Until recently, many zkVMs have remained experimental, slow, and costly to run at scale. With the release of zkMIPS 1.0, a supremely performant production-grade zkVM, ZKM aims to change that.
Built on MIPS32r2 and optimized for recursive STARK proving via Plonky3, zkMIPS 1.0 is a full-stack rewrite of the zkVM, delivering 6x to 19x performance gains over its previous version (0.3.0).
Here, we present the benchmark results comparing zkMIPS 1.0 with zkMIPS 0.3, ROVM 2.0.1, and SP1 4.1.1 across standard workloads. All benchmarks were run on an AWS r6a.8xlarge (CPU) instance.
A basic recursive computation used to evaluate performance over long chains of dependent instructions.
zkMIPS 1.0 consistently outperforms SP1 across all sizes, peaking at a 199% improvement at small input sizes. At scale (n = 58,218), zkMIPS still shows a >100% performance edge, making it the fastest zkVM on CPU for this class.
A widely used hash function benchmarked across increasing input sizes.
zkMIPS 1.0 outpaces SP1 by a wide margin - over 200% on 256-byte inputs, and still 158% faster at 2048 bytes. This demonstrates zkMIPS’ optimized memory model and trace compression under real cryptographic workloads.
Known for its complexity, SHA3 is a tougher proving workload.
zkMIPS 1.0 holds its advantage even here, delivering >200% improvements on small inputs and maintaining >100% edge on 2048 bytes. This confirms consistent efficiency across both lightweight and heavy cryptographic operations.
When heavy ops like Keccak are abstracted into precompiles:
zkMIPS maintains a strong lead - up to 56% faster than SP1 - showing how its precompile architecture reduces prover burden. This highlights the benefit of ZKM’s modular chip design and native precompile integration.
Stress tests programs with large memory access ranges.
zkMIPS 1.0 delivers a 74% speedup over SP1, showing strong performance in workloads requiring extensive memory reads/writes - a result of its multiset hashing model and efficient memory consistency checking.
A chained cryptographic workload useful for rollup compression and recursive proof aggregation.
zkMIPS 1.0 achieves an 81% improvement over SP1, suggesting strong recursive performance and layout compactness.
One of the most demanding proving tasks.
Even here, zkMIPS performs competitively, matching SP1 on large workloads and outperforming it by 22% when precompiles are used. This reinforces the advantage of offloading complexity to dedicated logic circuits.
Across every category tested - from simple arithmetic to full cryptographic chains - zkMIPS 1.0 is now the most performant CPU-based zkVM in production. With GPU support underway and impressive results starting to show on Ethproofs, ZKM is positioned to lead the field in scalable, low-cost verifiable compute.
Developers and researchers can replicate these benchmarks or explore more in the open-source zkvm-benchmarks suite.
For more details, visit docs.zkm.io.
Until recently, many zkVMs have remained experimental, slow, and costly to run at scale. With the release of zkMIPS 1.0, a supremely performant production-grade zkVM, ZKM aims to change that.
Built on MIPS32r2 and optimized for recursive STARK proving via Plonky3, zkMIPS 1.0 is a full-stack rewrite of the zkVM, delivering 6x to 19x performance gains over its previous version (0.3.0).
Here, we present the benchmark results comparing zkMIPS 1.0 with zkMIPS 0.3, ROVM 2.0.1, and SP1 4.1.1 across standard workloads. All benchmarks were run on an AWS r6a.8xlarge (CPU) instance.
A basic recursive computation used to evaluate performance over long chains of dependent instructions.
zkMIPS 1.0 consistently outperforms SP1 across all sizes, peaking at a 199% improvement at small input sizes. At scale (n = 58,218), zkMIPS still shows a >100% performance edge, making it the fastest zkVM on CPU for this class.
A widely used hash function benchmarked across increasing input sizes.
zkMIPS 1.0 outpaces SP1 by a wide margin - over 200% on 256-byte inputs, and still 158% faster at 2048 bytes. This demonstrates zkMIPS’ optimized memory model and trace compression under real cryptographic workloads.
Known for its complexity, SHA3 is a tougher proving workload.
zkMIPS 1.0 holds its advantage even here, delivering >200% improvements on small inputs and maintaining >100% edge on 2048 bytes. This confirms consistent efficiency across both lightweight and heavy cryptographic operations.
When heavy ops like Keccak are abstracted into precompiles:
zkMIPS maintains a strong lead - up to 56% faster than SP1 - showing how its precompile architecture reduces prover burden. This highlights the benefit of ZKM’s modular chip design and native precompile integration.
Stress tests programs with large memory access ranges.
zkMIPS 1.0 delivers a 74% speedup over SP1, showing strong performance in workloads requiring extensive memory reads/writes - a result of its multiset hashing model and efficient memory consistency checking.
A chained cryptographic workload useful for rollup compression and recursive proof aggregation.
zkMIPS 1.0 achieves an 81% improvement over SP1, suggesting strong recursive performance and layout compactness.
One of the most demanding proving tasks.
Even here, zkMIPS performs competitively, matching SP1 on large workloads and outperforming it by 22% when precompiles are used. This reinforces the advantage of offloading complexity to dedicated logic circuits.
Across every category tested - from simple arithmetic to full cryptographic chains - zkMIPS 1.0 is now the most performant CPU-based zkVM in production. With GPU support underway and impressive results starting to show on Ethproofs, ZKM is positioned to lead the field in scalable, low-cost verifiable compute.
Developers and researchers can replicate these benchmarks or explore more in the open-source zkvm-benchmarks suite.
For more details, visit docs.zkm.io.
