Source link: https://x.com/Zypher_Network/status/1871383005505396853
From Goth16 to Plonk, the entire zk industry has been dedicated to designing more efficient proof systems. However, how to objectively evaluate the performance of a zk proof system has always been a complex issue, as there is currently no unified benchmark standard. In July this year, StareWare released the next-generation proof system Stwo, which introduced the metric of hashes per second (HPPS) to directly measure system performance. Tests showed that Stwo can prove 620,000 Poseidon2 hashes per second on an M3 laptop, which was once considered the fastest proof system. However, this record was soon broken. In August, Expander released the latest benchmark report, stating that its throughput reached 2,160,000 HHPS on an AMD Ryzen 7950X3D machine, far exceeding Stwo's 960,000 HHPS, becoming the world's fastest prover.
Zypher Network is an all-chain gaming platform driven by AI Agents and ZKP technology, continuously focusing on industry developments and actively introducing more efficient proof systems. More efficient proof systems not only significantly improve the user experience, but also provide unlimited possibilities for building high-performance decentralized applications on-chain. Currently, we are working with the inventor of Expander, Polyhedra Network, to actively explore the use of Expander to build the next-generation on-chain decentralized infrastructure, injecting continuous innovation into the all-chain gaming ecosystem. As a leader in the field of zero-knowledge research and application, Polyhedra has built the foundation technology of trust and scalability in AI and blockchain systems through its EXPchainLayer1. Relying on the efficiency of Expander, they have realized verifiable on-chain AI, providing the industry with new possibilities and solutions.
What is Expander
Most SNARKs use univariate polynomials, which is mainly determined by their underlying polynomial commitment (such as KZG, FRI) schemes, and polynomial commitments are usually the main bottleneck of SNARK performance. In recent years, some new proof systems have emerged, such as the GKR+sumcheck-based schemes, which significantly reduce the prover's complexity by cleverly using multivariate polynomials.
The GKR protocol was proposed in 2008 and further improved by Libra, which proposed the first zero-knowledge proof scheme with strictly linear prover time. Based on this, the Polyhedra team, the authors of Libra, developed the Expander proof system. This system combines the efficient GKR protocol with polynomial commitment schemes based on multivariate polynomials (such as Orion and Brakedown), so that the prover's computation only involves hash operations and field operations that are linearly related to the polynomial size. This proof system does not require FFT operations, so it can use the currently more popular small fields, which also means lower prover overhead and better hardware compatibility. Expander currently mainly supports the following fields:
Writing GKR circuits is relatively complex, so Expander provides the ECC circuit compiler, which allows circuits written in gnark to be transformed into hierarchical circuits, and then use the Expander prover for proof, the entire development process is as shown in the following figure:
Delving into the GKR Protocol
The complete protocol flow is as follows, let:
Zypher's Solution
Expander performs excellently in terms of performance, especially in the field of AI. It is understood that Expander can quickly generate proofs for an 80 billion parameter Llama-3 model in a few minutes using high-performance CPUs, which is an exciting achievement and paves the way for the application of AI Agents.
Zypher Network is an all-chain gaming platform driven by AI Agents and ZKP technology, and is using AI Agents to create intelligent non-player characters, opponents, and powerful game bosses, conducting daily character training, and realizing automated PvP battles.
With the help of Expander, Zypher can not only achieve more complex and practical AI Agents and faster proof generation with fewer computing resources, but also more efficiently verify the behavior of AI agents and players, ensuring the security and fairness of the system, thereby elevating decentralized on-chain gaming to a new level and bringing a higher level of user experience.
