Decentralized compute network

gpusharing.org operates a decentralized GPU compute network that connects developers needing computational resources with owners of idle NVIDIA GPUs. The platform provides a single API that accepts tasks such as video processing, image generation, LLM inference, and coding workloads, then routes them to available GPU providers. Developers pay per completed task, while GPU owners earn compensation for each idle cycle their hardware processes.

The system uses an orchestrator layer to authenticate applications, queue tasks, match them to suitable GPUs based on task type and VRAM, and return results via webhook. Providers join by installing a desktop agent that supports workloads including ComfyUI for video/images and Ollama/vLLM for text & code.

Consumers can access the network through vertical apps like CinemaHub.ai, which own the UX and billing and send universal tasks to the API. These apps handle cinematic renders, image generation, LLM completions, and coding tasks through a single POST /tasks interface.

Providers can choose to run Video & Images workloads using ComfyUI or Text & Coding workloads using Ollama/vLLM, or both. The desktop agent runs quietly in the tray, allows pausing anytime, and only downloads models for the modes the user enables.

The architecture consists of three layers: consumers using vertical apps, the orchestrator which authenticates apps and queues tasks in Postgres, and providers running desktop agents on real machines. Apps never touch hardware directly, and providers never touch the users' data.

Every job, whether a cinematic render, an image, an LLM completion, or a coding task, uses the same POST /tasks shape. This universal API allows adding new workloads without touching the provider side.

The network currently supports four task types: video (comfyui_workflow), image (sd_image), LLM (llm_inference), and code (coming soon). The first anchor client in production is CinemaHub.ai, an AI video studio.

Billing is strictly per finished task for consumers, while providers get paid for every idle cycle their card finishes. There is no bidding or lock-in for providers, and they earn for every task their hardware processes.

Some reviewers on Reddit mention that the network is part of a broader movement to decentralize the cloud, one device at a time, similar to other projects like Acurast that leverage consumer hardware for compute.

The orchestrator matches every task to a capable GPU, collects the result, and calls the client back via webhook, ensuring a clean separation between the API and the hardware layer.

The website states that the same network that renders cinematic video today will run coding agents, chatbots, and image tools tomorrow without changing a line on the provider side.

To get started as a provider, users download the host installer, pick the workloads they allow, and start earning for every task their card finishes.

The official website of Decentralized compute network is gpusharing.org.

Decentralized compute network has no reviews on Trustburn yet. Anyone who has used it can write the first review on its Trustburn page.