What is Nosana?

Nosana, a distributed GPU network, allows anyone to rent computing power. Its unique decentralized architecture offers a new model for global users to rent computing power, providing affordable GPUs for AI users and generating income for GPU owners. Designed to validate Nosana’s operation across multiple GPU devices, the GPU Test Grid platform aims to gather comprehensive user feedback to ensure a smooth transition to the mainnet.

By tapping into underutilized GPU resources globally, such as gaming PCs, mining rigs, and MacBooks, Nosana enables on-demand access to GPU computing power and effectively alleviates the market’s GPU resource shortage. Additionally, Nosana is setting up a decentralized network tailored specifically for AI inference (e.g., when ChatGPT responds and replies after a request is input) workloads.

The fundamental objective of Nosana is to integrate AI technology with blockchain, channeling the disruptive capabilities of blockchain into the AI field to fulfill the escalating demand for GPU units in AI inference. The team has observed numerous issues arising from centralized cloud computing in AI development, such as high power consumption in global data centers and high public centralized cloud computing costs. Consequently, Nosana is dedicated to improving resource utilization efficiency and supporting innovation and development in the AI field.

Solana partially supports Nosana’s protocol, offering users multiple protocols. By holding Nosana tokens, users can stake, earn rewards, participate in the open computing market, register nodes, and join the staking pool on the Nosana platform.

Unlike io.net, a DePIN project running on the Solana network, Nosana is more suitable for AI inference applications, such as large language model (LLM) inference. While io.net similarly provides infrastructure for AI computing power by acting as a GPU aggregator and facilitating AI inference, its primary focus is computing clusters, offering users the capability to deploy and rent GPU clusters on io.net.

Roadmap

Nosana’s development plan is structured around five significant release cycles, each named after a galaxy to symbolize its ambition to explore vast opportunities.

Genesis: The Birth of the Platform

The Genesis phase includes the alpha and beta stages, involving launching the GPU network and replicating the entire “demand > supply > demand” process to test the system from end to end. This phase ensures the platform’s stability, reliability, and performance.

Galactica: Expansion Phase

In the second phase, the mainnet will go live. The CLI and SDK will be released, allowing anyone to connect to the network. Additionally, container nodes for consumer-grade GPUs will be introduced. The focus throughout this phase will be on scaling the network and engaging with a broader user base.

Triangulum: Enhanced Interoperability

This version will focus on integrating major machine learning protocols. The release of community connection libraries and official connectors for PyTorch, HuggingFace, TensorFlow, etc., will happen. The platform’s usability and compatibility with existing machine-learning tools and libraries will be boosted in this phase.

Whirlpool: Expanding Hardware Support

The platform’s coverage will further expand to support more types of GPUs, including those from AMD, Intel, and Apple Silicon, aiming to make Nosana the world’s largest computing grid. This phase will present significant technical challenges, as the platform should be made to handle various hardware and use cases.

Sombrero: Supporting Enterprise Users

This version will add support for medium and large enterprises, including fiat access, billing, and team functionalities. As the platform has to attract and support a broader range of users and use cases, this phase will involve technical development and business and community growth.

Source: Nosana

Composition of Nosana

AI Inference

AI inference, a critical component in the AI area, refers to using a trained AI model to analyze and make predictions on new data. This process is the key step in transitioning AI from theory to practical application, as it enables models to function in real-world scenarios such as recognizing images, translating language translation, and recommending content.

However, inference workloads typically demand substantial computational resources, especially when dealing with complex models or large-scale data processing. This necessitates high-performance hardware, such as GPUs (Graphics Processing Units), which provide the parallel processing capabilities required to complete inference tasks swiftly.

Nosana Network, as a platform for running AI inference workloads, offers a solution to meet these needs. Users must first acquire NOS tokens (the currency used for transactions and operations on the Nosana platform). Once holding NOS tokens, users can begin running various AI inference workloads.

Users can utilize the Nosana Network to run AI models like LLama 2 or Stable Diffusion. Stable Diffusion is a popular open-source generative AI model for generating images. With the Nosana Network, users can set up and run these models quickly and conveniently without using expensive hardware.

The detailed documentation provided by Nosana Network assists users in understanding how to leverage platform resources, configure workloads, and perform AI inference computations effectively. This platform is highly valuable for researchers, developers, and any enterprise or individual engaging in AI inference tasks, as it lowers the barrier to entry, increases efficiency, and potentially reduces costs.

Nosana Nodes

Nosana provides financial incentives, allowing users to earn NOS tokens as rewards for participating in its network services. They can join the NOS reward distribution pool if they engage with the test grid and help run AI models. In the initial testing phase, Nosana has allocated a generous pool of up to 250,000 $NOS tokens.

