It’s been exactly one month since our @eth_proofs Summit, organized by House of ZK and hosted during @BerBlockWeek - where the leading builders across the zkVM ecosystem and @ethereumfndn researchers came together to confront the hard questions around benchmarks, proof verification, and infrastructure maturity.
As the use of zero-knowledge proofs expands across @Ethereum and beyond, we set out to discover how to compare proving systems meaningfully and standardize what we’re actually measuring.
Across twelve sessions - spanning benchmark methodologies, zkVM design, and execution-model rethinks - the summit examined how Ethereum can build toward a world of real-time, decentralized proving with reproducible metrics and shared standards.
This article aims to capture the key themes, open questions, and areas of early convergence.
The summit opened with a broad keynote by @corcoranwill, Research Coordinator at the EF, framing Ethproofs as a public benchmarking initiative for zk infrastructure. Its purpose: to create open, adversarial, and reproducible benchmarks that go beyond performance marketing: Opening Keynote
The topic then moved on to a deep-dive into the physical limits of proof generation. In the panel on Real-Time Proving and the Limits of Parallelization, excellently moderated by @alexanderlhicks (EF), leading thnkers from @0xPolygon, @RISCZero, @0xLita, and @SuccinctLabs discussed memory bottlenecks, recursion depth, and the difficulty of coordinating distributed prover orchestration under sub-12s latency constraints: Panel 1
As @jbaylina emphasized, real-time proving changes what’s required of consensus. Proof size and generation time become critical to validator propagation assumptions, and hardware limitations begin to influence protocol parameters like “delta”. This theme recurred across sessions: zero-knowledge is no longer just a cryptographic primitive - it’s a systems-level constraint.
Following this, @NorbertVadas introduced @thezkcloud’s approach to general-purpose zk infrastructure. His talk focused on the challenges of building prover networks that can support verifiable workloads from multiple zkVMs, and the role of shared APIs and interfaces in scaling zk compute: Norbert’s Keynote
One of the most revealing panels, zkVM Design and Proof System Tradeoffs, explored how ISA selection and constraint-system choices shape real-world performance: Panel 2
Speakers from @ProjectZKM, @LineaBuild, @StarkWareLtd, and @zksync laid out the consequences of their respective design paths - whether through fixed-width RISC-based ISAs, custom constraint reductions, or low-abstraction circuits optimized for GPU throughput.
This panel, moderated by @asn_d6 (EF), revealed a recurring idea: standardizing verification logic across zkVMs may enable proof markets to emerge, but architectural divergence remains critical for specialized applications. In other words: convergence in verification, diversity in proving.
Following the panel, @alexand_belling expanded on Linea’s constraint system optimizations, proving backend design, and ongoing work in proof compression. He emphasized tradeoffs between SNARK recursion and STARK parallelism, and how protocol-level design affects the latency/cost curve: Alexandre Keynote
The session on Benchmarking Methodologies and Verification Standards got to the core of the Ethproofs initiative: Panel 3
Moderated by @alicelingl (House of ZK) with speakers from StarkWare, zkCloud, @NexusLabs, and @Khovr of the Ethereum Foundation, the panel surfaced clear priorities:
This is where Ethproofs aims to make the most impact: moving zk infrastructure beyond speed claims, toward a model where metrics are public, reproducible, and adversarially tested.
The mid-point of the summit was characterized by a series of keynotes by @l2beat's @donnoh_eth, ZKM's @sd_eigen, and the returning Jordi Baylina.
Jordi introduced ZisK, Polygon’s fully open-source zkVM and proving toolstack designed for high-throughput, verifiable execution in real-time environments. ZisK is a 64-bit, RISC-V-based architecture with support for high-level languages like Rust - offering full-stack control from low-level primitives to aggregation logic: Jordi Keynote
The system proves 35M gas Ethereum blocks in ~11 seconds using 32 GPUs, with native-speed execution via precompiled instructions (~1.5 GHz), parallelized witness generation, and sub-2s aggregation. ZisK targets 128-bit security, and is open-source across prover, circuits, and tooling.
Luca Donno then challenged a common assumption in zk discourse: that verifiable computation is equivalent to trustless computation. In reality, he argued, trust assumptions persist - often hidden in infrastructure layers like RPC endpoints, verifier keys, circuit tooling, and audit processes.
His keynote called for a more precise framing: minimizing trust, not eliminating it: Luca Keynote
Stephen Duan then introduced zkMIPS (now Ziren), the first zkVM built on the MIPS32 instruction set - selected for its 40-year legacy, stable semantics, and native support for special-purpose registers and opcodes like MADD/MSUB. The result: reduced circuit complexity, smaller compiled programs, and more predictable performance profiles.
zkMIPS (Ziren) features a modular proving architecture that decomposes programs into parallelizable chips. Using Plonky3 over KoalaBear Prime Field, it enables efficient table commitments and enforces cross-table consistency at scale. Benchmarks show zkMIPS sustaining 2.6 million cycles per second on a single 4090 GPU.
