nx_gpu_power.nx

nx_gpu_power.nx -- sovereign GPU POWER + THERMAL GOVERNOR (operator 2026-06-20: "fix the energy and heat output if we need to as nvidia fries their gpus"). NEVER-FRY BY CONSTRUCTION (rule #26) + resource-aware (rule #21): clamps the power cap to the hardware [min,max], DERATES under thermal stress, and EMERGENCY-caps to minimum at the critical temperature. Also yields the LOCKED CLOCKS the benchmark methodology needs (reproducible measurement). The governor LOGIC is pure INTEGER (no-float, deterministic) -- gated locally. APPLYING the cap = the device shim's RM_CONTROL power-limit command (nx_nv on native-Linux, RM NV2080_CTRL_CMD_PERF; or nvidia-smi -pl as the incumbent, admin) = TODO_DEVICE. Real 5080 range probed: pmin=5W, pmax=175W, default=80W. Pure funcs, no main. license_tier: ORIGINAL