AI Mining Precompiles

GPU attestation and compute verification

AI Mining Precompiles

Native precompiles for AI compute attestation and mining operations.

Address Range

AI precompiles occupy addresses 0x0300 through 0x0302.

Precompiles

Address	Precompile	Gas	Description
`0x0300`	AI_MINING	100,000	ML-DSA signature verification, rewards
`0x0301`	NV_TRUST	50,000	GPU TEE attestation (NVTrust)
`0x0302`	MODEL_REGISTRY	25,000	AI model registration/verification

Hardware Trust Tiers

Level	Hardware	Multiplier	Description
Sovereign (4)	Blackwell	1.5x	Full TEE, highest trust
Confidential (3)	H100/TDX/SEV	1.0x	Hardware-backed confidential compute
Private (2)	SGX/A100	0.5x	Trusted execution environment
Public (1)	Consumer GPU	0.25x	Stake-based soft attestation

Usage

import {IAIMining} from "@luxfi/precompile/ai/IAIMining.sol";
import {INVTrust} from "@luxfi/precompile/ai/INVTrust.sol";

contract ComputeProvider {
    IAIMining constant MINING = IAIMining(0x0300);
    INVTrust constant NVTRUST = INVTrust(0x0301);

    function startSession(bytes calldata teeQuote) external returns (bytes32 sessionId) {
        // Verify TEE quote from GPU
        require(NVTRUST.verifyQuote(teeQuote), "Invalid TEE quote");

        // Start mining session
        sessionId = MINING.startSession(msg.sender, teeQuote);
    }

    function heartbeat(bytes32 sessionId) external returns (uint256 reward) {
        // Submit proof of work and receive rewards
        return MINING.heartbeat(sessionId);
    }
}

NVTrust Verification

The NVTrust precompile verifies NVIDIA Confidential Computing attestations:

// Verify a TEE quote
bool valid = NVTRUST.verifyQuote(teeQuote);

// Get hardware info from quote
(
    uint8 trustLevel,
    bytes32 gpuId,
    uint256 computeUnits
) = NVTRUST.parseQuote(teeQuote);

Model Registry

import {IModelRegistry} from "@luxfi/precompile/ai/IModelRegistry.sol";

IModelRegistry constant REGISTRY = IModelRegistry(0x0302);

// Register a model
bytes32 modelId = REGISTRY.register(
    "llama-3-70b",
    modelHash,
    metadata
);

// Verify model execution
bool valid = REGISTRY.verifyExecution(modelId, inputHash, outputHash, proof);

Lux Standard Contracts - AI token and mining contracts
A-Chain Documentation - Attestation chain details
LP-2000 - AI Mining specification

AI Mining Precompiles

On this page