Problem Solving

Some Issues in Web2:

What is the Problem we are solving?

Problem 1: High Costs and Lack of Flexibility in Centralized Cloud Services

In the AI era, numerous models require training, ranging from basic image recognition to complex natural language processing tasks. These training efforts often necessitate adjusting various parameters to meet specific requirements. However, the large GPUs and CPUs provided by traditional centralized cloud service providers are not only expensive but also lack the flexibility for customization and adjustment. This poses significant financial challenges for small to medium enterprises and independent researchers, hindering the speed and breadth of innovation in the field.

Problem 2: Underutilized High-Performance User Equipment

On the other hand, within cryptocurrency communities, many users possess high-end GPU and CPU devices that remain idle for extended periods. These powerful computational resources are not being fully utilized, representing a waste of personal assets and a missed opportunity to leverage societal computational power. Under the current framework, there is no straightforward and effective way for these users to share or lease their computational power to others who need it, thus missing out on potential earnings and the chance to contribute to research and development in AI.

How We Solve These Problems

To address the aforementioned challenges, our solution is to provide a decentralized computational power leasing platform that fully utilizes idle user devices, offering unused computational resources to those who genuinely need them. This approach creates a win-win scenario for both resource providers and lessees. Leveraging the strong community of EVM and Ton blockchain, our dApp will initially launch as a web version, followed by development within the Telegram mini-app environment. This integration will enable all Telegram users to conveniently rent out or lease others' machines directly within Telegram.

Our dApp is scheduled to go live before April 1st, commencing epoch 1 of our early rewards program. All machines joining our network will be rewarded based on their online duration, encouraging participation and contribution to the network. This strategic rollout aims to streamline the process of computational resource sharing, making it accessible, efficient, and profitable for participants, thereby fostering a vibrant ecosystem for AI and machine learning development.

Some Issues in Web3:

Problem 1 : Internet Speed

When using idle user resources, insufficient bandwidth often prevents datasets from being properly uploaded during AI training. This issue arises because AI training typically requires the transfer of large datasets to the computational resource. When the internet speed is slow or bandwidth is limited, these data transfers become bottlenecks, causing delays or failures in the training process. This can significantly hinder the efficiency and effectiveness of AI model development, as prolonged upload times or incomplete uploads disrupt the workflow and can lead to suboptimal training results.

Our solution:

Once a user joins our TonGPU network, they must first undergo our node certification process, which includes periodic internet speed tests. If the upload and download bandwidth do not meet a certain threshold, we will refuse to add them to our TonGPU node network. Consequently, token rewards will not be distributed to that node.

This is similar to staking 32 ETH to become an Ethereum staking node; you must ensure your server meets the online requirements to qualify as a node. We will announce the full node requirement specifications later.

Problem 2 : Absolute computational power

The computational power of machines with the same model can vary between different users. Setting a uniform price for a specific model may be unfair.

Our solution:

This issue is addressed through our node verification strategy. Before officially joining the TonGPU network, machines will undergo a professional benchmark test using the official Cinebench tool to ensure consistent absolute computational power. Pricing will be based on this absolute computational power. After joining the TonGPU network, we will also periodically verify the quality of the nodes, and those that do not meet the standards will be removed from the network.