Skip to content

Latest commit

 

History

History
55 lines (33 loc) · 5.05 KB

RNP-010.md

File metadata and controls

55 lines (33 loc) · 5.05 KB
Raw

RNP-010: Exabits Compute Client

RNP # Title Category Author Created Status
010 Exabits Compute Client Core Jonathan and Zack 19-03-2024 Initial Proposal Vote

Overview

Within the swiftly evolving sphere of technological innovation in Artificial Intelligence and Media, Graphics Processing Units (GPUs) are the base hardware for transforming how we experience digital content. As the boundaries of innovation are continuously challenged, the demand for GPUs has risen exponentially, exposing the market's inefficiencies and adversely affecting costs and accessibility for an indispensable component driving this era.

Exabits, a decentralized cloud computing network, is introducing middleware that accelerates and optimizes GPUs and various layers of the compute stack to alleviate the inefficiencies in the market for complex compute workloads.

Render, a decentralized GPU marketplace servicing "rendering" media workloads, has been a recipient of this rise in demand by creating an ecosystem for consumers and GPU owners to interact fluidly in introducing the benefits of decentralization to the media industry.

THIS RENDER NETWORK PROPOSAL (RNP) AIMS TO LEVERAGE THE EXPERTISE AND NETWORK OF BOTH ECOSYSTEMS BY: INCREASING UTILIZATION OF IDLE GPUs IN THE RENDER NETWORK BY INTEGRATING WITH THE EXABITS NETWORK

Exabits is built on open collaboration and has a shared goal with the Render Network to democratize access to reliable computing resources to support the advancement and innovation in the industry. Exabits capabilities will enable Render compute contributors to achieve a diversified source of revenue by seamlessly integrating their resources to the Exabits platform via Render Network: Resource Aggregation: Render Network GPU suppliers can integrate their resources to a dedicated network within the Exabits ecosystem, to which Exabits will allocate appropriate workloads to this network. A time- and quality-based mechanism will determine the earning potential of these contributions depending on the resources' capabilities. Dividing complex workloads to digestible tasks for varying GPU specifications via resource management and elastic training methods proprietary to Exabits.

Motivation

We are witnessing exponential growth in the demand for GPUs, with no time horizon for supply and demand to reach a healthy equilibrium. In the long term, GPUs will become commoditized rather than a perpetual hunt for the golden goose; for now, the viable solution to hardware shortages is software powered by the Exabits platform. As the industry grows, platforms like Exabits and Render Network must deploy software to scale access to GPUs that contribute to their respective ecosystems.

Render has onboarded GPUs from the community with varying capabilities, each with unique specifications. GPU owners are expected to be unique in their contributions and fickle in nature, favoring opportunities that present the highest earning potentials, affecting the reliability of supply within their networks.

Exabits aims to mitigate the risks associated with a decentralized network, such as varying hardware specifications contributing to the network, fickle suppliers, and exit-en-masse due to changing market conditions, by creating a collaborative incentive mechanism between Render Network and Exabits. Our goal at Exabits is to deploy resources effectively and to strive to achieve a steady inflow of workloads from consumers, increase the uptime of idle GPUs, and create a feedback loop that incentivizes all market participants to contribute their resources to these ecosystems as their de facto platform.

Stakeholders

The proposal impacts all market participants in the Render Network.

Implementation

INCENTIVIZING IDLE GPUs IN THE RENDER NETWORK TO CONTRIBUTE TO THE EXABITS NETWORK WHEN DEMAND IS LOW.

Exhibits network users access Render Network GPU through an API, achieving the following functionalities:

  • Render Network GPUs can be accessed and correctly identified within the Exabits Network;
  • Render is capable of contributing GPUs of which Exabits is responsible for orchestrating AI tasks on Render GPUs in exchange for corresponding revenue;
  • Exabits can execute AI tasks assigned to Render GPUs and return the results;.

We propose a live feed of earning potentials and rails that enable seamless network contribution to support this initiative. We propose a compute client that performs the following function:

  • The compute client will integrate with the Render Network via API;
  • The compute client will enable users to choose Render Network for running the tasks;
  • The compute client is capable of orchestrating AI tasks to the Render Network and producing results to the users;
  • The compute client will provide transparent pricing and enable the users to pay RENDER Credits for the compute.

Exabits will start as soon as possible with the goal towards launching within a quarter.