Repository navigation
Operations
Create a persistent signing key once with cargo run --release --locked -- keygen --output miner.key. Pass --key-file miner.key on subsequent starts. Use --wallet-address for the reward address. A new ephemeral signing identity on each restart can split worker statistics. The miner checks each job's network parameters and authorizes the reward keys in the candidate header.
Each accepted SOLO nonce must satisfy both the fixed neural gate and the block’s final BLAKE3 scratchpad threshold. Reward authorization binds the miner’s receiving keys.
CPU mining is the default, including when an accelerators directory exists next to the executable. Set --threads N to limit CPU workers. To build CUDA, run cmake -S native -B native/build -DHYPHEN_ENABLE_CUDA=ON followed by cmake --build native/build --config Release. Pass --accelerator-dir DIRECTORY to each GPU invocation. HYN3 uses a single host controller and runs the full search on the device. A missing, incompatible or failed plugin stops GPU mining with an error.
Each nonce passing the BLAKE3 gate uses 1 MiB of device scratchpad and runs 32,768 fill steps followed by 32,768 dependent reads. The plugin processes finalists in groups capped at 512 MiB of scratchpad and one quarter of reported free VRAM. --batch-size limits the number of candidate nonces per search window. Older network profiles support CPU mining. To run the hardware regression tests, set HYPHEN_TEST_ACCELERATOR_DIR to the plugin directory and run cargo test --release --locked native_neural_ -- --ignored --test-threads=1.
For GeForce GTX 1060 (Pascal, compute capability 6.1), configure a fresh build with CUDA 12.x. The default targets include native sm_61, sm_75, sm_86 code and compute_61 PTX. With Ninja, pass -DCMAKE_BUILD_TYPE=Release at configure time. CUDA 13 dropped offline compilation for Pascal. On the target machine, run hyphen-miner --accelerator-dir DIRECTORY accelerator-check --batch 32768 --timings to compare every score and winning nonce with Rust. Compare gpu_search_ns and cpu_search_ns for the complete searches over the same nonce window.
Run hyphen-miner --accelerator-dir DIRECTORY accelerator-check --offset 125 --batch 1 to reproduce the first-height HYN3 vector. The check compares every returned device score with Rust, applies the memory gate and prints scores=1 accepted=1 for this nonce. Reference vector
For an operator-provided SOCKS5 or HTTP CONNECT proxy use --via-proxy socks5://HOST:PORT or --via-proxy http://HOST:PORT. Set --pool to the destination pool address. Job changes invalidate outstanding work for that parent.
首次运行使用 keygen --output miner.key 生成签名密钥,此后用 --key-file miner.key 复用该身份,用 --wallet-address 指定奖励地址。默认使用 CPU,可通过 --threads 限制线程数。即使程序旁边已有 accelerators 文件夹,也要显式传入 --accelerator-dir DIRECTORY 才会启用 GPU。HYN3 的神经评分和每个 nonce 的全部 BLAKE3 计算均在显卡完成,每个通过初筛的 nonce 使用 1 MiB 显存。插件缺失、版本不兼容或运行失败时,矿工报错退出。
Hyphen documentation · Protocol and node · Miner · Pool · Wallet · Security reporting