[Long Thread] Cysic Research Report: The Path of ZK Hardware Acceleration in ComputeFi

Chainfeeds Guide:

Zero-Knowledge Proofs (ZK), as a new generation of cryptographic and scaling infrastructure, have demonstrated vast potential in blockchain scaling, privacy computing, as well as emerging applications such as zkML and cross-chain verification. However, the enormous computational workload and high latency in the proof generation process have become the biggest bottlenecks for industrial adoption.

Source:

Jacob Zhao

Opinion:

Jacob Zhao: GPU has become the core computing resource for both AI and ZK. In the field of Artificial Intelligence (AI), GPU, relying on its powerful parallel computing architecture and mature ecosystem, has almost become the irreplaceable mainstream hardware. Especially in deep learning and neural network training and inference, GPU demonstrates its unparalleled advantages. During training, neural networks require a large number of matrix operations and highly parallel computations, which are exactly the tasks GPUs excel at. Through the CUDA (Compute Unified Device Architecture) programming model and deep learning frameworks such as PyTorch and TensorFlow, GPUs can achieve extremely high computational efficiency. This makes GPUs the ideal choice for large AI models (such as GPT, BERT, etc.), whether during training or inference at deployment. In the ZK field, GPUs also play an important role. Zero-Knowledge Proof (ZK) is a cryptographic algorithm that allows one party to prove the authenticity of certain information without revealing the information itself. In ZK computational tasks, GPUs, with their high parallelism and large throughput, have become the mainstream computing resource, especially in the early stages, where GPUs are the ideal choice due to their lower cost and accessibility. However, the limitations of GPUs are also obvious. Although GPUs have advantages in many ZK proof algorithms, for certain specific tasks, such as large integer modular operations, MSM (Multi-Scalar Multiplication), and FFT/NTT (Fast Fourier Transform / Number Theoretic Transform), the storage bandwidth and memory bandwidth of GPUs become bottlenecks. These computational tasks have very high requirements for storage and bandwidth, and the GPU architecture is not fully optimized for these bottlenecks. Therefore, although GPUs dominate the ZK field, in the long run, more specialized hardware solutions are still inevitable. FPGA (Field Programmable Gate Array), as a programmable hardware, has long been considered a solution between GPU and ASIC. Compared to GPUs, FPGAs offer greater flexibility, allowing developers to program and customize the hardware as needed. This flexibility enables FPGAs to deliver excellent performance in many application scenarios, especially during algorithm development and optimization stages. The hardware programmability of FPGAs makes them ideal for tasks such as ZK proof algorithm verification and iteration, prototype verification, and some low-latency demand scenarios (such as high-frequency trading, 5G base stations). In the ZK field, the application of FPGAs has great potential. As ZK proof algorithms continue to evolve, many research teams adjust and optimize algorithms according to specific needs, and the flexibility of FPGAs precisely meets this demand. Developers can customize hardware architectures for different ZK algorithms to maximize performance. In addition, FPGAs also have certain advantages in terms of power consumption and latency, especially in low-power edge computing scenarios with high computational resource requirements. Cysic Network is a decentralized network based on the ComputeFi concept, aiming to financialize computing resources (such as GPU, ASIC, and mining machines), break the limitations of traditional computing resources, and realize programmable, verifiable, and tradable computing resources. This network is built on the Cosmos SDK (Software Development Kit) and Proof-of-Compute (PoC) mechanism, constructing a decentralized task matching and multi-verification marketplace that uniformly supports computing demands such as ZK proofs, AI inference, mining, and high-performance computing (HPC). Cysic's goal is to provide a new kind of infrastructure for the Web3 ecosystem, especially in the field of computing power, promoting the liquidity and decentralization of computing resources. A key advantage of Cysic Network lies in its unique vertical integration capability. Relying on its self-developed ZK ASIC, GPU clusters, and portable mining machines, Cysic can provide efficient computing resources. By combining the advantages of GPU and ASIC, the Cysic team can provide customized computing power support for different application scenarios, further enhancing the network's flexibility and scalability. In addition, Cysic adopts a dual-token mechanism, namely CYS and CGT. CYS is mainly used for network governance and reward mechanisms, while CGT is used for computing power trading and liquidity support.