Theta EdgeCloud: Kickstarting the Next Generation of AI Computing

The Theta team is currently developing the Theta EdgeCloud, an innovative decentralized software platform designed for periphery computing. This will enable them to provide cost-effective connectivity for developers, researchers, and businesses, enabling them to utilize GPU processing capabilities in the context of AI and video work. The Theta Edge Network enhances the performance.

The first phase of EdgeCloud is scheduled to be unveiled on May 1, 2024, as mentioned in Theta’s roadmap. The foundation upon which EdgeCloud’s AI computing framework is built has been under development for a considerable period of time.

With Mainnet 3.0, the Theta Edge Network debuted in 2021, catering to GPU-intensive video processing requirements for allocation, encoding, and transcoding. The community currently operates nearly 10,000 active edge nodes comprising the global Theta network. It is recognized internationally as one of the most extensive allocated GPU computing power collections.

GPUs with 1,000 nodes and high performance deliver 36,392 TFLOPS, while GPUs with 2,000 nodes and moderate performance deliver 28,145 TFLOPS and 13,002 TFLOPS, respectively. Roughly 77,538 TFLOPS are accumulated in total.

More than 2,500 NVIDIA A100s can be made available due to the combination of Theta’s expansive processing capability and connectivity to its cloud partners, which provides an additional 800+ PetaFLOPS. Training and maintaining the largest language modules (LLMs) requires only this amount. The hybrid cloud decentralized infrastructure of EdgeCloud supports GenAI text-to-image and text-to-video modules, including Stable diffusion and Llama2, as well as cutting-edge text-to-3D components.

Connectivity with allotted GPU processing power cannot modify the AI computing space. Theta filed the first patent application for the Edge computing platform, which is supported by a blockchain network based on smart contracts, in 2021. This paved the way for developing a cutting-edge hybrid computing framework in which computing tasks are recorded on a blockchain and routed to an edge computing node inside a decentralized computing network via a secure peer-to-peer link. The patent was received in September of 2023.

In 2022, many AI associates were formed to work on natural language processing (NLP) and machine learning. This resulted in the partnership with Lavita.AI, FedML, and Google Cloud to create an AI module pipeline for video-to-text applications.

AI developers can quickly select and position popular modules such as stable diffusion and Llama 2 to create AI-powered applications. Additions can be added to the templates to harness sought-after AI modules like chatbots, generative AI, and others. Theta will shortly release their scaled edge node software, providing elite and booster features for EENs with 500,000TFUELs staked. This enables node operators to share rewards from all EdgeCloud AI compute activity.

In the future, AI developers will be able to handle the entire AI pipeline construction process, from prototyping with Jupyter notebooks to training AI modules such as hyperparameter tuning, neural framework search, and model refinement. After that, the models will be ready for deployment over EdgeCloud’s GPU network. Finally, these will be functional on ray clusters.