Earn TON with Your GPU

Provide compute power to the Cocoon network and receive TON for processing AI inference requests. Secure, transparent, and fully automated.

Download Cocoon Worker Read Documentation

How It Works

Download & Install

Download the Cocoon Worker distribution, unpack it, and configure your settings.

Configure GPU & Wallet

Set up your NVIDIA GPU for confidential computing and provide your TON wallet address.

Earn TON

Start the worker and receive automatic TON payments for every AI request processed.

Hardware Requirements

GPU Requirements

GPU: NVIDIA H100 / H200
VRAM: 80GB+
CC: Confidential Computing support required
RTX series NOT supported for production

Server Requirements

CPU: Intel (TDX)
Linux: 6.16+
QEMU: 10.1+
RAM: 128GB+

Software & Accounts

Updated VBIOS (via NVIDIA support)
Hugging Face token
TON wallet for payouts

Earnings Calculator

Select GPU

AI Model

Network Utilization: 50%

~125 TON

per month (estimated)

These are estimated values. Actual earnings depend on network demand, model selection, and uptime.

Quick Start

1. Download and unpack

wget https://ci.cocoon.org/cocoon-worker-release-latest.tar.xz
tar xzf cocoon-worker-release-latest.tar.xz
cd cocoon-worker

2. Start seal-server (required for production)

./bin/seal-server --enclave-path ./bin/enclave.signed.so

3. Configure worker

cp worker.conf.example worker.conf
# Edit configuration:
# owner_address = YOUR_TON_WALLET_ADDRESS
# gpu = 0000:01:00.0
# model = Qwen/Qwen3-0.6B
# hf_token = hf_xxx...

4. Start the worker

./scripts/cocoon-launch worker.conf

For detailed instructions, see the full setup guide.

How Payments Work

👤 Client Wallet

→

🔄 Proxy Contract

→

⚙️ Worker Contract

→

💰 Your Wallet

Automatic Payouts

Smart contracts on TON blockchain handle all payments automatically. No manual claims needed.

Transparent & Verifiable

All transactions are recorded on-chain. View your earnings anytime via TON explorer.

Security & Privacy

🔒

Intel TDX

Workers run inside hardware-isolated virtual machines. Host cannot access request data.

🛡️

Data Privacy

All prompts and responses are encrypted. You never see user data.

✅

GPU Attestation

Hardware verification ensures only genuine GPUs participate in the network.

📄

Smart Contracts

Open-source contracts on TON ensure fair and transparent payments.

Full Setup Guide

1 Prerequisites

Before starting, ensure your hardware and software meet the following requirements for running a Cocoon worker.

Hardware Requirements

GPU: NVIDIA H100 / H200 (Confidential Computing)
CPU: Intel Xeon (TDX support)
RAM: 128GB+ DDR5
Storage: NVMe SSD, 500GB+
Network: 1Gbps+, stable connection

Software Requirements

Linux: Kernel 6.16+ (TDX support)
QEMU: 10.1+ (TDX patches)
NVIDIA Driver: 560+

Note: Consumer GPUs (RTX series) are not supported due to lack of Confidential Computing capabilities.

2 VBIOS Update

GPU attestation requires a special CC-enabled VBIOS. Standard VBIOS versions do not support attestation features needed for production deployment.

Update Process

Contact NVIDIA Enterprise Support with your GPU serial number
Request CC-enabled VBIOS for your specific GPU model (H100/H200)
Follow NVIDIA instructions to flash the new VBIOS
Verify update with nvidia-smi and check CC mode availability

Warning: Incorrect VBIOS flashing can damage your GPU. Follow NVIDIA instructions carefully.

3 TDX Configuration

Intel Trust Domain Extensions (TDX) provides hardware-isolated virtual machines. TDX must be enabled in BIOS for production workers.

BIOS Settings

Enter BIOS setup (usually F2/Del on boot)
Navigate to Security or Advanced CPU settings
Enable "Intel TDX" or "Trust Domain Extensions"
Save and reboot

Verify TDX is Active

dmesg | grep -i tdx

4 Configuration File

The worker.conf file (INI format) contains all settings for your worker. Copy the example file and edit it with your values.

cp worker.conf.example worker.conf
nano worker.conf

Required Parameters

Parameter	Description
`type`	Must be "worker"
`model`	AI model to serve (e.g., Qwen/Qwen3-0.6B)
`owner_address`	Your TON wallet address for payouts
`gpu`	PCI address of GPU (find with lspci \| grep -i nvidia)
`hf_token`	Hugging Face API token for model download
`node_wallet_key`	Base64-encoded worker wallet private key

Optional Parameters

Parameter	Default	Description
`instance`	0	Worker instance number for multi-GPU setups
`worker_coefficient`	1000	Price multiplier (1000 = 1.0x, 1500 = 1.5x)
`persistent`	auto	Path to persistent disk image

Example Configuration

type = worker
model = Qwen/Qwen3-0.6B
owner_address = EQC...your_ton_address
gpu = 0000:01:00.0
hf_token = hf_xxxxxxxxxxxxx
node_wallet_key = base64_encoded_key
ton_config = ./mainnet-config.json
root_contract_address = EQ...

