Complete Guide

Mining Guide

Configure and optimize TensorChain mining on Apple Silicon, Kneron accelerators, and sharded pools.

Overview

TensorChain Mining

TensorChain is a Proof of Useful Work algorithm designed to favor consumer NPUs over industrial GPU farms. It exploits the "Batch-1 Efficiency Gap" where unified memory architectures outperform PCIe-bound systems.

Memory-Bound

Matrix size targets 75% of system RAM, creating a Proof of Memory Capacity

INT8 Operations

Saturates Neural Engine tensor cores with integer arithmetic

Fast Verification

Freivalds' algorithm verifies in O(n²) vs O(n³) compute

Apple Silicon

MLX Backend Setup

Requirements

  • Apple Silicon Mac (M1/M2/M3 series)
  • macOS 13.0+ (Ventura or later)
  • Xcode Command Line Tools
  • 64 GB+ Unified Memory recommended

Installation

# Install MLX framework
pip install mlx
 
# Build tensor-miner with MLX backend
cd po8-core/miner
cargo build --release --features mlx

Configuration

# ~/.po8/miner.toml
 
[backend]
type = "mlx"
 
[tensorchain]
# Matrix dimension (auto-calculated from RAM if not set)
matrix_size = 8192
 
# Target RAM usage (0.75 = 75%)
memory_target = 0.75
 
# Use INT8 precision
precision = "int8"
 
[performance]
# Pin to performance cores
pin_cores = true
 
# Number of mining threads
threads = "auto"

Running the Miner

# Start mining
./target/release/tensor-miner \
  --config ~/.po8/miner.toml \
  --node http://localhost:26657
 
# With verbose logging
RUST_LOG=debug ./target/release/tensor-miner --config ~/.po8/miner.toml

Performance Tuning

Memory Allocation

Close other applications to maximize available unified memory. The miner performs best with uninterrupted memory access.

Thermal Management

Ensure adequate cooling. Sustained mining can throttle if temperatures exceed 100°C.

Power Settings

Disable sleep and use "High Power Mode" on MacBooks for consistent performance.

Kneron

Edge Accelerator Setup

Supported Devices

KL520
0.3 TOPSUSB 3.00.5W

Entry-level, image classification workloads

KL720
1.5 TOPSUSB 3.01.2W

Recommended for TensorChain shards

KL730
4 TOPSUSB 3.02W

High-performance edge mining

Installation

# Install Kneron SDK
pip install kneron-plus
 
# Build tensor-miner with Kneron backend
cd po8-core/miner
cargo build --release --features kneron
 
# Verify device detection
./target/release/tensor-miner --backend kneron --list-devices

Configuration

# ~/.po8/miner-kneron.toml
 
[backend]
type = "kneron"
device_id = 0
 
[sharding]
# Enable sharded mining for memory-constrained devices
enabled = true
 
# Shard size (smaller = less memory required)
shard_size = 512
 
# Pool coordinator address
pool_url = "https://pool.po8.network"

Sharded Mining

Kneron devices participate in sharded mining pools where large matrices are decomposed into sub-blocks.

A₁
A₂
A₃
A₄
Device 1
Device 2
Device 3
Device 4
C
Aggregated Result
InferNet

Useful Work Tasks

Beyond TensorChain entropy, InferNet lets miners earn rewards by executing useful AI inference tasks.

Supported Models

MobileNetV2
Image Classification
INT8 Quantized~3.4M params
ResNet50
Image ClassificationINT8 Quantized
INT8 Quantized~25M params
BERT-Tiny
Text Embedding
INT8 Quantized~4.4M params
Llama-3-8B
Text Generation
INT8 QATApple Silicon only

Enable InferNet

# Add to miner.toml
 
[infernet]
enabled = true
 
# Models to support
models = ["mobilenetv2", "resnet50"]
 
# Stake for optimistic verification
stake = 100
 
# Maximum concurrent tasks
max_tasks = 4

Verification Flow

1
Execute

Miner runs inference, posts result + state hash

2
Challenge Window

Fishermen can dispute within challenge period

3
Bisection

Disputes narrow to single divergent instruction

4
Resolution

On-chain execution determines truth, loser slashed

Monitoring

Mining Dashboard

Built-in Metrics

# View mining statistics
./target/release/tensor-miner status
 
# Output:
# Mining Status: Active
# Backend: MLX (Apple M2 Ultra)
# Matrix Size: 8192x8192
# Memory Usage: 98.5 GB / 128 GB
# Hashrate: 1.2 H/s
# Shares Submitted: 847
# Shares Accepted: 842
# Power Draw: ~85W
# Temperature: 78°C

Prometheus Metrics

# Enable metrics endpoint
[metrics]
enabled = true
port = 9090
 
# Available metrics:
# po8_miner_hashrate
# po8_miner_shares_submitted
# po8_miner_shares_accepted
# po8_miner_power_watts
# po8_miner_temperature_celsius
# po8_miner_memory_used_bytes
Economics

Mining Rewards

Block Rewards

Base emission for valid TensorChain proofs

InferNet Fees

Task fees from useful inference work

Mixnet Relay

Bandwidth rewards for privacy relay

Profitability Factors

  • Hardware efficiency (Joules per valid share)
  • Memory bandwidth utilization
  • Network difficulty adjustment
  • Electricity costs in your region
  • InferNet task availability

Ready to Mine?

Set up your node and start contributing to the Po8 network.

Node SetupNetwork Operations