Omniverse Compute (OMC) is a decentralized physical infrastructure network (DePIN) that turns idle GPUs worldwide into a high-throughput, low-cost, censorship-resistant compute grid — settled on BNB Chain.
Pool of 20,000,000 OMC. Taking part is free and open now — register entries on the airdrop page; claiming opens with mainnet on 1 January 2027 (UTC+8).
AI demand is exploding — LLM training, Sora-grade video generation, real-time 3D rendering. Centralized clouds answer with prohibitive CapEx, quota limits, and vendor lock-in. OMC answers with a permissionless GPU grid.
Hundreds of thousands of consumer GPUs (RTX 4090/3090), data-center spare capacity and post-PoW mining rigs stay underutilized. OMC aggregates them with zero marginal onboarding cost.
DePINA dedicated BNB Chain L2 rollup for node heartbeats, task allocation and per-second / per-task micropayments is planned but not yet deployed — today these functions run on BNB Chain, with the L1 anchoring final settlement and disputes.
L2 RollupPer-job verification policy: spot-check re-execution, optional 2–3 node redundancy, and hardware TEE for confidential workloads.
Verification policyLLM fine-tuning, LoRA training & inference — without enterprise discounts or GPU quota walls.
24/7 autonomous agent inference, single-node fine-tuning and batch rendering, natively crypto-paid.
Unreal Engine 5, Blender batch rendering and Gaussian Splatting at extreme compute spikes.
Plug idle hardware into the protocol and get paid in OMC for the verified compute you provide.
BNB Chain mainnet — OMC token issuance, staking smart contracts, DAO governance and final dispute arbitration.
L2 Rollup (planned, not deployed) — high-frequency node heartbeats, task-allocation state mapping, state-channel settlements and low-cost micropayments. Until it ships, these functions run on BNB Chain.
Off-chain containerized grid — Docker/Kubernetes sandboxes with CUDA/ROCm acceleration, NVLink inter-GPU links and P2P data transport.
Every job is scored and routed by a weighted bipartite matching engine.
Requesters lock OMC fee tokens in the escrow contract before dispatch.
Nodes matched by VRAM, TFLOPS, bandwidth, latency & reputation score.
Jobs run in isolated containers; outputs verified redundantly & via ZK proofs.
Currency for AI training, inference and 3D rendering jobs.
Collateral proportional to hardware tier before receiving jobs.
Vote on protocol parameters and treasury allocation.
50% of every 8% protocol fee is burned outright — 4% of settled volume. Slashing burns a further 20%.
Quote a reference rate and the contract settles at or below it, refunding the difference. No published price list — the station shows the arithmetic, not an offer.
A provider can only take the job if its staked OMC meets the hardware tier. The same stake pays you back if it fails to deliver.
Verified results release payment automatically. Anything contested goes to the 48-hour window, and slashed collateral is split 50% to you, 30% to the treasury, 20% burned.
Five pairs a day. One photograph is real, the other was generated by a model. No install, no account — just a scoreboard and a streak. A small demo of the question our whole network exists to answer: which compute result do you actually trust?
Claim your share of the 20,000,000 OMC community airdrop, or plug your idle GPU into the testnet today.