New: --wait-ready
For rigs where model bring-up is slow — typically low-RAM hosts with many cards running --low-ram — the miner previously declared each model to the pool the moment the first card finished loading. The pool would immediately start routing OPoI challenges while the remaining cards were still staging, so those challenges ran on a starved host, missed their deadlines, collected inference strikes, and could stall the rest of the bring-up entirely.
With --wait-ready, the miner neither mines nor declares any OPoI capability until every card is fully set up (model staged + walk table installed). Until then the pool has nothing to challenge and staging gets the host to itself. The gate opens once — later model swaps or fault recoveries never re-suspend a working rig.
- Progress is logged every 30 s (
2/4 cards ready (waiting on [2, 3])…), each card logs when it becomes ready, and the gate announces itself when it opens. - Safety valve: if a card can never finish (unsupported tier, dead card), mining starts anyway after
KERYX_WAIT_READY_TIMEOUT_SECS(default 2700 = 45 min) with whichever cards are ready, and the log names the stragglers. - Recommended for low-RAM rigs:
--low-ram --wait-readytogether.
Live-validated on a 4-GPU mixed rig: first pool declaration goes out the exact second the gate opens; 4 accepted / 0 rejected within the first two minutes of mining.
Includes everything from v0.11.17 (chaseless PoM solver, +8-14% on GDDR cards).