What is Dispersed?

Dispersed is a distributed GPU compute network. It is a subnet of the RENDER NETWORK ecosystem focused on machine learning workloads (model inference and training) and other compute beyond the specialized task of rendering images. The GPUs in the network stay productive by distributing data processing tasks securely across GPUs on a global scale ensuring that your GPUs are profitable even if there are no 3D images or movies to render.

The network stays productive by distributing data processing tasks securely across GPUs on a global scale.

Dispersed Service Consumers (that is, end users) essentially rent powerful GPU’s to perform heavy computational tasks at an hourly rate. They can:

• Run heavy computational tasks such as model training, machine learning, analyzing large data sets, simulations etc.
• Forego buying expensive hardware upfront. Rent, execute, and release!

What is the purpose of Dispersed?

Dispersed is designed to open wider access to high-performance computing or grouped computing power. By replacing traditional, top-down control with the circular economy of a decentralizsed model, it distributes value among network participants.

At its core, the platform allows users to scale indefinitely, allocating as many nodes as needed to tackle tasks far beyond the reach of standard consumer machines, and meanwhile allows providers to earn rewards for contributing their spare computing power. By offering powerful GPUs at a competitive cost, Dispersed is an affordable, on-demand solution for heavy computational workloads, massive data crunching, and specialised use-cases that would otherwise be too slow or expensive to run locally.

Who can use Dispersed?

From individuals and hobbyists to researchers, scientists, developers and global enterprise users, anyone in need of compute can leverage powerful GPUs to bridge the gap between high-level ambition and limited local hardware. Dispersed is for anyone whose projects and ideas need a scalable solution to level up.

• Software Engineers / Researchers for model training and/or machine learning.
• Data Scientists for crunching large amounts of data (analyzing large data sets).
• Game Developers for running heavy workloads like NPC/physics simulations.
• Engineers for running math-heavy simulated models.

Real world application

Task	Why	Benefit
Model training	Train large models	Faster than personal computers. No upfront hardware.
Image generation	Generate thousands of images quickly.	Able to scale quickly, no upfront hardware.
Big data data analysis	Crunch massive datasets quickly	Fast, no upfront hardware.
Scientific / Engineering simulations	Run math-heavy models for Physics/Chemistry	Efficient, scalable, no upfront hardware, fast.
Stress testing	Run heavy compute tasks	Able to scale, no need to maintain your own CPU cluster, fast, cost effective.

Real-world user stories

1. Developer Chatbot Training “I need to train a chatbot for my startup, but my laptop can’t handle weeks of computation. I rent GPUs on Dispersed for a few hours, train the model, and save thousands of dollars compared to buying my own hardware.”
2. Data Scientist – Data Analysis “I have a dataset of 50 million records I need to analyze. My PCs can’t handle it. I cook up a GPU instance on Dispersed, crunch the numbers overnight, and only pay for the time I use.”
3. Artist – Image Generation “I want to use neural rendering with my concept art in order to generate thousands of unique, high-fidelity images for a game. Instead of buying a high-end GPU, I rent cloud GPUs on Dispersed for a day and generate all images super fast and cheap!”
4. Engineer – Scientific Simulation “I’m running a fluid dynamics simulation for a design project. It would take my workstation days to run. Using rented GPUs on Dispersed, the simulation only runs for a couple of hours.”
5. Game Developer – Stress Testing “I need to simulate 10,000 players on my multiplayer game server to test performance. I rent GPUs on Dispersed, run the stress test and save thousands of dollars and time without having to buy expensive server hardware.”
6. ML Researcher – Experimenting with Models “I want to try multiple ML model architectures on large datasets. I rent GPUs on Dispersed to test multiple models in parallel, paying only for the hours I use instead of buying hardware for each experiment.”