Welcome

Welcome to the Sundara project! Sundara is a decentralized collaborative mining network designed to optimize GPU resource utilization and provide efficient, cost-effective, and reliable computational power for AI tasks within the Bittensor ecosystem. By leveraging containerization, virtualization, and task scheduling, Sundara enables Bittensor subnets to access high-performance computing resources for various AI applications, with an initial focus on model inference.

Sundara aims to support the growth and scalability of the Bittensor network by providing a decentralized solution for GPU resource aggregation and optimization. As the project matures, we plan to expand our services to other decentralized AI networks, fostering a more accessible and efficient AI computing landscape.

Key Features

• Inference Optimization: Sundara optimizes computing power scheduling at the task level, prioritizing the most suitable resources and allowing a single device to run multiple models, catering to the diverse needs of Bittensor subnets.

• GPU Virtualization: By leveraging vGPU (virtual GPU) and virtualization technology, Sundara partitions large-scale GPU devices into smaller, more efficient units suitable for various inference scenarios within the Bittensor network.

• Parallel Computing: For large-scale training tasks, Sundara offers task slice clusters based on vGPU technology, enabling parallel computing by decomposing and combining tasks based on layers or data.