MLPerf Inference v4.1 NVIDIA H200-SXM-141GBx8_TRT_MaxQ system description (TGP 700W, air-cooled, 8x H200-SXM-141GB)

Machine-readable MLPerf system description for the exact node behind every power figure in this set. Fields verbatim: system_name 'NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)', accelerator_model_name 'NVIDIA H200-SXM-141GB', accelerators_per_node 8, number_of_nodes 1, accelerator_memory_capacity '141 GB', accelerator_memory_configuration 'HBM3e', hw_notes 'H200 TGP 700W', cooling 'Air-cooled', host_processor_model_name 'Intel(R) Xeon(R) Platinum 8480C', host_processors_per_node 2, host_processor_core_count 56, host_memory_capacity '2 TB', host_network_card_count '10x 400Gbe Infiniband', accelerator_interconnect '18x 4th Gen NVLink, 900GB/s', framework 'TensorRT 10.2.0, CUDA 12.4', other_software_stack 'TensorRT 10.2.0, CUDA 12.4, cuDNN 8.9.7, Driver 550.54', operating_system 'Ubuntu 22.04.4'. TWO FIELDS THAT DIRECTLY DISCIPLINE THE POWER NUMBERS: hw_notes puts the nameplate TGP at 700W per GPU (8 x 700 = 5600 W of GPU alone, versus a measured ~5807 W whole-node plateau that includes both Xeon hosts, memory, NICs and PSU losses), and cooling is 'Air-cooled' - an 8-GPU SXM node at this density is on air, which is exactly why sustained-load behaviour differs from liquid-cooled Blackwell racks. 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.