How to Mine Keryx (KRX) — Mining Setup Guide for suprnova.cc

Step-by-step guide to start mining Keryx (KRX) at krx.suprnova.cc. Works on any modern NVIDIA or AMD GPU with the latest keryx-miner-supr release. VarDiff stratum on ports 4401 (low) and 4404 (recommended), plus SSL on 4405. 1% pool fee, PPLNS payouts, automatic 20% coinbase escrow claiming on your behalf.

Miners online

group

Workers online:

Blockchain height

database

Network Difficulty:

Pool Hashrate

bar_chart

Nethash:

How to mine Keryx (KRX) in three steps

1% fee · hourly PPLNS payouts · no password

Paste your KRX wallet address

keryx:…

Every command on this page fills in your address automatically.

No wallet yet? See Coin Infos. Addresses start with keryx:.

Pick your GPU

Download, then run this

keryx-miner-supr-modern-0.15.2-linux-x86_64.tar.gz (1.1 GB) · NVIDIA driver 570 or newer, RTX 20xx and newer, H100

tar xzf keryx-miner-supr-modern-0.15.2-linux-x86_64.tar.gz && cd keryx-miner-supr-modern
./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

keryx-miner-supr-legacy-0.15.2-linux-x86_64.tar.gz (1.3 GB) · GTX 10-series, Tesla V100, CMP 100 to 210, driver 550 or newer. P100 is not supported.

tar xzf keryx-miner-supr-legacy-0.15.2-linux-x86_64.tar.gz && cd keryx-miner-supr-legacy
./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

keryx-miner-supr-amd-0.15.2-linux-x86_64.tar.gz (43 MB) · AMD OpenCL for the walk, Vulkan for inference, both included

tar xzf keryx-miner-supr-amd-0.15.2-linux-x86_64.tar.gz && cd keryx-miner-supr-amd-0.15.2
./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

keryx-miner-supr-windows-nvidia-pom.zip (886 MB, NVIDIA) · keryx-miner-supr-windows-amd.zip (41 MB, AMD) · extract, open PowerShell in that folder

.\keryx-miner-supr.exe -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

rig01 is the worker name, change it per rig. No tier flag is needed: every card loads the heaviest AI model its VRAM fits. Your rig shows on Your Stats within a few minutes.

First start takes a few minutes. Keryx mining serves AI inference, so the miner first downloads the model for your card (about 6.5 GB for the entry tier, up to 28 GB for the top tier) and builds its possession index. During that time it shows 0 hash/s (preparing). That is normal. No model, no mining: if it logs OPoI: no models ready, the card has too little VRAM for the chosen tier.

Stratum Servers

Region	Host	Ports	Ping
EU (Germany)	`krx.suprnova.cc`	4404 GPU · 4401 Low · 4405 SSL	Getting Ping
EU2 (Ukraine)	`stratum-eu2.suprnova.cc`	4404 GPU · 4401 Low · 4405 SSL	Getting Ping
US-East	`stratum-us.suprnova.cc`	4404 GPU · 4401 Low	Getting Ping

Use the region closest to your rig; the other two work as backup pools. Port 4404 adapts the difficulty to any rig automatically (VarDiff), 4401 starts lower for small or slow cards, 4405 is 4404 over TLS (stratum+ssl://). No password: leave it empty and do not set d=, a static difficulty makes small cards look like they never find a share. Why ping barely matters.

Setup by platform

Wallet and worker go in as keryx:ADDRESS.worker. Current release v0.15.2, all downloads on the release page.

keryx-miner-supr recommended

v0.15.2 · one archive per GPU line: modern (RTX 20xx and newer, H100), legacy (GTX 10xx, V100, CMP 100 to 210), amd (OpenCL). Each archive contains a run.sh launcher that sets the library path, so start the miner with ./run.sh. The CUDA runtime ships inside the archive; only the NVIDIA driver is needed.

./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

Select GPUs, run in the background, share one model store, stop safely

# only GPU 0 and 2 (CUDA order); omit --cuda-device to use all cards
./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --cuda-device 0,2
# headless, log to a file, models on a shared NFS store (one download for the whole farm)
setsid nohup ./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --no-tui --model-dir /mnt/share/keryx-models >~/keryx.log 2>&1 &
# local stats JSON for dashboards (also /mmpos)
./run.sh -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --api-bind 127.0.0.1:4067

Stop with Ctrl-C or SIGINT only (pkill -INT -x keryx-miner-supr). The miner has a built-in dashboard in a terminal; --no-tui gives plain log lines. Clocks and power can be set from the miner too: --cuda-power-limits, --cuda-lock-core-clocks, --cuda-fan-speed.

