Skip to main content

Miners

Miners provide AI inference compute for Orbinum Network, operating GPU infrastructure to execute AI models within specialized Orbits. Earnings are based on quality scores, with top performers receiving the majority of requests and rewards.

Miners power the AI capabilities that enable multichain agents to make intelligent decisions and execute complex workflows across multiple blockchains.

→ Learn more about Cross-Chain Agents

What Miners Do​

ResponsibilityDescription
AI InferenceExecute AI models for agent requests within registered Orbits
Quality CompetitionOptimize output quality, latency, and availability for ranking
InfrastructureOperate GPU hardware, manage models, maintain uptime
Proof SubmissionGenerate and submit cryptographic proofs of execution

Orbit Specialization​

Miners register for domain-specific Orbits, competing on quality within each specialization.

OrbitTasksTypical ModelsVRAM
NLPText generation, translation, analysisGPT, BERT, T5 (7B-70B)16-80 GB
Computer VisionClassification, detection, generationResNet, YOLO, Stable Diffusion8-24 GB
Predictive AnalyticsTime series, forecasting, anomaly detectionLSTM, XGBoost4-16 GB
Code AnalysisCode completion, bug detectionCodeLlama, StarCoder16-48 GB
Audio ProcessingSpeech-to-text, synthesisWhisper, Tacotron8-16 GB
EmbeddingSemantic search, similaritySBERT, E54-8 GB

Strategy: Focus on 1-2 Orbits where you can maintain competitive quality scores. Specialization enables infrastructure optimization and model expertise.

→ Learn more about Orbits

Hardware Requirements​

GPU Requirements​

TierGPUVRAMUse CaseExamples
EntryNVIDIA CUDA 7.0+16 GBSingle OrbitRTX 3090, RTX 4090, A5000
ProfessionalMulti-GPU setup40-80 GB/GPUMulti-OrbitA100, H100, RTX 6000 Ada
EnterpriseGPU cluster4-8 GPUs/nodeHigh volumeA100 clusters, NVLink

System Requirements​

ComponentMinimum (Single Orbit)Recommended (Multi-Orbit)
CPU8 cores @ 2.5 GHz16+ cores @ 3.0 GHz
RAM32 GB DDR4128 GB DDR4/DDR5
Storage500 GB NVMe SSD2 TB NVMe SSD (RAID 0)
Network1 Gbps, <50ms latency10 Gbps with redundancy

Software Requirements​

CategoryRequiredRecommended
OSUbuntu 22.04 LTS, Debian 11+Ubuntu 22.04 LTS
NVIDIADrivers 525.x+, CUDA 12.0+, cuDNN 8.9+Latest stable
AI FrameworkPyTorch 2.0+ or TensorFlow 2.13+PyTorch 2.0+
LibrariesTransformers 4.30++ vLLM, TensorRT, DeepSpeed
OrbinumMiner node, blockchain client+ Monitoring tools

Note: Orbinum node software is not yet publicly available. Installation instructions will be provided upon release.

Getting Started​

Requirements​

TypeRequirement
Economic1,000 $ON per Orbit + transaction fees
TechnicalGPU, models, blockchain node
NetworkPublic endpoint for request assignment

Step 1: Generate Keys & Register​

# Generate miner keys
orbinum-miner keys generate --output /secure/location/

# Register for Orbit (1,000 $ON stake per Orbit)
orbinum-miner register \
  --orbit nlp \
  --endpoint https://miner.example.com:8080 \
  --stake 1000 \
  --keys /secure/location/miner-keys.json

Step 2: Start Miner​

# Start miner node
orbinum-miner start \
  --orbit nlp \
  --model-path /models/llama-70b \
  --gpu 0,1,2,3 \
  --max-batch-size 32

Step 3: Immunity Period​

New miners get 12-hour protection (7,200 blocks):

  • ✅ Cannot be deregistered
  • ✅ Assigned median quality score
  • 🎯 Time to prove performance

Competitive Registration​

Orbit Capacity: Fixed slots (e.g., 1,024 miners per Orbit)

When full: New registrations replace lowest-ranked non-immune miners

  • Lowest miner auto-deregistered
  • Stake returned after unbonding period
  • Strategy: Maintain competitive quality scores to avoid deregistration

Auto-Staking: Enable to auto-compound rewards → increases protection against deregistration

# Enable auto-staking in config
orbinum-miner config --auto-stake true

Miner Commands​

# View your miner info
orbinum-cli orbit miners --orbit nlp --uid <YOUR_UID>

# Check leaderboard position
orbinum-cli orbit leaderboard --orbit nlp --top 20

# Monitor status
orbinum-miner status --miner-id <YOUR_ID>

Quality Evaluation​

Validators score miners on multiple metrics to determine ranking and request routing. Higher quality scores result in more requests and higher earnings.

Key Metrics:

  • Output Quality (50% weight): Accuracy, relevance, correctness
  • Latency (25% weight): Response time performance
  • Availability (15% weight): Uptime and acceptance rate
  • Cost-Effectiveness (10% weight): Price relative to quality

→ Learn more about Quality Consensus & Detailed Metrics

Request Distribution by Rank​

Ranking TierRequest ShareStrategy
Top 25%70% of requestsOptimize all metrics, premium hardware
25-50%20% of requestsFocus on reliability and consistency
50-75%8% of requestsImprove latency and model quality
Bottom 25%2% of requestsRisk deregistration if Orbit is full

Optimization Tips:

  • Quality: Use latest models, fine-tune for domain
  • Latency: Quantization (FP16/INT8), TensorRT, batching
  • Availability: Redundancy, monitoring, auto-failover
  • Cost: Optimize GPU utilization, efficient serving

Economics​

Revenue Sources​

Miners earn from multiple sources based on performance:

  • Block Emissions: 80% of total emissions distributed proportionally to quality scores
  • Inference Fees: 63.7% of user fees for handling requests
  • Quality Bonuses: Additional rewards for top performers

Expected Returns: Varies significantly by ranking tier and Orbit usage. Top 25% of miners receive ~70% of requests.

→ See detailed breakdown in Economic Model

Cost Structure​

CategoryItems
InfrastructureGPU hardware, electricity, cooling, bandwidth
Operational1,000 $ON stake/Orbit, transaction fees, maintenance
OpportunityStake lock-up cost, hardware depreciation

Slashing Penalties​

See the Slashing (Negative Incentives) section for a complete and up-to-date table of miner penalties, offenses, and prevention strategies.

Next Steps​