The Fusion of AI and DePIN: The Rise of Distributed GPU Networks

Since 2023, artificial intelligence and decentralized physical infrastructure networks ( DePIN ) have become two major trends in the Web3 space. Both areas encompass various protocols that serve different needs. This article will explore the intersection of the two and examine the development of related protocols.

In the AI technology stack, the DePIN network empowers AI by providing computing resources. Due to the GPU shortage resulting from the development of large tech companies, other developers find it difficult to obtain sufficient GPUs for AI model training. The traditional approach is to choose centralized cloud service providers, which requires signing inflexible long-term contracts and is inefficient.

The DePIN network provides a more flexible and cost-effective alternative. It uses token rewards to incentivize resource contributions, crowd-sourcing GPU resources from individual owners into the network, forming a unified supply for users who need access to hardware. This not only offers developers customizability and on-demand access, but also provides additional income for GPU owners.

There are various AI DePIN networks on the market, each with its own characteristics. Below we will explore the features and goals of several major projects:

Overview of AI DePIN Network

Render is a pioneer in the P2P network providing GPU computing power, initially focused on content creation rendering, and later expanded to AI computing tasks.

Interesting point:

• Founded by the cloud graphics company OTOY, which has Oscar-winning technology.
• GPU networks have been used by major companies in the entertainment industry.
• Collaborate with Stability AI and integrate AI models with 3D content rendering.
• Approve multiple computing clients and integrate more GPUs from the DePIN network.

Akash is positioned as a "super cloud" alternative to traditional cloud platforms, supporting storage, GPU, and CPU computing.

Interesting part:

• A wide range of computing tasks from general computing to network hosting
• AkashML allows running a large number of models on Hugging Face.
• Hosted some well-known AI applications, such as Mistral AI's LLM chatbot.
• Supports metaverse, AI deployment, and federated learning platforms

io.net provides access to distributed GPU cloud clusters specifically designed for AI and ML use cases.

Interesting aspect:

• The IO-SDK is compatible with mainstream AI frameworks and can automatically scale according to needs.
• Supports the creation of 3 different types of clusters, can be quickly launched.
• Integrate GPU resources in collaboration with other DePIN networks

Gensyn provides GPU computing power focused on machine learning and deep learning.

Interesting point:

• Significantly reduce GPU computing costs
• Support fine-tuning of pre-trained base models
• Provide a decentralized, globally shared foundational model

Aethir specializes in providing enterprise-grade GPUs, primarily used in fields such as AI, machine learning, and cloud gaming.

Interesting point:

• Expand into the cloud phone service field
• Establish partnerships with large Web2 companies such as NVIDIA
• There are multiple partners in the Web3 field.

Phala Network serves as the execution layer for Web3 AI solutions, using the trusted execution environment ( TEE ) to address privacy issues.

Interesting part:

• Act as a verifiable computing coprocessor protocol
• AI agent contracts can access top large language models
• Multiple proof systems will be supported in the future.
• Plan to support H100 and other TEE GPUs

Project Comparison

| | Render | Akash | io.net | Gensyn | Aethir | Phala | |--------|-------------|------------------|---------------------|---------|---------------|----------| | Hardware | GPU & CPU | GPU & CPU | GPU & CPU | GPU | GPU | CPU | | Business Focus | Graphics Rendering and AI | Cloud Computing, Rendering and AI | AI | AI | Artificial Intelligence, Cloud Gaming and Telecommunications | On-chain AI Execution | | AI Task Type | Inference | Both | Both | Training | Training | Execution | | Work Pricing | Performance-Based Pricing | Reverse Auction | Market Pricing | Market Pricing | Bidding System | Equity Calculation | | Blockchain | Solana | Cosmos | Solana | Gensyn | Arbitrum | Polkadot | | Data Privacy | Encryption& Hashing | mTLS Authentication | Data Encryption | Secure Mapping | Encryption | TEE | | Work Fees | 0.5-5% per job | 20% USDC, 4% AKT | 2% USDC, 0.25% reserve fee | Low fees | 20% per session | Proportional to staked amount | | Security | Proof of Render | Proof of Stake | Proof of Computation | Proof of Stake | Proof of Render Capability | Inherited from Relay Chain | | Completion Proof | - | - | Time-Lock Proof | Learning Proof | Rendering Work Proof | TEE Proof | | Quality Assurance | Dispute | - | - | Verifier and Reporter | Checker Node | Remote Proof | | GPU Cluster | No | Yes | Yes | Yes | Yes | No |

Importance

Availability of Cluster and Parallel Computing

The distributed computing framework has implemented GPU clusters, which can train complex AI models more efficiently. Most projects have now integrated clusters for parallel computing. io.net has successfully deployed a large number of clusters. Although Render does not support clusters, its operation is similar. Phala currently only supports CPUs, but allows for the clustering of CPU workers.

Data Privacy

Developing AI models requires the use of large amounts of data, which may involve sensitive information. Most projects use some form of data encryption to protect privacy. io.net has launched Fully Homomorphic Encryption (FHE), allowing encrypted data to be processed without the need for decryption. Phala Network introduces Trusted Execution Environment (TEE), which prevents external processes from accessing or modifying data.

Completion Proof and Quality Inspection

Some projects will generate proofs to indicate that work has been completed and undergo quality checks. Gensyn and Aethir use validators and check nodes to ensure service quality. Render recommends using a dispute resolution process. A TEE proof will be generated after Phala is completed.

Hardware Statistics

| | Render | Akash | io.net | Gensyn | Aethir | Phala | |-------------|--------|-------|--------|------------|------------|--------| | Number of GPUs | 5600 | 384 | 38177 | - | 40000+ | - | | Number of CPUs | 114 | 14672 | 5433 | - | - | 30000+ | | H100/A100 Quantity | - | 157 | 2330 | - | 2000+ | - | | H100 Cost/Hour | - | $1.46 | $1.19 | - | - | - | | A100 Cost/Hour | - | $1.37 | $1.50 | $0.55 ( Estimated ) | $0.33 ( Estimated ) | - |

Requirements for High-Performance GPUs

AI model training requires the best-performing GPUs, such as Nvidia's A100 and H100. The inference performance of the H100 is 4 times faster than that of the A100, making it the preferred choice. Decentralized GPU market providers must compete with their Web2 counterparts not only by offering lower prices but also by meeting actual market demands.

io.net and Aethir each received over 2000 H100/A100 units, which are more suitable for large model computations. The costs of these decentralized GPU services are already much lower than centralized services.

Despite the memory limitations of GPU clusters connected over the network, they remain a powerful choice for users who require flexibility. By offering more cost-effective alternatives, these networks create opportunities to build more AI and ML use cases.

provides consumer-grade GPU/CPU

In addition to high-end GPUs, consumer-grade GPUs and CPUs also play a role in AI model development. Considering that a large number of consumer GPU resources are idle, some projects also serve this market by developing their own niche.

Conclusion

The AI DePIN field is still relatively new and faces challenges. However, the number of tasks executed on these decentralized GPU networks and the hardware significantly increased, highlighting the demand for alternatives to Web2 cloud provider hardware resources.

Looking ahead, the AI market has a broad prospect, and these decentralized GPU networks will play a key role in providing developers with cost-effective computing alternatives. By continuously bridging the gap between demand and supply, these networks will make significant contributions to the future landscape of AI and computing infrastructure.