Validators

Security through useful work.

Validators on Amadeus are selected through UPoW score and performance — they propose blocks, collect network fees, and rotate every epoch. Traditional proof of work burns energy on useless hashes; Amadeus miners perform real AI computation that doubles as the work securing the chain.

Selected by UPoW score + performance Rotation every epoch (~14h) BLS12-381 aggregate signatures >67% supermajority finality Tolerates 33% faulty trainers Slashing for downtime & equivocation
1
Work

Nodes solve an AI task — matrix multiplication — and submit the result on-chain.

2
Verification

Other trainers verify the solution and award points to the solver.

3
Mining

Top-scoring trainers become validators: they propose blocks and collect fees.

Operating a node

What validators sign up for.

Hardware

Modern AI GPUs

UPoW mining runs Tensorcore matrix multiplication. Participating compute should be modern AI hardware — consumer RTX 4090/5090 class or datacenter H100/H200 & B100/B200 class.

  • GPU clusters or single nodes
  • TEE-capable GPUs earn premium rewards (phasing in)
Performance

Uptime & throughput

Trainers are assigned entry slots each epoch. Missing slots, slow proposals, or downtime risks slashing — performance is measured continuously.

  • ≥95% uptime expected under grant terms
  • Entries in ~0.5s; delays >8s mid-epoch are slashable
  • Epochs of 100,000 entries (~14 hours)
Rewards

Emissions & fees

Validators collect network fees; miners earn UPoW emissions. After the Shenron Upgrade, net miner rewards rise from 43k to 186k AMA per epoch by epoch 810.

  • Mining fee falling from 85% to 25% (epochs 750–810)
  • Earmarked TEE pool for GPU-TEE operators
  • Grants of 100k–200k AMA available to bootstrap

Bring compute. Secure the agent economy.

Whether you run a single RTX rig or a multi-geo H100 cluster, the network — and the grants program — has a place for you.

Apply for a miner grant Talk to the Foundation