not the *host* variable. The planned scraps v2 / smallpieces v3 transition (12-core Zen 4/5 with PCIe Gen 4 x2/x4 lanes) will create an unstarved host environment for a cleaner cross-host comparison against origami's EPYC ROMED8-2T.
• RTX PRO 4000 Blackwell — 2 card(s), 541 lifetime WUs with measured power:
• *Project-normalized baseline across 4 shared project(s) (P15239, P15240, P15241, P18264), 252 like-for-like WUs: class median 85,183 PPD/W. Each card is scored as the WU-weighted mean of its per-project median PPD/W measured against that project's class median, so a card cannot win on assignment luck. Sample gate 68 lifetime WUs, 2/2 cards ranked; a 🏆 requires ≥ +1.5% overall AND beating the class median on ≥ 3/4 shared projects.*
• origami (e1600107): +0.45% normalized (85,568 PPD/W over 108 shared WUs, won 4/4 projects) — lifetime 79,301 PPD/W raw across 111 WUs 🏆 *(Class Leader — within noise)*
• origami (6f99ccc4): -0.47% normalized (84,786 PPD/W over 144 shared WUs, won 0/4 projects) — lifetime 83,899 PPD/W raw across 430 WUs
• *Leadership unchanged since 2026-09-16: 🏆 origami e1600107 at +0.68% normalized (86,834 PPD/W).*

3. System & Service Health

3.0 Work Unit Reconciliation (F@H Authoritative)

• Local (deduplicated): 22,972 WUs vs F@H authoritative: 23,311 WUs (98.55% coverage).
• Per host: o 15,286, sp 5,625, i 1,191, s 870
• Global rank: #423 | Score: 13,754,956,783 | Active clients (7d): 4
• 🟢 Coverage nominal. The standing shortfall predates the oldest retained client.db and is accepted by design.

3.1 F@H Client Warning & Error Summary

None

3.2 fah-runaway Actions

None

3.3 PCIe Bus Disconnections

None

3.4 Manual & System Restarts

None

4. Combined Live PPD Timeline (24-Hour Window: 08:00 to 08:00 PDT)

*(Note: Transient 0-rate socket poll timeouts are forward-filled with the last known valid PPD rate)*