A brief guide to the basic arguments of Krig
Krig doesn't expose overclocking flags — manage clocks/power/fans with your GPU vendor's own tools. By default it runs both backends and mines every NVIDIA and AMD GPU it finds; through arguments you can pick a vendor, choose which GPUs mine, and expose stats.
--no-cuda — disable the CUDA backend (skip NVIDIA GPUs) — use it on an AMD-only rig.
--no-rocm — disable the ROCm backend (skip AMD GPUs) — use it on an NVIDIA-only rig.
--amd-igpu — also include AMD integrated GPUs (APU/iGPU), which are excluded by default.
-d 0,2 — mine only on GPU indices 0 and 2 (0-based, matches --list-devices order). Default is all.
--devices-pci 01:00.0,0a:00.0 — same as -d, but selecting GPUs by PCI bus address instead of index — useful on rigs where GPU order isn't stable across reboots.
--list-devices — prints every detected GPU (index, name, PCI address) and exits, without starting to mine.
--api-port 12000 — exposes the miner's stats API and a Prometheus /metrics endpoint on the given port.
--api-host 127.0.0.1 — restricts the stats API to localhost instead of all interfaces (default).
If you have multiple arguments, list them in a line, separated by spaces. For example, mine only on the first two GPUs and expose stats locally on port 12000:
-d 0,1 --api-port 12000 --api-host 127.0.0.1
This means: mining restricted to GPU 0 and GPU 1, with the stats API reachable only from the local machine on port 12000.
Check which GPU is which before selecting indices:
--list-devices
