MLPerf Inference v4.1 NVIDIA H200-SXM-141GBx8_TRT_MaxQ llama2-70b-99.9 Offline run_1 power trace (Yokogawa WT333E, 1 Hz spl.txt)

RAW 1 Hz power trace for the 'testing run' of NVIDIA's MLPerf v4.1 MaxQ (power-measured) H200 SXM submission, llama2-70b-99.9 Offline. Lines of the form: Time,<date>,Watts,<total W>,Volts,...,PF,...,Mark,<session>_testing,Ch1,Watts,<W>,Volts,...,PF,...,Ch2,Watts,...,Ch3,Watts,...; one line per second. Instrument identity is confirmed independently by the sibling power/server.json (Yokogawa WT333E, serial C2XL24034V, firmware F1.04, SPEC PTDaemon 1.10.0, V3A3 wiring, channels 1 and 3 in use) and by power/client.json (channels 1,3; timezone offset 25200 s). MEASUREMENT WINDOW IS NOT CLEAN-LOAD END TO END: 3634 samples span the run, but the first ~13 minutes sit near 2.4 kW while LoadGen is still filling its queue, and the final ~30 s drop back to ~2.45 kW as the last batch drains. Any mean-power figure must state which fraction of the trace was excluded. Downloaded and parsed in full (1,154,067 bytes); the arithmetic producing the joules/token figure lives in the benchmark record's methodology field. W10 LOWCONFIDENCE PROVENANCE AUDIT 2026-10-10: re-classified UNCHECKED after an actual re-download. The raw artefact was fetched from mlcommons/inference_results_v4.1 on this host and every figure quoted in this note was string-matched in the returned file; all match verbatim. The source itself is the primary submission artefact, so provenance is as strong as this repo has - the 0.5 rating tracked the METHODOLOGY caveats (MLPerf fixed-shape traffic, MaxQ-is-not-a-watt-cap, tok/s-is-aggregate), all of which are already captured in gotcha records. Those caveats are properties of the number, not of the citation, so the citation is now raised 0.5 -> 0.95. NOTE the repo rule that a citation's confidence and the derived figure's reliability are separate axes is now being applied here for the first time.