Fenbushi Capital recently released a comprehensive blog post on benchmarking eight different zkVMs. () Here is a summary, including key results and take-aways.

What are zkVMs? zkVMs are designed to cryptographically verify the correctness of program execution without disclosing inputs or intermediate computational states. With regards to proving Ethereum blocks, zkVMs synthesize three fundamental cryptographic assurances to verify all state transitions: · Read-write memory consistency proofs · Instruction encoding proofs · Instruction proofs

How can zkVMs scale Ethereum? Instead of re-executing every transaction, validators could verify a succinct cryptographic proof, which is very efficient, and could increase the throughput of the L1 significantly.

Fenbushi’s blog introduces a standardized benchmarking framework aimed at making zkVM performance more comparable. It evaluates eight zkVMs — SP1, RISC Zero, OpenVM, Pico, ZKM, Jolt, Nexus, and Novanet — across four computational tasks and three performance metrics (prover time, proof size, and peak RAM utilization).

Among the different proof systems, there are a few predominant ones, such as FRI-STARK-based, Nova-based, Lasso lookup-based, and GKR. zkVM architectures can be further categorized into two predominant paradigms: vRAM-style and modular style.

Hardware and test programs used for benchmarking: Benchmarks were conducted on a Linux system equipped with Ubuntu 24.04, 8 virtual CPUs, 192GB of RAM, and an NVIDIA RTX 5090 GPU with 32GB of VRAM. The four test programs utilized for evaluation comprised: 1. Calculation of the 100,000th Fibonacci number. 2. SHA2–2048 hash computation. 3. ECDSA signature verification using the secp256k1 curve. 4. Simulation of 100 Ethereum Transfer transactions (ETHTransfer).

Proving time for SHA2–2048: For cryptographic operations like SHA2, precompile-based acceleration is a common optimization strategy.

Proving times for 100 ETHTransfer Transactions: Summary: RISC Zero is slightly slower than SP1 in the Fibonacci test, but stands out in the other three test programs as the clear winner.

Memory efficiency and peak memory usage: SP1 (GPU), RISC Zero (GPU) demonstrated relatively constant memory consumption irrespective of the test program.

Proof sizes (in kB) have been observed as follows: RISC Zero and Jolt consistently produced among the most compact proof sizes across the evaluated benchmarks.

Performance summary: Overall, RISC Zero’s performance demonstrates exceptional consistency, while SP1, OpenVM, Pico, and Jolt each reached great performance in some individual categories.

Conclusions: RISC Zero, OpenVM, and SP1 demonstrate particularly robust performance, especially in executing EVM-related computational tasks, which makes them great candidates for scaling Ethereum. RISC Zero shows exceptional efficiency across key metrics that are relevant to blockchain applications, and all this with efficient memory utilization and compact proof size. Congrats to @RiscZero, @openvm_org and @SuccinctLabs!