A Nosana node is the software users run to connect their hardware to the grid. To register for the test grid, users need to operate a Nosana node with an NVIDIA GPU. It is important to note that Nosana nodes are currently in the pre-release stage. Nosana officially recommends running nodes in a secure environment or virtual machine and using a Solana address containing only a small amount of SOL.

Source: Nosana

Nosana Protocol

Staking: Nosana’s staking protocol allows users to earn continuous rewards by staking Nosana tokens. This helps maintain the security and decentralization of the Nosana network as well as provides participants with economic incentives.

Staking process: Firstly, users need to possess a certain amount of Nosana tokens. They can then stake these tokens via the Nosana node software or wallet interface. The staking process involves selecting the amount to stake and setting a lock-in period. After staking, users receive xNOS. Once the lock-in period ends, users can opt to exit the staking process and reclaim their staked tokens. Upon exiting, users also receive the accumulated reward tokens.

Source: Nosana

Staking rewards: The longer the staking period, the higher the reward multiplier relative to the staked token amount. Users can stake for a minimum of two weeks (14 days) and up to one year (365 days). After staking, users receive continuous NOS token rewards. The reward amount is calculated based on the staked amount and duration and is automatically distributed to the user’s wallet address periodically.

Additionally, Nosana’s staking protocol employs security measures such as multi-signature and distributed storage to ensure asset safety. Node operators must adhere to certain integrity standards, or they may face penalties.

Nosana pool: The Nosana Pool offers a decentralized liquidity and trading mechanism where users can deposit assets to facilitate the exchange and conversion of various assets. The pool comprises multiple asset types, including NOS tokens and other cryptocurrencies, each with its respective quantity and value ratio within the pool.

Trading: When users wish to exchange an asset, they withdraw the corresponding amount from the pool. Then, the asset ratios will be adjusted dynamically to maintain value equilibrium. Users can also become liquidity providers by depositing assets into the pool, earning a share of the transaction fees as a reward. The pool is collectively managed by nodes in the Nosana network to ensure its security and fairness. The nodes must follow certain protocol rules, such as minimum deposit amounts and fee rates.

Nosana Jobs: Nosana’s Jobs Protocol provides a decentralized distributed computing platform. Job submitters can post various computational tasks, such as machine learning and data analysis, which the nodes in the network will execute. Nodes automatically accept and process tasks in parallel, with the computation results uploaded back to the platform, where they are integrated to form the final output.

As an incentive, job submitters pay NOS token rewards to the nodes executing the tasks, encouraging broader participation. This enhances computational efficiency and offers participants economic benefits, fostering the healthy development of the Nosana network. Additionally, Nosana verifies and evaluates task results to ensure they meet expected standards. Malicious or low-quality nodes may be penalized by deducting NOS token rewards.

Nosana tokens: Nosana’s native token, NOS, is the key asset driving the entire Nosana ecosystem. The total supply of NOS tokens is 10 billion, distributed and incentivized through various channels.

Users can use NOS tokens to pay for network services such as transaction and task fees, stake NOS for continuous rewards, or become liquidity providers to earn a share of transaction fees. The future issuance, burning, and parameter adjustments of NOS tokens will be governed by a decentralized autonomous organization (DAO), achieving complete token decentralization. Nosana aims to maintain the economic health of the network through the supply and demand mechanism of NOS tokens, creating sustained value for participants and driving the prosperity of the Nosana ecosystem.

Distribution: airdropped tokens constitute 5% of the total NOS supply. Users can obtain these during token distribution or through Nosana’s incentivized testnet program. Nodes executing Nosana AI workloads will receive 20% of the total NOS supply, distributed linearly over 24 months (2 years), with a maximum of 20,000,000 tokens. Tokens distributed to the team account for 20% of the total NOS supply, released linearly over 48 months (4 years). Nosana Corporation receives 25% of the total NOS supply, released linearly over 36 months (3 years), with 10% directly released at distribution for platform development. Tokens for fans account for 17% of the total NOS supply, released linearly over 9 months (0.75 years), with an initial 10% unlocked at distribution.

Source: Nosana

Conclusion

Leveraging underutilized GPU resources worldwide, Nosana, an innovative distributed GPU network, provides an affordable computing power rental model for AI users while creating income opportunities for GPU owners. Nosana’s vision is to drive the AI revolution by combining AI and blockchain technologies, addressing issues of high energy consumption and costs associated with centralized cloud computing.

Nosana is evolving into a completely decentralized AI computing platform. Utilizing its distributed GPU network, it provides substantial computing resources for AI users and opens new income streams for GPU owners. As Nosana’s roadmap unfolds, it is set to become a pivotal player in fostering innovation and growth in the AI sector, offering users worldwide a secure, efficient, and cost-effective computing solution. With ongoing technological progress and community growth, Nosana’s future has infinite possibilities.