Tip: Use lspci | grep -i nvidia to find your GPU PCI address.

5 Seal Server

seal-server provides secure key derivation for TDX guests using an SGX enclave. It must run on the host before starting workers.

Why seal-server is Required

Derives cryptographic keys tied to the TDX image
Keys persist across reboots
Ensures host cannot access worker secrets

Start seal-server

./bin/seal-server --enclave-path ./bin/enclave.signed.so

Note: One seal-server instance can serve multiple workers. Keep it running in a separate terminal or as a systemd service.

6 Launch Worker

After configuring worker.conf and starting seal-server, you can launch the worker in different modes.

Execution Modes

Mode	Command
Production	`./scripts/cocoon-launch worker.conf`
Test (real TON)	`./scripts/cocoon-launch --test worker.conf`
Test (mock TON)	`./scripts/cocoon-launch --test --fake-ton worker.conf`

Command-line Overrides

./scripts/cocoon-launch \
  --instance 0 \
  --worker-coefficient 1500 \
  --model Qwen/Qwen3-8B \
  worker.conf

7 Multi-GPU Setup

Run multiple workers on a single server, each handling one GPU. Use the --instance flag to assign unique identifiers.

Port Assignment

Instance	Port	CID
0	12000	6
1	12010	16
2	12020	26

Launch Multiple Workers

# GPU 1
./scripts/cocoon-launch --instance 0 --gpu 0000:01:00.0 worker.conf &

# GPU 2
./scripts/cocoon-launch --instance 1 --gpu 0000:41:00.0 worker.conf &

Tip: Each instance gets separate persistent storage automatically.

8 Monitoring

Monitor worker health and performance through HTTP endpoints and the health-client utility.

HTTP Endpoints

Endpoint	Description
`/stats`	Human-readable status
`/jsonstats`	JSON format for automation
`/perf`	Performance metrics

curl http://localhost:12000/stats
curl http://localhost:12000/jsonstats | jq

Health Client Commands

./health-client --instance worker status
./health-client -i worker gpu
./health-client -i worker logs cocoon-vllm 100
./health-client -i worker all

9 Troubleshooting

Common issues and their solutions when running Cocoon workers.

Worker Does Not Start

Verify seal-server is running
Check GPU is in CC mode
Confirm VBIOS is updated
Ensure TDX is enabled in BIOS

Diagnostic Commands

# Check GPU CC mode
nvidia-smi -q | grep "Confidential"

# Verify seal-server
ps aux | grep seal-server

# Check TDX
dmesg | grep -i tdx

# View logs
./health-client -i worker logs cocoon-vllm 50

Resources

Frequently Asked Questions

How much can I earn?

Earnings depend on your GPU model, the AI models you serve, network utilization, and uptime. Use our calculator above for estimates. As the network grows, demand and earnings potential will increase.

Which GPUs are supported?

Currently, only NVIDIA H100 and newer GPUs with Confidential Computing support are accepted for production. Consumer GPUs (RTX series) are not supported due to lack of CC capabilities.

Is it safe for my hardware?

Yes. Cocoon runs AI inference workloads, not traditional mining. There's no excessive wear on your GPU. The TEE environment ensures secure, isolated execution.

How do I set up multiple GPUs?

Run separate worker instances for each GPU using the --instance flag. Each instance gets unique ports and handles one GPU. Example: ./scripts/cocoon-launch --instance 0 --gpu 0000:01:00.0 worker.conf

What is seal-server and why is it required?

seal-server provides secure key derivation for TDX guests using an SGX enclave. It ensures your worker's cryptographic keys survive reboots and are tied to the verified image. Without it, production workers won't start.

How do I update VBIOS for GPU attestation?

Contact NVIDIA support to request a CC-enabled VBIOS for your GPU model. Standard VBIOS versions may not support attestation. This is required for production deployment.

What if my worker doesn't start?

Check that: 1) seal-server is running, 2) GPU is in CC mode (use setup-gpu-vfio script), 3) VBIOS is updated, 4) TDX is enabled in BIOS. See the deployment docs for troubleshooting.