@Cysic open-sourced Venus, a graph-based foundation for high-performance zkVMs, based on ZisK, designed to improve proving performance through global optimization and hardware-native execution. @Cysic 团队开源了Venus项目,这是一个基于图架构的高性能zkVM基础框架,代码基于ZisK,旨在通过全局优化和硬件原生执行提升证明性能。
Notes
Venus uses a graph-first architecture, encoding proving as a computational graph instead of traditional HAL.
Enables global optimization across the full proving pipeline, not isolated kernels.
Hardware-native execution maps directly to GPU models like CUDA Graphs.
Improves auditability with structured and inspectable proving flows.
Performance tests show ~6%+ throughput improvement over ZisK 0.16.1.
Provides a foundation for hardware-software co-design across GPU, FPGA, and ASIC.
Venus采用图优先架构,将证明流程编码为计算图,替代传统的HAL方法
支持全局优化,可跨整个证明管道进行性能提升,而非孤立内核
硬件原生执行,直接映射到GPU执行模型如CUDA Graphs
提升审计性,使证明流程结构化且可检查,增强透明度
性能实测显示,相比ZisK 0.16.1,吞吐量提升约6%以上
为硬件-软件协同设计奠定基础,支持GPU、FPGA和ASIC等硬件
零知识证明zkDaily
Q&A Deep Dive 💬今日要点 深入解析 💬
Thu星期四
04.09
2026
Why use a graph-based design? 为什么用图结构?
A graph-based design represents the full proving pipeline uniformly, enabling global optimization and reducing CPU-GPU synchronization overhead, especially for repetitive zkVM workloads. 图结构可以把整个证明流程统一表示,支持全局优化,减少 CPU 和 GPU 同步开销,特别适合 zkVMs 的重复计算任务。
How does Venus optimize sum-check? Venus 如何优化 sum-check?
Venus encodes multiple sum-check rounds into one computation graph, enabling batched scheduling and state reuse while lowering kernel launch cost. Venus 将多轮 sum-check 编码进统一计算图,支持批量调度和状态复用,降低 kernel launch 成本。
Why is Venus better for high-throughput zkVMs? Venus 为什么更适合高吞吐 zkVMs?
It supports cross-stage memory reuse, kernel fusion, and global batch evaluation optimization, reducing latency and bandwidth pressure. 它支持跨阶段 memory reuse、kernel fusion 和 batch evaluation 全局优化,降低延迟和带宽压力。