# ProxiChorizo — how to mine Proxima (PROX) with it ProxiChorizo is a CUDA miner ("nonce seeker") for the Proxima fair-launch mine chain. It works with a **pool** (the pool runs the wallet side and pays you per share) or **solo** against your own Proxima node and `proxi node mine`. One NVIDIA GPU of the Turing, Ampere, Ada or Blackwell generation is enough; several GPUs in one rig are fine. Requirements: NVIDIA driver 550 or newer (CUDA 12.x capable), Linux x64 or Windows 10/11 x64. Nothing else is installed by the miner. --- ## 1. Pool mining — example: suprnova.cc The pool is https://proxi.suprnova.cc. A Proxima transit is signed with the key that mined it, so a pool cannot let you mine with your own wallet key: at registration it creates a **mining key** for your account and gives you a **token**. Rewards are credited to your PROX address by the pool (PPLNS, 1 % fee) and paid out to it. 1. Open https://proxi.suprnova.cc/StartMining.html and register with your **PROX address** (the 64-hex holder ID of your wallet, shown by `proxi node balance` as `a/`; type it without the `a/`). 2. The page gives you a **token** and a **key file**. Save the key file as `proxima.key` next to the miner. This key can only mine for your account; it holds no funds. 3. Start the miner: Linux ```bash ./proxichorizo --proxi http://proxi.suprnova.cc:8100 --token YOUR_TOKEN --key-file proxima.key --name rig1 ``` Windows ```bat proxichorizo.exe --proxi http://proxi.suprnova.cc:8100 --token YOUR_TOKEN --key-file proxima.key --name rig1 ``` `--name` is the worker name shown on the pool; use a different one per rig/GPU. 4. After a minute the log shows lines like ``` GPU 0: 16.90 MH/s avg over 30s | in kernel 95% ... | idle: no job 4%, stale 0% job 1792…: slot 11234, k=28, ttl 9680 ms ``` and the pool's dashboard lists your worker. Shares are accepted silently and counted in the 30-second rate line; a rejected share is logged. Several GPUs in one rig: run one process per GPU — it gives the steadiest rates and the pool hands every worker its own nonce range: ```bash CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=0 ./proxichorizo ... --name rig1-gpu0 & CUDA_DEVICE_ORDER=PCI_BUS_ID CUDA_VISIBLE_DEVICES=1 ./proxichorizo ... --name rig1-gpu1 & ``` (or one process with `--gpus 0,1`; each GPU still searches its own range). Instead of `--key-file` you may pass the key as `--seed-hex ` or in the environment variable `NONCE_SEEKER_SEED` — that is what the HiveOS/mmpOS/Docker packages do. ## 2. Solo mining — your own node and `proxi node mine` Solo, the reference miner `proxi node mine` builds, proves and submits the transits with your wallet key and ProxiChorizo only does the nonce search for it (the "external nonce seeker" protocol of Proxima). You need a synced Proxima access node and the `proxi` wallet (see https://lunfardo314.github.io/#/participate/proxi). 1. In the wallet profile `proxi.yaml` set the seeker server: ```yaml mine: seeker: listen: 0.0.0.0:8200 # where seekers fetch jobs (127.0.0.1:8200 if on the same box) token: some-long-random-string ``` 2. Start the miner without CPU workers (`--workers N` would add CPU search beside the GPUs): ```bash proxi node mine --workers 0 ``` 3. Start ProxiChorizo with the **same key file as the wallet** and the token from step 1: ```bash ./proxichorizo --proxi http://your-node:8200 --token some-long-random-string --key-file proxima.key ``` If the key file is passphrase-encrypted, pass the seed with `--seed-hex` instead. **Library upgrade at slot 29000 (2026-10-21 17:01 UTC, proxima v1.1.0-take1):** from that slot transits are stamped at tick 83 and carry miner version 5. `proxi node mine` does this for you, but only from v1.1.0-take1 on — update `proxi` and your node before the slot, or every transit your older `proxi` builds after it is rejected. ProxiChorizo itself needs no update (the nonce search is unchanged); pool miners have nothing to do. `proxi node mine` logs `SOLVED …` / `confirmed` for every transit; the first rewards of a new account are 15 PROX "opening" payouts until the account is known on the ledger, then 50 and 95 PROX (see the Proxima docs on mine payout options). Expect contests: several miners solve the same height and the smallest VRF output wins, so a small solo miner wins rarely but exactly in proportion to its hashrate. ## 3. HiveOS Custom miner, installation URL = the `proxichorizo-hiveos.tar.gz` asset of the release, miner name `proxichorizo`. Flight sheet: | Field | Value | |---|---| | Pool URL | `proxi.suprnova.cc:8100` (or your own `proxi node mine` seeker address) | | Wallet / template | your pool **token** (or the mine.seeker.token) | | Pass | the **seed**: first 64 hex characters of `private_key` in your key file | | Extra config | optional `KEY=VALUE` lines: `DEVICES=0,1`, `VARIANT=fe32_w4_lb2_c`, `THREADS=256`, `ITERS=2`, `EXTRA_ARGS=...