Official keryx-miner from Keryx Labs (v0.5.4 or newer) works here too

The stock reference miner speaks our stratum and is fully accepted; builds older than v0.5.4 are refused at login. The Suprnova build above adds the legacy line, the Docker image, per-card tier selection and VarDiff-friendly reconnects.

./keryx-miner -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404

keryx-miner-supr NVIDIA

v0.15.2 · keryx-miner-supr-windows-nvidia-pom.zip (886 MB) · RTX 20xx and newer · extract, open PowerShell in that folder

.\keryx-miner-supr.exe -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

keryx-miner-supr AMD

v0.15.2 · keryx-miner-supr-windows-amd.zip (41 MB) · OpenCL build with the Vulkan inference route

.\keryx-miner-supr.exe -a YOUR_KERYX_ADDRESS.rig01 -s stratum+tcp://krx.suprnova.cc:4404 --backup-pool stratum+tcp://krx.suprnova.cc:4401

Windows Defender may flag miners on first run; allow the folder if it does. Models are stored in the models folder next to the exe (change with --model-dir D:\keryx-models). GTX 10-series users: the Windows package is the modern line; use the Linux legacy build for those cards.

One image for all three: ocminersupr/keryx-miner-supr:0.15.2 (1.25 GB, NVIDIA RTX 20xx and newer, driver 570 or newer). It starts mining on its own; only the wallet is required. Mount a volume at /models so the multi-GB model is downloaded once.

Your own Docker host

Needs the NVIDIA Container Toolkit. One line:

docker run -d --name keryx --restart unless-stopped --gpus all \
  -e KERYX_WALLET=YOUR_KERYX_ADDRESS -e KERYX_WORKER=rig01 \
  -v keryx-models:/models \
  ocminersupr/keryx-miner-supr:0.15.2

All environment variables

`KERYX_WALLET`	required, your `keryx:` address
`KERYX_WORKER`	worker name, default: container hostname
`KERYX_POOL`	default `stratum+tcp://krx.suprnova.cc:4404`; use `stratum-us.suprnova.cc` or `stratum-eu2.suprnova.cc` for a host near those
`KERYX_BACKUP`	backup pool, default `stratum+tcp://krx.suprnova.cc:4401`
`KERYX_TIER`	`auto` (default), `very-light`, `light`, `default`, `high`, `very-high`
`KERYX_MODEL_DIR`	model directory inside the container, default `/models`
`KERYX_DEVICES`	CUDA device list such as `0,1`; default all
`KERYX_EXTRA_ARGS`	anything else, appended to the miner command line
`ROOT_PASSWORD`	SSH root password inside the container (default `keryx`); only matters if you publish port 22

Rent GPUs on Clore.ai

Rent an NVIDIA machine with 8 GB VRAM or more, choose a custom Docker image and fill in two variables. The image finds the pool on its own.

Image	`ocminersupr/keryx-miner-supr:0.15.2`
Environment variables	`KERYX_WALLET=YOUR_KERYX_ADDRESS KERYX_WORKER=clore01`
Command	leave empty, the image has its own entrypoint
Ports	none needed; add `22` tcp only if you want SSH into the container
Disk	at least 35 GB free for the top tier model, 10 GB for the entry tier

KERYX_WALLET must be your real address. Without it the container idles and prints an error; with a wrong one the pool rejects the login with "Invalid wallet". A rental that connects but never appears on Your Stats is almost always this.

Pick a US host? Add KERYX_POOL=stratum+tcp://stratum-us.suprnova.cc:4404. The first minutes are spent downloading the model, so very short rentals waste time; a persistent disk keeps the model between orders.

Rent GPUs on Octa.space

Marketplace → Custom image. Same image, same two variables.

Image	`ocminersupr/keryx-miner-supr:0.15.2`
Environment variables	add `KERYX_WALLET` = your address and `KERYX_WORKER` = a name, then deploy
Start command	leave empty, the image has its own entrypoint

Choose a node with an RTX 30, 40 or 50 series card and enough disk for the model. Everything else behaves like the Clore setup above.

