Chainfeeds Briefing:
zkVM will become the foundational technology for Ethereum scaling - but how do they actually perform in real-world environments?
Article Source:
https://x.com/tcoratger/status/1937990534293033100
Article Author:
tcoratger
Perspective:
tcoratger: Fenbushi's test project includes 8 zkVMs: SP1, RISC Zero, OpenVM, Pico, ZKM, Jolt, Nexus, and Novanet. The tests covered 4 types of programs: calculating the 100,000th Fibonacci number, obtaining SHA2-2048 hash, secp256k1 ECDSA signature verification, and simulating 100 Ethereum transfers. The fastest prover runtimes were RISC Zero (GPU) and SP1 (GPU), which performed consistently well in various scenarios; OpenVM (CPU) was also highly competitive without GPU. Jolt and Pico's performance varied depending on the test task. Fibonacci benchmark (100,000th item): A computationally intensive, low-memory task. The fastest were: SP1 (GPU): 3.4 seconds. RISC Zero (GPU): 3.6 seconds. OpenVM (CPU): 7.5 seconds. This shows that GPU-accelerated zkVMs lead in such tasks. Simulating 100 Ethereum transfers (EVM simulation): RISC Zero (GPU) 7.3 seconds; OpenVM (CPU) 7.6 seconds; SP1 (GPU) 13 seconds; Jolt (GPU) 82 seconds. Achieving such speeds with ordinary hardware is impressive. Scalability test (Fibonacci from 10 to 100k): SP1 (GPU) had a moderate performance decline; RISC Zero and OpenVM had moderate slowdown; Jolt, Pico, ZKM experienced sharp performance drops. This shows that execution continuation and batch proof mechanisms are crucial for handling larger-scale tasks. Proof size: Smaller proofs mean lower on-chain verification costs. RISC Zero: around 222KB. Jolt: around 232KB. SP1 Fibonacci: 1.8MB, potentially up to several MB in other tests. Novanet/Nexus: tens of MB. Therefore, platforms like SP1 and Nexus trade larger proofs for flexibility or higher abstraction capabilities.
Content Source
