In order to effectively match GPU requests to providers, the RNDR network incorporates node reputation history and node power in its automated assignment process. Using Multi-Tier Pricing, creators select from a menu of preferences for cost, speed, and security, enabling the network to optimally sort jobs based on the nature of demand at any given moment. The assignment processes incentives GPU providers to maintain high node success rate and allows creators to optimize their preferences. Job assignment and Multi-tier pricing are based on a tier system. Nodes are grouped into Tiers which are based on attributes like priority and performance benchmarks and reputation score. This enables jobs to be matched with the appropriate region or tier. For example, a user selecting Tier 2 work is matched with any nodes meeting the conditions of Tier 2 nodes prior to any work from Tier 3 being assigned to those nodes. Only after all Tier 2 work is assigned to Tier 3 nodes, will Tier 3 work be assigned.
Job Assignment (within each tier) aims to create a balance across the amount of OB assigned to each job at any given time. To do this, rather than assigning frames in the order they were created or just randomly selecting an available frame, the assignment process looks at the amount of nodes actively working on each job. By doing this, the system is able to determine which job is most in need of power and prevents situations where two jobs with drastically different render times get assigned at the same pace. Additionally, as the network of nodes settles across all available jobs, an equilibrium is met where jobs are getting equal attention and in turn creates a stickiness of the node-to-job pairing - meaning that a node will continue to be paired with a job that it is being processed successfully. As all active jobs have equal power attached to them, a node completing a frame will then cause the job with the frame that was just completed to have the lowest amount of power attached to it - meaning the node will get assigned a frame from the job it just completed. This is advantageous because the node will not need to re-download assets and can spend a greater portion of its time rendering.
The network uses intelligent automated assignment to keep network supply and demand in equilibrium, efficiently matching jobs to nodes.