Flight sheet: any coin, your keryx: wallet, Pool Configure in miner, Miner Custom with the fields below. Save, apply to the rig, then watch the miner log: the first start downloads the model before shares appear.

keryx-miner-supr recommended

v0.15.2 · one package per GPU line, do not rename the file: the line name is part of the miner name

Installation URL (RTX 20xx+)	`https://github.com/ocminer/keryx-miner-supr/releases/download/v0.15.5/keryx-miner-supr-modern-0.15.5.tar.gz`
Installation URL (GTX 10xx)	`https://github.com/ocminer/keryx-miner-supr/releases/download/v0.15.5/keryx-miner-supr-legacy-0.15.5.tar.gz`
Installation URL (AMD)	`https://github.com/ocminer/keryx-miner-supr/releases/download/v0.15.5/keryx-miner-supr-amd-0.15.5.tar.gz`
Miner name	filled from the URL (`keryx-miner-supr-modern`, `-legacy` or `-amd`)
Hash algorithm	`keryxhash`
Wallet and worker template	`YOUR_KERYX_ADDRESS.%WORKER_NAME%`
Pool URL	`stratum+tcp://krx.suprnova.cc:4404`
Pass	leave blank
Extra config arguments	leave blank for per-card auto; `--very-high` or another tier flag pins every card to one tier; `--force-model very-high,light` sets a model per card

Keep the stratum+tcp:// scheme in the pool URL. The package stores models under /hive/miners/custom/models, so they survive miner updates. Also available: SimpleMiningOS packages keryx-miner-supr-smos-modern-0.15.2.tar.gz and -smos-legacy-.

Add a Custom Miner with the archive for your GPU line, then set the fields. Worker is %WORKER% in mmpOS.

keryx-miner-supr recommended

Custom miner archive	`https://github.com/ocminer/keryx-miner-supr/releases/download/v0.15.2/keryx-miner-supr-modern-mmpos_0.15.2.tar.gz` also `-legacy-mmpos_` for GTX 10xx and `-amd-mmpos_` for AMD
Algorithm	`keryxhash`
Pool URL	`stratum+tcp://krx.suprnova.cc:4404`
Wallet	`YOUR_KERYX_ADDRESS.%WORKER%`
Extra args	`--api-bind 127.0.0.1:4067` then point the mmpOS agent (4.0.18 or newer) at `http://127.0.0.1:4067/mmpos` for per-GPU stats

SimpleMiningOS: use keryx-miner-supr-smos-modern-0.15.2.tar.gz (or -smos-legacy-) as a custom miner with the same wallet, pool and algorithm fields. macOS Apple Silicon: keryx-miner-supr-macos-arm64-0.15.2.tar.gz.

Model tiers: which AI model your card runs

Keryx pays a higher block reward for heavier models, and the pool passes that bonus through in full (Tier Rewards). With no flag every card loads the heaviest model its VRAM fits, which is what you want.

Flag	Model	VRAM needed	Download	Block reward
`--very-light`	Qwen3.5-9B	7 GB	6.5 GB	base
`--light`	GLM-4-9B	11 GB	7 GB	+17%
`--tier default`	Gemma-4-12B	15 GB	9 GB	+33%
`--high`	Qwen3.6-27B	22 GB	17 GB	+50%
`--very-high`	Kimi-Linear-48B	28 GB	28 GB	+67%

Examples: an 8 GB RTX 3070 runs Qwen3.5-9B, a 16 GB 5070 Ti runs Gemma-4-12B, a 24 GB 3090 or 4090 runs Qwen3.6-27B, a 32 GB RTX 5090 runs Kimi-Linear-48B. Mixed rigs get a different model per card automatically; --force-model very-high,light overrides it per card in CUDA order. The reward column is the network's tier bonus over the base model.

If something looks wrong

`0 hash/s (preparing)`	Model download and index build on first start. Several minutes, longer for the top tiers. Wait.
`OPoI: no models ready`	No tier fits this card or the download failed. Check VRAM against the table above and free disk space in the model directory.
"Invalid wallet" at login	The address is mistyped or a placeholder. It must start with `keryx:`; the worker name follows a dot.
Rejects "low difficulty" after a reconnect	Seen with the official miner 0.5.x, which keeps hashing at the old difficulty after a VarDiff change. Use keryx-miner-supr, which handles the retarget.
Connected but no shares for a long time	Remove any `d=` password and let VarDiff set the difficulty; small cards belong on port 4401.
Miner exits at start on an RTX 50 card	NVIDIA driver 570 or newer is required for RTX 50 and H100 with the modern line.
Still stuck?	Ask on Discord with the first 50 lines of the miner log, or open an issue on GitHub.

Building your own Keryx miner?

Full wire-protocol specification — JSON-RPC kaspad-dialect, OPoI tag, Proof-of-Model (PoM) extension, error codes, reference miners, source references. Everything you need to ship a third-party Keryx mining client against this pool.

Stratum Protocol Spec v3 →