How to Mine TensorCash (TSC) — GPU mining setup guide

Miners online

group

Blockchain height

database

Pool Hashrate

bar_chart
PROOF-OF-INFERENCE
TensorCash mining is proof-of-inference — your GPU runs a real LLM

Your GPU earns TSC by running real LLM inference (Qwen3-8B) — there is no SHA-style hashing. You need one card with 24 GB or more and a wallet address. Everything else is one command.

Start mining in three steps

One GPU with 24 GB or more, a wallet address, one command.

1
Check your GPU 24 GB VRAM or more and Ampere or newer (RTX 3090 / 4090 / 5090, A6000, H100). The model runs on your card — it needs the memory.
2
Get a wallet address Any TensorCash mainnet address, starting with tc1. That is your username — no signup, payouts go straight to it.
3
Run the miner Pick your platform below and copy one command. Your worker shows up here a few minutes after the model loads.
Pool address

Pick the port that matches your card. That is the only difference between them.

Region Use this if you run Address
EU 🇩🇪 LIVE Getting ping… 4090 / 5090 / A6000 / H100 stratum+tcp://tsc.suprnova.cc:3310
3090 and other slower cards stratum+tcp://tsc.suprnova.cc:3309
EU2 🇺🇦 LIVE Getting ping… 4090 / 5090 / A6000 / H100 stratum+tcp://stratum-eu2.suprnova.cc:3310
3090 and other slower cards stratum+tcp://stratum-eu2.suprnova.cc:3309

Username is <your tc1 address>.<worker name>, password anything. Difficulty adjusts itself — you do not set it.

Pool terms
  • Fee1%
  • Payout schemePPLNS
  • Minimum payout0.1 TSC
  • Payoutsevery 60 min
  • Registrationnone
Install the miner

Click your platform to open its instructions.

Source code: github.com/ocminer/supr-meow-tsc

Docker — Linux command line recommended

You need Docker with the NVIDIA Container Toolkit, driver R570+, and about 40 GB free disk (8 GB image + 16 GB model). One container per GPU.

docker run -d --name meow0 --restart unless-stopped \
  --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0 \
  -e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
  -v /opt/models:/models \
  -e POOL_URL=stratum+tcp://tsc.suprnova.cc:3310 \
  -e WALLET=tc1your_address_here \
  -e WORKER=rig1 \
  -e MODEL_URL=https://huggingface.co/ocminer/Qwen3-8B-9c925d64-bf16-GGUF/resolve/main/Qwen3-8B-9c925d64-bf16.gguf \
  -e MODEL_SHA256=fef5847c18f860086007cec0b08960f206c13c5cba69e3ed6292ee1e02ed7e44 \
  ocminersupr/supr-meow-tsc:latest

Replace the pink values with your own. Nothing else needs changing.

The model is downloaded once per rig (16.4 GB, 8 parallel ranges) and the checksum is verified before mining starts — a mismatch makes the miner refuse to mine rather than submit bad shares. If HuggingFace is slow or blocked, use our mirror: https://www.suprnova.cc/models/Qwen3-8B-9c925d64-bf16.gguf (same file, same checksum).

Tuning (optional)

The miner reads your card and applies a measured profile — leave these unset unless you have a reason. SLOTS (windows per GPU), MEOW_GROUPS (sampler threads), DEVICES=0,1, SPLIT_MODEL=1. Never use GROUPS as the variable name — it is a bash built-in and always arrives as 0. The miner reads MEOW_GROUPS.

More than one card?

Run it again with --name meow1 and NVIDIA_VISIBLE_DEVICES=1. Keep the same -v /opt/models:/models so the 16 GB model downloads only once.

Is it working?

Run docker logs -f meow0. In the first minute you should see, in this order:

auto-tuned for <your card>: --slots N --groups M
model loaded: Qwen3-8B-...
pool model: Qwen/Qwen3-8B@9c925d64d727
share accepted

If something is wrong the miner prints the reason and stops, rather than mining at zero. Read the last few lines before restarting it.

HiveOS

Use the supr-meow-tsc custom miner package. Settings in the flight sheet:

Miner name          : supr-meow-tsc
Installation URL    : https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5.tar.gz
Wallet and worker   : %WAL%.%WORKER_NAME%
Pool URL            : stratum+tcp://tsc.suprnova.cc:3310
Pass                : x

Extra config arguments

MODEL_URL=https://huggingface.co/ocminer/Qwen3-8B-9c925d64-bf16-GGUF/resolve/main/Qwen3-8B-9c925d64-bf16.gguf
MODEL_SHA256=fef5847c18f860086007cec0b08960f206c13c5cba69e3ed6292ee1e02ed7e44

Do not rename the package file. HiveOS parses <name>-<version>.tar.gz and then looks for a directory of the same name inside. Any added suffix makes it read the version wrongly and the miner is never found.

