Temperatures
Hardware monitoring is on by default (--no-hwmon turns it off). Hashwarp reads NVIDIA GPUs through the NVIDIA driver, and AMD GPUs on Linux through the kernel driver: core, junction and memory temperatures, fan, power and clocks. On Windows, AMD GPUs older than Vega report their temperature, fan and power; newer AMD GPUs report nothing on Windows.
The temperature limits take one value per GPU, or one for all. A paused GPU is logged with a warning, and resumes by itself once it has cooled down:
--mem-temp-stop needs a GPU that reports its memory or hotspot temperature. Most AMD GPUs on Linux report both, but NVIDIA’s driver doesn’t report them for many GeForce cards. When a GPU reports neither, Hashwarp warns that --mem-temp-stop has no effect on it. The statistics table (s) shows which temperatures each GPU reports.The watchdog
The watchdog restarts a GPU whose mining failed, and stops Hashwarp when a GPU hangs, so that a supervisor can start it again:
A hung GPU can’t be recovered by Hashwarp: the watchdog exits so that Hashwarp, or the whole rig, can be restarted. Run Hashwarp under a supervisor that starts it again: HiveOS or mmpOS, a systemd service, or a loop in a script. Overclocks are restored on a watchdog exit.
Checked results
Every solution a GPU finds is checked on the CPU before it is sent. A wrong one, usually from too high a memory overclock, isn’t submitted: it counts asF (failed) in the share counts and is logged:
hashwarp --selftest runs the same check offline and says PASS or FAIL per GPU: see GPUs & Tuning.