ZKsync Airbender: The Fastest Open-Source RISC-V zkVM

·

ZKsync has unveiled Airbender, a high-performance universal ZK prover designed to meet the practical demands of interoperability, decentralization, and scalability. As the fastest open-source RISC-V zkVM, Airbender outperforms leading competitors—proving Ethereum blocks in under 35 seconds on a single GPU and leveraging ZKsync OS to validate average Ethereum block execution in just 17 seconds.

Introducing ZKsync Airbender: The World’s Fastest RISC-V zkVM

Today, we proudly present ZKsync Airbender—a next-generation ZK prover engineered for speed, efficiency, and versatility without compromise.

Airbender isn’t just fast; it’s revolutionary. Benchmark results demonstrate its superiority:

These aren’t just numbers—they signal a future where:
✔ Chains coordinate and settle in real time.
✔ Developers build client apps with native proof generation.
✔ Ecosystems benefit from cheaper, decentralized infrastructure.


Unmatched Performance: By the Numbers

Independent benchmarks confirm Airbender’s dominance in two key phases:

  1. Base Proof Generation:

    • H100 GPU: 21.8 MHz (vs. 3.45 MHz for SP1 Turbo).
    • L4 GPU: 8.5 MHz (vs. 1.1 MHz for RiscZero).
  2. STARK Recursion (End-to-End):

    • 2.5–4x faster than SP1 Turbo.
    • 8.5–11x faster than RiscZero.

👉 Reproduce these benchmarks yourself using our open-source Fibonacci example.


Breakthrough: Near-Real-Time Ethereum Proofs on One GPU

By integrating Airbender with ZKsync OS, we achieved:
17-second Ethereum block proofs (pre-recursion).
35-second end-to-end proofs (with recursion).

Why This Matters:


Use Cases Beyond ZKsync

Airbender’s RISC-V zkVM and open-source design unlock:
🔹 Cross-chain interoperability (e.g., Abstract, Sophon, GRVT).
🔹 ZK-native dApps (gaming, identity, decentralized AI).
🔹 Cost-efficient proofs ($0.0001 per tx—10x cheaper than Boojum).

👉 Build with Airbender today—testnet demos available!


Technical Deep Dive

How Airbender Works

  1. Execution Model:

    • RISC-V 32I+M instruction set.
    • 4M-cycle processing blocks stitched via memory parameters.
  2. Proof Pipeline:

    • Witness commitment → Lookup/Memory → STARK polynomials → DEEP construction → FRI.
  3. Key Optimizations:

    • Mersenne31 field arithmetic.
    • RAM-backed registers.
    • Custom precompiles (e.g., Blake3, big-integer math).

FAQ

Q: How does Airbender compare to SP1 Hypercube?
A: Airbender matches SP1’s performance using 1 GPU vs. 50+ GPUs, thanks to deeper stack integration.

Q: Can I use Airbender for non-EVM chains?
A: Yes! Its RISC-V zkVM supports arbitrary programs—from WASM to custom VMs.

Q: When will Airbender launch on mainnet?
A: Currently in public testnet; mainnet upgrades follow after audits.


Acknowledgments

Credits to RiscZero, StarkWare, and Polygon for pioneering RISC-V zkVMs, STARKs, and Mersenne31 optimizations.

Explore the codebase to dive deeper!


ZKsync: Scaling Ethereum with Zero-Knowledge. 🚀