Weir

# Weir · Bittensor Subnet 36

Weir is a GPU rental marketplace on Bittensor. Miners bring GPU machines, validators check that the cards are real and fast, and renters on [weircompute.com](https://weircompute.com) rent them by the hour.

This repo has both sides:

| Program | Who runs it | What it does |
| --- | --- | --- |
| `weir-node` | Miners | Answers GPU checks and starts rental containers on your machine |
| `weir-validator` | Validators | Checks every listed GPU and sets weights |

**At a glance**

| | |
| --- | --- |
| Network | `finney`, netuid `36` |
| Validator hotkey | `5Ec48Jz5mS5kY7198cJFSDHDNPEAvDHnsCNaiB9bXq1Ux2iX` |
| Platform hotkey | `5ERf3azzYiJEdUdgufdM9bnz4UU94igLh362gBSBXWVdJehe` |
| Cards that earn | H100 80GB · A100 80/40GB · A800 80/40GB · RTX 5090 · RTX 4090 48GB · RTX 4090 |
| Rent is paid in | USDT or USDC, weekly |
| Questions | [Telegram](https://t.me/+qmO92A3EUt5hYWMx) · [email protected] |

`weir-node` already knows these values (`weir_subnet/network.py`), so you don't type them.

---

## Quick start for miners

Ubuntu 22.04 or 24.04 with an NVIDIA driver and Docker. Register a hotkey first (`btcli subnet register --netuid 36`). Only the hotkey goes on the GPU machine; leave your coldkey on your own computer.

```bash
# 1. GPUs inside Docker: NVIDIA container toolkit
curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg
curl -fsSL https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list \
  | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
  | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list
sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker && sudo systemctl restart docker

# 2. Sysbox, so renters get root and Docker inside their container (restarts Docker)
curl -fsSL -o sysbox.deb https://github.com/nestybox/sysbox/releases/download/v0.6.6/sysbox-ce_0.6.6-0.linux_amd64.deb
sudo apt-get install -y ./sysbox.deb

# 3. weir-node
git clone https://github.com/weircompute/weir-subnet && cd weir-subnet
python3 -m venv .venv && .venv/bin/pip install -e '.[gpu]'

# 4. Self-test: the same GPU check the validator runs
.venv/bin/weir-node selftest --payout-address <USDT/USDC address>

# 5. Run, then publish your address on chain
.venv/bin/weir-node run --wallet miner --hotkey default --price 0.35 \
    --public-host <public ip> --payout-address <USDT/USDC address>
.venv/bin/weir-node announce --wallet miner --hotkey default --ip <public ip> --port 8091

# 6. See how your machines are doing
.venv/bin/weir-node status --wallet miner --hotkey default
```

Run step 5 as a systemd service so it comes back after a reboot.

> **Self-test says "The NVIDIA driver on your system is too old"?** pip installed a PyTorch built for a newer CUDA than your driver. Look at the CUDA version in the top right of `nvidia-smi` and install the matching build. For CUDA 12.4 (driver 550):
>
> ```bash
> .venv/bin/pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cu124
> ```

### Ports

Open the node port (default `8091`) and the renter SSH range (`--ssh-ports`, default `42000-42999`) to the internet.

Behind a provider that maps ports (external 30282 → internal 8800, for example), run the node on the internal port and announce the external one. For `--ssh-ports`, use ports that map one to one.

### More than one machine

Use the same hotkey everywhere and give each machine its own `--name`. On the machine whose address is on chain, add `--fleet main=http://1.2.3.4:8091,box2=http://1.2.3.5:8091`. A machine in the list that doesn't sign with your hotkey is ignored.

---

## Activation fee

Each GPU pays a one-time activation fee, set per card. Fees change with demand, so they are not listed here. Get the current ones, with the price range each card may be listed at:

```bash
.venv/bin/weir-node quote --wallet miner --hotkey default
```

A quote is held for that hotkey for 7 days: GPUs it activates in that time pay at most the quoted fee. No hotkey yet? Sign in at [weircompute.com](https://weircompute.com), create an API key under **API keys**, and ask with it:

```bash
curl -H "Authorization: Bearer <API key>" "https://weircompute.com/v1/miner/quote?hotkey=<your hotkey>"
```

Pay from the hotkey's coldkey to `5Dk1ftRx4wpwZsiufim8KvHPaeY7k9RdkXJmii2ADv9uHfZ9`. Payments add up as credit, and each GPU is activated once it passes its first check. A GPU that later reports a dearer card owes the difference within 7 days. Activation fees are not refunded.

---

## How you earn

Every weight-setting round, the miner share of emissions is split three ways:

| Pool | Share | Who gets it |
| --- | --- | --- |
| Rented | 10% | GPUs rented on Weir, by GPU-hours actually used × card weight. Rentals under 10 minutes don't count, and a miner gets at most the platform fee its rentals paid |
| Idle | up to 90% | Verified GPUs waiting for a renter: card weight × 90% ÷ 1134 each, until the pool is full |
| Network fund | what is left | The reserve hotkey (below) |

On top of emissions, **rent goes to you**: what the renter paid, minus refunds and Weir's 5% fee. It becomes payable 48 hours after the rental ends and is sent weekly in USDT or USDC once it reaches $5.

**Card weights:** H100 4.0 · A100 80GB 2.5 · A800 80GB 2.4 · A100 40GB 1.8 · A800 40GB 1.7 · RTX 5090 1.3 · RTX 4090 48GB 1.2 · RTX 4090 1.0

**What an idle GPU needs before it earns:**

- 14 days of passed checks in a row. A missed check, a changed GPU set or a wrong answer starts the count again.
- Your price between half the cap and the cap for that card.
- Disk at least 1.5× the machine's total VRAM.
- Sysbox as the container runtime.
- A node that answers at least half the time.

Weir may review GPUs periodically. A reviewed GPU can earn before its probation ends,