Apus Network is building a decentralized and trustless GPU network designed for secure, efficient and scalable AI agents inference, aiming to address the key challenges faced by decentralized GPU networks, ensuring GPU integrity, AI model trustworthiness, and verifiable inference results.

A decentralized GPU scenario involves utilizing multiple independent GPU resources within a distributed network to perform computational tasks, rather than relying on a single data center or provider. In this model, anyone can contribute their idle GPU power to the network and receive corresponding rewards, while users in need of computing power can rent these resources on-demand. This model not only improves the efficiency of computing resource utilization but also significantly reduces costs, providing more possibilities for various application scenarios such as AI training and inference, video rendering, and scientific computation. Compared to traditional centralized GPU clusters, decentralized GPU networks offer enhanced fault tolerance, scalability, and privacy protection.

Decentralized GPU not only provides powerful computing capabilities but also ensures higher computational efficiency and reliability through distributed networks. It has a wide range of applications, including but not limited to AI training and inference, high-performance computing (HPC), big data analysis, game graphics rendering, and video processing and encoding. This model makes high-performance computing resources affordable for small and medium-sized enterprises as well as individual developers, thereby fostering innovation and technological advancement.

Using decentralized GPUs as an example, DePHY DID solves the problem of universal and unique identification of devices on the network. It also provides Apus' economic model executors and GPU devices with a uniquely verifiable DID. At the same time, DePHY's restaking mechanism ensures the security of Apus' economic model. Thus led to the creation of the Fisherman role, also protected by DID, to perform multiple verifications on data computation results and execute the economic model.

DePHY DID associates each GPU node with a unique on-chain identity, where metadata can include hardware specifications declared by the device owner. It can combine with RWA, abandoning anonymity to provide real-name information disclosure, which serves as a traceable basis for computation incentives and penalties (Slashing). The core technology of Apus lies in its perfect integration with the DePHY DID solution:

To ensure the economic efficiency of GPU computation, a deterministic model has been designed. This model employs the AlexNet model, known for its significant performance difference between GPU and CPU operation. The process involves randomly selecting several GPU nodes or confirming certain nodes as GPU anchor nodes. Using VRF (Verifiable Random Function), random parameters are generated to create a GPU computation task. The GPU being tested and the anchor nodes execute this task separately and submit their results. In the smart contract, an arbitration process verifies the accuracy of the results and checks the performance metrics and benchmarks of the task to ensure they fall within an acceptable deviation range. Upon successful verification, a mark is added to the DID, proving the computational capability of the GPU node is authentic and trustworthy. To maintain ongoing validity of the verification, the certification might have an expiration date, requiring random revalidation. This mechanism ensures the authenticity and computational reliability of every GPU node in the network, thereby enhancing the reliability and trustworthiness of the entire decentralized GPU network.

Looking ahead, Apus Network will use the Web3 identity system, DePHY ID, to associate each GPU node with a unique on-chain identity. This is not just an ordinary identity credential but a crucial component in establishing the trust foundation of Apus. With a unique digital identity for each GPU node based on DePHY ID, which can be linked to real-world attestation, a range of innovative scenarios will emerge. For instance, GPU nodes can be used for staking and consensus, where node operators need to stake funds for GPU nodes as a financial guarantee for honest service. Nodes with excellent historical records and higher stakes will receive more task allocations and rewards, incentivizing long-term integrity. Different GPU hardware configurations can accept differentiated staking strategies (known as “rainbow staking”), balancing contributions and rewards in a heterogeneous hardware environment.

Moreover, Apus will introduce a series of peripheral services based on the trusted GPU, such as hardware functionality verification contracts, to objectively assess the capabilities of the GPU. Combining the genuine GPU with reliable verification mechanisms, Apus demonstrates infinite potential and a promising future.