ocminer/keryx-miner-supr v0.9.7

on GitHub

v0.9.7 — per-GPU efficiency stats

Field request implemented: the per-device hashrate line now carries the GPU health and an efficiency figure on ONE line, so tuning for efficiency no longer means correlating two log lines:

Device #0 (NVIDIA RTX A4000): 18.87 Mhash/s | 62°C fan=99% 114.8W core=1530MHz mem=8001MHz vram=8084/16376MB | eff 0.164 MH/s/W
Rig efficiency: 245.2 W total, 0.214 MH/s/W

• The NVML sample is taken in the same reporting tick as the hashrate, so the MH/s/W pairs a rate with the wattage that produced it.
• Safe on every rig: if NVML is unavailable (AMD/CPU), a sensor is missing (e.g. CMP fan), or the CUDA↔NVML device mapping can't be verified by name, the affected fields are simply omitted — never guessed.
• Stats API (--api-bind, / and /stats): devices[] now includes temp_c, fan_pct, power_w, core_mhz, mem_mhz, vram_used_mb, vram_total_mb, efficiency_mhs_per_w, plus top-level total_power_w and total_efficiency_mhs_per_w. Additive — existing consumers and /mmpos are unchanged.

Also included since v0.9.6's original cut (shipped as v0.9.6 repacks, listed here for visibility): the solo safety gate that HOLDS escrow claims and warns when keryxd < 1.4.4 (pre-1.4.4 nodes are consensus-diverged and reject blocks carrying claims — upgrade your node), per-GPU transient-fault recovery, batched escrow claims with boot-time state validation, GGUF download hardening, and the OPoI repetition penalty.

Linux NVIDIA assets attached (modern sm_75+/legacy sm_70+/pascal sm_60 × HiveOS/SMOS/mmpOS/linux + SHA256SUMS). Windows/macOS follow via CI; AMD assets follow from the AMD line.