Golem Network AI/GPU Roadmap Update
Earlier this year, we published our AI/GPU roadmap. We are excited to announce an updated version designed to maximize benefits for the AI market. This revised roadmap focuses on initiatives with market validation, enabling us to allocate resources to the most impactful projects. The updated roadmap is available on our website.
The high demand for computing power in the AI/ML industry, particularly GPU-based, has led us to tailor our solutions to meet these needs. To provide better context for the updated roadmap, here's some background information.
Solutions Tailored for the AI Industry
Below you can find the solutions we would like to focus on based on feedback from developers, startups and AI-based companies.
Modelserve
A service designed to run inference of AI models at scale, affordably. Modelserve offers seamless deployment and inference through scalable endpoints.
Learn more about Modelserve here.
Golem-Workers
This API enables direct access to AI Workers for fine-tuning small AI models on single nodes of the Golem Network. It supports a variety of tasks beyond AI/ML payloads, offering flexibility and power to users.
Learn more about Golem-Workers here.
| Addressing AI Industry Needs | |
| Market Demands | Golem Solution |
|
Data Processing and Preparation for Training Models Fine-Tuning Small AI Models Low-Level Access to Golem GPU and CPU Resources for AI computation and more |
Golem-Workers |
|
Deployment and Inference of AI Models Synthetic training data generation |
Modelserve AI |
Ensuring the Supply of GPU Resources for AI Industry Needs
GPU Provider
This provider enables the rental of a Virtual Machine with GPU, deployment of an image, and performing any calculations requiring parallel processing, not limited to AI/ML tasks.
AI Provider GamerHash
This provider facilitates the execution of specific AI tasks like inference.
| Supply of GPU Resources for AI Needs | |
| Provider | Use Case |
| GPU Provider | Any GPU computation |
| AI Provider GamerHash | AI models inference |
Adapting our Scope to Market Demands
Due to the market validation of our projects in the AI/GPU roadmap, we have adjusted our scope to better respond to market demands. This involves prioritizing projects with demonstrated potential and pausing those with less traction. This refocusing aims to enhance the success rate of our selected initiatives, based on market feedback from developers, startups and AI companies.
Below detailed information about each of the projects reflected on the updated roadmap:
GPU Provider Linux
Since January 29, we have invited 67 individuals to beta test this service from over 300 waitlisted users. By the end of June, we achieved a milestone by opening access to the GPU Provider in an experimental version. For installation instructions, check out our documentation.
We are also working on multi-GPU support for the GPU Provider.
AI Provider Gamerhash
We have developed an internal Proof of Concept. Read about the project in this article.
Modelserve
This service makes it easy and affordable to run AI models inference at scale. We've confirmed interest through market validation with over a dozen companies. Our inference tests, comparing consumer-grade and commercial-grade GPU cards, highlighted the value of consumer-grade GPUs for smaller models like Whisper (audio to text models), Stable Diffusion (diffusers models) and small LLMs.
Beta testing with several startups and companies has also helped us fine-tune the service.
Golem-Workers
Golem-Workers provides direct access to GPU resources on the Golem Network, offering greater flexibility than services focused solely on deployment and inference of AI models.
