Fenbushi Capital Releases zkVM Benchmark Report: Key Findings and Strategic Insights

zkVM, or Zero-Knowledge Virtual Machine, is a pioneering technology designed to handle complex computational processes off-chain while providing proof of correct execution to the blockchain. This method ensures both trust and privacy by hiding the execution process and only verifying the results. This innovative approach addresses Ethereum(ETH)'s scalability issues.

Ethereum validators currently execute all transactions and computations redundantly, resulting in high fees for complex transactions or applications. zkVM optimizes this by processing computations off-chain and recording a concise proof on-chain, thereby significantly reducing fees.

Blockchain investment firm Fenbushi Capital has released a performance benchmark report on zkVM technology, assessing eight projects based on standardized criteria such as computational speed, memory efficiency, and proof size.

The Benchmark: Methodology and Participants

The eight zkVM projects evaluated were Succinct's SP1, RISC Zero, OpenVM, Pico, ZKM, Jolt, Nexus, and Novanet. The assessment focused on proof generation time, maximum memory usage, and proof size. Real-world calculations, including the 100,000th Fibonacci number, SHA2–2048 hashing, ECDSA signature verification, and the simulation of 100 Ethereum transactions, were used to reflect actual dApp demands.

Here are the highlights:

RISC Zero: Outstanding GPU-Based Performance

RISC Zero demonstrated superior speed and efficiency across all benchmark categories in GPU environments. It consistently recorded the fastest results with minimal memory usage and compact, uniform proof sizes.

For example:

• Generating a proof for the 100,000th Fibonacci number took just 3.6 seconds, using 0.63 GB of GPU memory and producing a proof size of 222 kB.
• The simulation of 100 Ethereum transactions was completed in 7.3 seconds.

RISC Zero's direct GPU integration optimized computations, providing high performance and efficiency.

Succinct's SP1: Exceptional Speed with GPU Focus, but On-Chain Limits

Succinct's SP1 also delivered strong performance in GPU-accelerated environments. It achieved the fastest Fibonacci calculation at 3.4 seconds and delivered promising results in other computations.

However, SP1 struggled with proof size efficiency in operations such as SHA2 hashing or ECDSA verification, with proof sizes exceeding 6 MB, posing limits for on-chain efficiency. Additionally, its intermediate server setup (Moongate) for GPU utilization simplifies maintenance but slightly reduces performance compared to RISC Zero's direct GPU integration.

OpenVM: Competitive Multi-Core CPU Results

OpenVM showcased competitive performance even without GPU acceleration, delivering upper-tier results for most tasks. For example:

• SHA2 hashing was completed in 0.99 seconds, and the Ethereum transaction simulation took 7.6 seconds—results comparable to some GPU-accelerated zkVM solutions.

Its modular architecture, defining task-specific circuits and handling only required computations, improved scalability and reduced memory usage.

Pico, Jolt, and ZKM: Performance Challenges in Specific Scenarios

While Pico and Jolt achieved solid results under certain conditions, performance fluctuated significantly depending on the task. Jolt's new "Lasso Lookup" mechanism aimed to simplify architecture but needed improvement in optimizing cryptographic operations like SHA2 hashing and ECDSA verification.

For instance:

• Jolt required 83 seconds to verify an ECDSA signature using 58 GB of memory.
• ZKM exhibited similar limitations, consuming 84 GB of memory for the same task, rendering it impractical on standard computing hardware.

Novanet and Nexus: Memory Stability at the Expense of Speed

As derivatives of the Nova zkVM family, Novanet and Nexus prioritized memory stability using "folding" techniques that capped memory usage at about 4 GB. However, this focus on resource efficiency resulted in slower computation times. These zkVMs could be well-suited for resource-constrained environments but are less effective for real-time or high-speed uses.

Guidance for zkVM Selection

Fenbushi Capital recommended zkVMs based on specific use cases and environmental constraints:

• Balanced Speed and Efficiency: RISC Zero is optimal for GPU environments, while OpenVM provides a strong CPU-based alternative.
• High-Performance Server Applications: SP1 excels in speed but requires GPUs with more than 24 GB VRAM.
• Scalability Across Functions: OpenVM's modular structure ensures robust performance for various computations.
• Resource-Constrained Environments: Novanet and Nexus excel in low-memory settings.

The report also emphasized evaluating security architectures and developmental maturity. For instance, SP1 has not undergone a comprehensive security audit for its latest version, while Jolt remains in its early beta phase (v0.1) and may face challenges in production deployment.

Implications for Blockchain Scalability

Fenbushi Capital highlighted zkVMs as a promising solution to Ethereum's computational inefficiency and high fees. The benchmark provides valuable insights into the technical maturity and practical feasibility of various zkVM solutions.

The firm stressed that zkVM performance should not rely on a single metric. Instead, factors like proof generation time, resource usage, and security frameworks must be collectively evaluated.

Fenbushi's report is expected to serve as a key reference for blockchain scalability scenarios, including zkVM-based rollups, application-specific coprocessor designs, and decentralized AI computation verification.