Equilibrium Infra Bulletin #35: Hardware Acceleration for Plonky, Intent-Centric Programming, The Value Of Blockchains, and more...
Equilibrium Labs builds the state-of-the-art of decentralized infrastructure. We are a global team of ~30 people who tackle challenges around security, privacy, and scaling.
🔍 What Makes Plonky So Fast on the VPU?
⚡️ Topic Summary
Fabric Cryptography is a ZK hardware company building a custom chip for cryptography called the Verifiable Processing Unit (VPU). Fabric has collaborated with Polygon Labs since 2022 and Polygon Labs also recently committed to buying $5m worth of VPU-based server systems to accelerate all ZK-powered Polygon-related protocols. The post dives into how the VPU accelerates Plonky2 and Plonky3 - foundational (open-source) ZK proof systems developed by Polygon Labs - along with other ZK schemes.
The main innovation behind Fabric’s VPU is software-hardware codesign across compute, memory, network, I/O, and software. It supports several cryptographic primitives such as generalized Merkle tree, Plonk, and more, while also supporting 900% more big-integer operations than a typical GPU. Some examples of general optimizations include:
Compute: Adding a RISC-V processor directly onto the VPU chip, which solves the PCIe bottleneck between CPUs and co-processors. Instead of handling workloads like witness generation separately on a server CPU, the VPU processes the full workload on the chip.
Specs of the FC1000 chip, which powers Fabric’s VPU. Each VPU card features 3 FC1000 chips (source). Memory: NTTs (Number Theoretic Transforms) are an important component of ZKPs, but memory-intensive and hence difficult to optimize on standard CPUs, GPUs, and FPGAs. The VPU's architecture includes enough on-chip memory to compute the entire ZK proof (end-to-end workloads) without slow intermediate data transfers.
The VPU can process an entire Plonky 2/3 workload (source) Recursive proving: Originally designed to reduce on-chain verification costs and make proving more parallelizable over transactions, the same techniques can be leveraged to use the memory resources inside the VPU more efficiently.
To speed up Plonky 2/3 specifically, the VPU uses custom instructions designed in close collaboration with the Polygon team that reduce the number of steps required.
🤔 Our Thoughts
Hardware acceleration is an important piece of the puzzle that gets us closer to “real-time proving” and accelerates the adoption of ZKPs. Today there is also more incentive to build ZK hardware with demand coming from a wider range of applications, proof mining (Aleo), and the emergence of decentralized proving networks.
Hardware acceleration also benefits privacy-preserving use cases, as many of them allow proof recursion to be done server-side. Collaborative proving (MPC x ZKP) is also getting more feasible, enabling outsourcing private proofs without sacrificing full privacy.
Besides being used in Polygon’s products, Plonky 2 & 3 are also leveraged by other industry players such as Lita (Valida zkVM) and Succinct (SP1 zkVM). In addition, many of the benefits of the VPU are generalized and not just relevant to Plonky.
💡 Research, Articles & Other Things of Interest
🤓 No Fish Is Too Big for Flash Boys! New MEV research related to frontrunning on DAG-based blockchains.
📚 Permissionless Capital, Global Consensus and New Markets: What ZeePrime Capital considers the main unique properties of blockchains.
📚 A First Look at Intent-Centric Programming: An overview of Pint, the intent-centric programming language developed by Essential.
📚 The Mechanics of Allocating and Slashing Unique Stake: Going deeper into EigenLayer’s new security model.
📚 An Architectural Overview Of Spherenet: How Sphere (in collaboration with Anza) aims to take on SWIFT and upgrade cross-border payments with a permissioned version of the SVM.
🎧 coSNARKs with Ais and Lukas from TACEO: Episode 341 from zeroknowledgeFM covers collaborative SNARKS, i.e. combining MPC and ZK.
🤌 Personal Recommendations From Our Team
📚 Reading: The Code Breaker - Walter Isaacson: A biography about Jennifer Doudna and others who significantly contributed to the development of CRISPR gene editing.
🎧 Listening: Giant Walker - Silhouettes: The title track from British progressive rock band Giant Walker’s second album.
💡 Other: The New Internet: A post by Tailscale on what’s wrong with the current internet and how we could improve it.