Multi-Tier Pricing System (MTP)

The Multi-Tier Pricing System (MTP) is StarMiner’s dynamic pricing framework for GPU computation. It replaces static billing models and unpredictable auction mechanisms with a programmable pricing engine that adjusts based on real-time network conditions, task parameters, and node performance.

MTP ensures that computing power is priced fairly, distributed efficiently, and remains accessible to a broad range of users from budget-conscious researchers to enterprise-grade AI labs.


Why MTP Exists

In decentralized compute networks, one of the biggest challenges is pricing diversity:

  • Not all tasks require the same performance

  • Not all providers offer equal infrastructure

  • Not all users can tolerate the same latency or reliability

MTP solves this by establishing multiple pricing tiers and applying real-time adjustments to match compute supply with demand, based on context, not just cost.


Core Pricing Variables

MTP calculates pricing using a blend of on-chain and off-chain inputs:

  1. Hardware Class

    • Higher-performing GPUs (e.g., A100, H100) command premium pricing tiers.

    • Nodes with verified TEEs or sustainable energy credentials may also be priced higher.

  2. Task Urgency

    • Tasks marked as time-sensitive (e.g., real-time inference, priority training) are charged higher AGPU rates for prioritized routing.

  3. Network Load

    • When the network is congested or GPU availability is low, MTP increases pricing to reflect scarcity.

    • During off-peak periods, AGPU requirements are reduced to incentivize demand.

  4. Job Type Complexity

    • Tasks involving multi-GPU parallelism, long runtimes, or memory-intensive models are factored into the tier curve.

  5. Geographic Routing Preferences

    • Users may choose lower-cost regions for less urgent jobs, or pay more for localized execution (e.g., jurisdictional compliance, data sovereignty).


Tier Structure Overview

StarMiner maintains three main pricing tiers:

Tier
Description
Use Cases
Cost Level

Economy

Low-latency tolerance, flexible timing

Batch inference, render queues

Low

Standard

Balanced SLA and latency expectations

Model training, real-time prediction

Medium

Premium

Fastest routing, guaranteed resources

Enterprise workloads, high-priority AI

High

Each tier operates with predictive AGPU price bands that shift based on demand similar to cloud spot pricing, but transparently encoded on-chain.


Price Discovery and Execution

  • When a task is submitted, it is automatically categorized and assigned to a pricing tier.

  • The MTP engine determines the current price band based on:

    • Active network demand

    • Supply in that tier

    • Node availability and historical performance

  • The task is accepted only if the user has sufficient AGPU balance to meet tier-specific pricing.

If tier conditions change mid-job, pricing remains locked for the duration of execution, avoiding surprises or volatility mid-compute.


Benefits of the MTP System

  • Fairness: Ensures that users only pay for what they actually use based on urgency and hardware.

  • Market Efficiency: Routes resources toward the highest-value use cases without central interference.

  • Predictability: Users know in advance what AGPU cost bands to expect for their job class.

  • Incentive Alignment: Encourages nodes to upgrade hardware or improve uptime to qualify for higher-paying jobs.


DAO Control and Future Customization

MTP parameters (e.g., tier thresholds, region multipliers, SLA scoring weights) are governed by AMAX AI token holders. This allows the protocol to:

  • Adjust incentive curves

  • Introduce domain-specific tiers (e.g., biomedical compute, defense)

  • Integrate sustainable compute discounts or compliance-based pricing structures

Through governance, the MTP system can evolve into a flexible economic engine, tailored to meet the changing needs of AI and decentralized compute infrastructure.

Last updated