keryx-miner-supr v0.6.7.0 — H3 hardfork (REQUIRED before DAA 43,450,000)
⚠️ Mandatory update. Keryx activates the H3 consensus hardfork at DAA score 43,450,000 (~2026-07-05 18:00 UTC). At the gate the Proof-of-Model folds change (the pre_pow_hash words feeding the walk are salted). A pre-H3 miner's shares are rejected after the gate — every miner must be on v0.6.7.0 (or newer) before then.
Just update and leave it running — it switches itself
This build auto-detects the fork and switches automatically. The H3 convention is decided per job from the block's DAA score, so:
- Before the gate it mines exactly as today (no change).
- At DAA 43,450,000 it flips to the post-fork convention on the next job — no restart, no reconnect, no re-download (the change only affects the walk's seed/pow folds, not the model weights).
- It logs the switch:
H3 hardfork ACTIVE … auto-switched to the post-fork PoM convention.
So you can update now and leave it mining across the fork.
What changed
- H3 salt (
POM_H3_PPH_SALT) applied host-side to the PoM folds at/after the gate — byte-exact tokeryx-nodev1.3.1. GPU kernels unchanged. - Gate
POM_LEVEL_ACTIVATION_DAA = 43,450,000(envKERYX_POM_LEVEL_ACTIVATION_DAAfor testing only). - No stratum/wire change —
mining.submitis unchanged.
Downloads — pick your line
Three NVIDIA build lines, each with its own bundled CUDA runtime (driver floor):
- modern — CUDA 12.9, driver R575+, native Blackwell (sm_120). 50-series / newest drivers.
- legacy — CUDA 12.4, driver R550+. Turing/Ampere/Ada on older drivers (forward-JITs to Blackwell).
- pascal — CUDA 12.4, sm_60. GTX 10-series / P100 (
--cpu-inference --light).
Each in HiveOS / SMOS / mmpOS / Linux formats; verify with the matching SHA256SUMS-<line>.txt. Windows zips (AMD, NVIDIA-PoM, keryxcuda) are attached automatically by CI.
Version label 0.6.7.0; pool user-agent bumped to 0.6.7 so pools can require H3-capable miners after the gate. (A macOS / Apple-Silicon build is coming as a follow-up asset.)