The package reports hashrate, per-GPU temperature and power, and accepted/rejected shares to the HiveOS dashboard automatically. Same GPU requirements as Docker.

MMPOS

Add as a custom miner using this package:

https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-mmpos.tar.gz

Settings are passed as environment variables:

POOL_URL=stratum+tcp://tsc.suprnova.cc:3310
WALLET=tc1your_address_here.rig1
MODEL_URL=https://huggingface.co/ocminer/Qwen3-8B-9c925d64-bf16-GGUF/resolve/main/Qwen3-8B-9c925d64-bf16.gguf
MODEL_SHA256=fef5847c18f860086007cec0b08960f206c13c5cba69e3ed6292ee1e02ed7e44
SimpleMining (SMOS)

Custom miner. Unpack to /home/miner/custom/supr-meow-tsc/; same variables as MMPOS.

https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-smos.tar.gz

The model cache defaults to /home/miner/models so it survives miner updates.

Plain Linux rig (no Docker)
wget https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-linux-x86_64.tar.gz
tar -xzf supr-meow-tsc-0.3.5-linux-x86_64.tar.gz
cd supr-meow-tsc
export LD_LIBRARY_PATH=$PWD/lib

./supr-meow-tsc -o stratum+tcp://tsc.suprnova.cc:3310 \
                -u tc1your_address_here.rig1 -p x \
                --model /path/to/Qwen3-8B-9c925d64-bf16.gguf

--list-devices shows the detected GPUs, --help lists the rest. Leave --slots and --groups unset — the miner reads the card and applies a measured profile.

All downloads & checksums v0.3.5

Direct links, and the SHA-256 of each package so you can verify what you downloaded.

HiveOS    https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5.tar.gz
MMPOS     https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-mmpos.tar.gz
SMOS      https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-smos.tar.gz
Linux     https://github.com/ocminer/supr-meow-tsc/releases/download/v0.3.5/supr-meow-tsc-0.3.5-linux-x86_64.tar.gz
Docker    docker pull ocminersupr/supr-meow-tsc:latest
fd0e970fc93951580b84aa00b9ba21e45afcaec67d7276252ae603e9cb61592f  supr-meow-tsc-0.3.5.tar.gz
029c47c897d568907ad88cef6a733a6f37b5be39764a8bdd32ac47747af324ae  supr-meow-tsc-0.3.5-mmpos.tar.gz
bb39785950a36c0fd5065b4a1809ee97ad0fb93b8a8bb0d1482d323cbcbbcc54  supr-meow-tsc-0.3.5-smos.tar.gz
5dd88078386985fd041ef954d56a751c285b491a8d1dbc2cc03cb5cdf34108ab  supr-meow-tsc-0.3.5-linux-x86_64.tar.gz
What your hardware needs
  • Ampere or newer. Turing cards (RTX 2080 Ti, Titan RTX, including 22 GB modded ones) cannot do bf16 and their shares are rejected. Check this first.
  • 24 GB VRAM per card — the model runs on the GPU and has to fit.
  • CPU with AVX2 (Intel 2013 or newer, any AMD Ryzen).
  • NVIDIA driver R570 or newer.
  • ~40 GB free disk for the image and the model.
Cards under 24 GB

Several smaller cards can be combined into one worker — the model is split across them. Add -e SPLIT_MODEL=1 and list them all in one container (NVIDIA_VISIBLE_DEVICES=0,1).

Cards must be identical. This pools their memory, it does not add speed — 2 × RTX 5080 measured 7.5 PoI/s. Do not also start one container per card; they would fight over the same GPUs.

What rate to expect

These are PoI/s — proofs per second, not hashes. Single or double digits is normal for proof-of-inference. If you come from a hash coin, nothing is broken.

GPUPoI/s
B20056
H100 / H20040
RTX PRO 6000 Blackwell26
A100 80GB21
RTX 509018.6
RTX 6000 Ada12.5
RTX A6000 / A409.5
2× RTX 5080 (split)7.5

Cards not listed still work — they are sized automatically from VRAM.

If something looks wrong
What you seeWhat it means
“this GPU cannot mine TensorCash — no bf16 support” Turing or older. Nothing to fix — the card cannot mine TSC.
Fails to allocate / OOM at startup Card too small for the model. Split it across several GPUs.
Stuck for several minutes before mining First run downloads the 16.4 GB model. Normal, once per rig.
“CHECKSUM MISMATCH — refusing to mine” Corrupted download. Delete the .gguf and restart; it refetches.
Rate looks far too low PoI/s, not MH/s. See the table on the left.
HiveOS says the miner is not found The package was renamed. It must stay supr-meow-tsc-<version>.tar.gz.

stratum+ssl:// works if you prefer TLS. ws:// and wss:// are rejected — that is a different protocol, not this pool. The miner exposes local stats on 127.0.0.1:21550 (--api-bind off disables it); HiveOS reads it for hashrate reporting.

Getting a TensorCash Wallet