Open-Sourcing Venus: A New Foundation for High-Performance zkVMs
Over the past months, we published a three-part series on how zkVMs should be built, from architectural philosophy to hardware…

Open-Sourcing Venus: A New Foundation for High-Performance zkVMs
Over the past months, we published a three-part series on how zkVMs should be built, from architectural philosophy to hardware acceleration. Today is the next step.
Why Venus Exists
Zero-knowledge proving systems are at an inflection point. As demand grows, from rollups to verifiable AI, the bottleneck has shifted. The constraint is no longer cryptography. The constraint is systems architecture.
Most zkVMs today follow a CPU-first design, later extended to GPUs through hardware abstraction layers (HAL). While convenient, this model limits deep optimization and creates barriers to scaling across different hardware.
Venus was built to challenge that approach.
It builds on ZisK’s proving foundations while fundamentally rethinking how execution should be structured. The changes affect performance, security auditability, and hardware portability at the system level.
From HAL to Graph: A Paradigm Shift
At the core of Venus is a fundamental shift in ideology: proving should be represented as a graph, not a sequence of function calls.
HAL-based zkVMs dispatch the same code path across CPU and GPU backends. The abstraction simplifies development but forecloses system-wide optimization. You can only ever tune what’s exposed to you.
Venus adopts a graph-first architecture, where the entire proving pipeline is encoded as a computational graph. The benefits of adopting the graph-based architecture are:
- Global optimization: Optimize across the full proving pipeline instead of isolated kernels.
- Hardware-native execution: Map directly to GPU execution models like CUDA Graphs.
- Auditability: Make proving flows structured and inspectable rather than opaque.
This shift mirrors what happened in modern AI systems, where graph execution unlocked significant performance gains.
Performance in Practice
Venus has been built and tested against real workloads. Running against the latest ZisK release (0.16.1), it delivers:
- ~6%+ throughput improvement
- Reduced CPU–GPU synchronization overhead
- Lower kernel launch cost
- More efficient execution in iterative workloads like sumcheck
The design does the work previously left to low-level tuning passes.
Beyond GPUs: A Hardware-Native Design
While GPUs are the immediate target, Venus is designed with a broader vision.
The hardware landscape for ZK proving spans a spectrum: GPUs offer flexibility and availability; FPGAs offer efficiency at the cost of programming complexity; ASICs offer peak performance at the cost of rigidity. Historically, moving along that spectrum meant rewriting your prover.
Venus changes how we think about this spectrum. By encoding proving as a graph:
- Execution becomes predictable and structured
- Hardware mapping becomes cleaner and more direct
- The prover becomes compilable to specialized hardware
This opens the door to a future where zk systems are co-designed with hardware.
Why Open-Source Venus
We believe the next generation of zk systems must be built collaboratively.
Closed systems in this space create trust problems that undermine the entire point of zero-knowledge proofs.
Open-sourcing Venus enables:
- A Shared Execution Model: A graph-based abstraction for ZK proving that others can extend and improve.
- Hardware–Software Co-Design: A foundation for engineers working across: GPU optimization, FPGA prototyping and zk-focused hardware design.
- Transparency and Verifiability: A more inspectable and standardized approach to building ZK systems.
The Bigger Picture: Verifiable Compute
The long-term vision behind Venus is simple:
Make compute verifiable at scale.
As systems evolve:
- AI workloads will require verification
- Cryptographic systems will demand higher throughput
- Distributed compute will need stronger trust guarantees
Venus provides a foundation for this future, where performance and verifiability go hand in hand.
Closing Thoughts
The zkVM trilogy laid out a direction:
- Rethink execution (Graph over HAL)
- Optimize across hardware (GPU → FPGA → ASIC)
Venus brings this vision into a practical, open system. Open-sourcing is just the beginning.