` | Hashrate, shares and per-GPU figures come from the miner's stats API on `127.0.0.1:4078`. ## 4. MMPOS and SMOS The Linux tarball `proxichorizo-.tar.gz` contains `mmpos-run.sh` and `smos-run.sh`: set `POOL_URL` (job server), `TOKEN` (pool token) and `SEED` (as above) in the miner config (`WALLET`/`PASSWORD` are accepted as aliases), optionally `DEVICES`, `VARIANT`, `EXTRA_ARGS`. MMPOS's API-based "external miner" format is served by `mmp-launch.sh` / `mmp-stats.sh` / `mmp-external.conf` from the same tarball (pool = job server, user = token, password = seed). ## 5. Docker ```bash docker run -d --name proxichorizo --gpus all --restart unless-stopped -p 4078:4078 \ -e TOKEN=YOUR_TOKEN -e SEED=FIRST_64_HEX_OF_private_key -e WORKER=docker1 ocminersupr/proxichorizo:latest ``` `POOL_URL` defaults to `proxi.suprnova.cc:8100`; `KEY_FILE` (a mounted key) replaces `SEED`; arguments after the image name go straight to the miner. Statistics at `http://localhost:4078/`. ## 5b. AMD GPUs (OpenCL) Build `opencl/proxichorizo-cl` (see README) on a machine with an OpenCL driver (ROCm, or the vendor ICD on HiveOS/mmpOS). Usage is identical: `./proxichorizo-cl --proxi ... --token ... --key-file proxima.key --name rig1`; `--list-devices` shows the GPUs, `--gpus 0,1` selects them, `--autotune` picks the fastest kernel variant. The first start compiles the kernels (about a minute per variant) and verifies them against the built-in test vectors; a variant that does not pass on your driver is disabled automatically. Expect roughly 10.5 MH/s on an RX 7900 XTX, 3.6 MH/s on an RX 7600 XT and 4.5 MH/s on an MI50 with the 9×29-bit field layout (`fe29_wnaf_c`, the default since 0.5.0; the 0.4.0 numbers were about half of that). `--iters 8` is the best launch geometry on every AMD card measured so far. ## 6. Options ``` --proxi URL job server: pool (http://proxi.suprnova.cc:8100) or your proxi node mine --token T bearer token from the pool registration / mine.seeker.token --key-file FILE unencrypted proxima.key (or --seed-hex HEX, or env NONCE_SEEKER_SEED) --name NAME worker name shown by the pool (default proxichorizo-) --gpus 0,1 GPUs to use (default all; CUDA order, set CUDA_DEVICE_ORDER=PCI_BUS_ID for PCI order) --api-bind [H:]PORT statistics: JSON at http://H:PORT/, mmpOS line at /mmpos (default off) --autotune [SECS] try every kernel variant on each GPU first, use the fastest (6 s per trial) --variant NAME force one kernel variant (--list-variants); default: per-GPU-generation table --threads N --iters N --blocks N launch geometry (defaults from the table) --bench [SECS] measure the hashrate and exit --selftest vectors.json verify the math ``` Typical rates: RTX 5090 ≈ 29 MH/s, RTX 5080 ≈ 21 MH/s, RTX 3070 ≈ 7.4 MH/s, CMP 170HX ≈ 7 MH/s. The work is pure integer ALU; memory, PCIe link width and risers do not matter. ## 6b. Kernel choice per card Measured generations (RTX 30xx/50xx, PRO 6000, A100/CMP 170HX; RX 7900 XTX, MI50) start with their measured kernel at once. Any other card is measured on first start (~15 s, two candidate kernels) and the result kept in `proxichorizo-autotune.cache` in the directory the miner runs from (HiveOS: the miner directory); delete the file to re-measure, or pass `--autotune` for the full sweep. The start-up line shows the kernel, the launch shape and the reason. ## 7. Troubleshooting - `cannot fetch the job: HTTP 401` — wrong token. `HTTP 409` — the pool moved on to the next height while you were submitting; normal, nothing to do. - `the miner mines under key … but this seeker holds …` — the key file does not belong to the token (pool) or to the wallet (solo). Use the key the pool gave you / the wallet's key. - `no job (miner idle)` for long — the job server is between rounds or unreachable; check the URL and the pool status page. - Low in-kernel percentage — too little work per launch: raise `--iters` (4 on most GPUs, 2 on a CMP 170HX), or run one process per GPU. - Windows: `start.cmd` expects `proxima.key` in the unpacked folder; `start.ps1 -KeyFile` points elsewhere. The source and issue tracker: https://github.com/ocminer/ProxiChorizo