One key deployment is with @GOATRollup, a Bitcoin-native zk rollup. zkMIPS powers recursive proofs of 100M+ Bitcoin blocks and emits compact Groth16 proofs for scalable, cross-chain BTC settlement - demonstrating its readiness for production-grade, yield-bearing Bitcoin infrastructure: Stephen Keynote
The returning Alexander Hicks of the Ethereum Foundation, alongside speakers from @NethermindEth and @Arbitrum discussed the inevitability of zkVM governance, the role of multiprover systems in fault tolerance, and the requirement for formal verification at scale.
The panel also surfaced coordination risks posed by open zkVM ecosystems and real-time proving’s hardware demands - highlighting the need for adaptive processes as ZK infrastructure becomes core to Ethereum’s consensus path: Panel 4
Following the panel, @portport255 introduced ZKsync Airbender and ZKsync OS - a full-stack architecture designed to unify the sequencer and prover under a single modular execution layer. ZKsync OS enables consistent VM behavior across sequencing and proving environments, allowing developers to write code once and execute it identically across multiple hardware and proof targets: Porter Keynote
ZKsync Airbender is a STARK-based, open-source zkVM achieving sub-35s proofs for full Ethereum blocks and sub-1s proofs for miniblocks - designed for high-throughput, low-latency settlement. With support for multiple reduced instruction sets and fully open GPU code, Airbender targets recursive interoperability across chains while preserving developer flexibility and rapid upgrade paths.
This panel, featuring speakers from ZKsync, Axiom, Powdr Labs, and Irreducible, surveyed the current state of zkVM development - from ISA modularity to compiler-level optimization - while spotlighting key performance bottlenecks that still limit real-world deployment. Speakers highlighted recent breakthroughs of their respective projects, but stressed the need for reproducible benchmarks that measure more than cost and latency.
Wonderfully moderated by Reka Medvecz (RISC Zero), the session balanced deep technical insight with levity, closing on a light note with panelists humorously citing “Vitalik” and “number go up” as their enduring motivators for building in ZK: Panel 5
In a live interview, @drakefjustin laid out why Ethproofs exists: to move zkVM benchmarking from toy examples to proving real Ethereum blocks. He emphasized that reproducible, production-grade benchmarks are key to evaluating zkVMs beyond speed - making security and verifiability first-class properties.
Justin also discussed real-time proving as essential to Ethereum’s future, previewing validator clients that can attest with ZK proofs within a single slot, describing multi-zkVM setups as a way to guard against correlated failures, and noted that delayed execution and optimistic attestation make ZK clients viable today.
He also outlined Ethereum’s long-term scale strategy: horizontal execution across thousands of rollups, with real-time ZK settlement as the foundation. The bottleneck now is power - not cost - with a goal of sustaining real-time proving within a Tesla-equivalent envelope (~11.5 kW).
Justin closed by encouraging zkVM teams to engage with the Ethereum Foundation directly. A new internal ZK org is forming, and real-time, open-source zkVMs may soon be treated as core infrastructure alongside execution and consensus clients: Justin Drake Interview
The final panel brought the conversation full circle: how do zkVMs interoperate, and what kind of economic infrastructure is needed to support decentralized proof markets?
ZKM's @daceconomy, Norbert from ZkCloud,@schmiddominik1 of Miden, and Lita's @carlo_cretino50 proposed models where zkVMs can verify each other’s proofs recursively, allowing applications to compose across systems without bridges.
But market design remains unsolved. Pricing, latency guarantees, and randomized prover selection all emerged as critical to avoiding centralization and enabling runtime zkVM optimization: Panel 6
Following the panel, Dominik outlined a sharp critique of Ethereum’s current structure: a global shared state leads to serialized computation, high coordination overhead, and limited privacy.
Miden proposes actor-based execution - each actor proving state transitions in isolation. Messages replace global calls. State is no longer globally composable by default, but massively parallel and privacy-preserving by design. This execution model reframes what we mean by scalability and introduces new primitives for ZK-native applications: Dominik Keynote
Jeremy Bruestle closed the summit with a candid look at what’s real, what’s missing, and what’s next in ZK. He highlighted the arrival of GPU-based real-time proving, orchestration layers, and efficient small-field hash-based proof systems - while underscoring persistent gaps in hardware performance, cost ceilings, and off-chain compute models.
@BruestleJeremy also introduced @boundless_xyz, a protocol for verifiable compute delivery across chains, and made the case for building zkVMs that can eventually run full Linux systems. His message: despite rapid progress, ZK infrastructure is still early - and the most important work is ahead: Jeremy Keynote
The summit surfaced five notable priorities:
Benchmarks must reflect production.
Diversity enables security and speed.
Proving infrastructure shapes protocol design.
Validator-side proving is becoming practical.
Security must be first-class.
Huge thanks to all the incredible speakers, attendees, and the @HouseofZK team for making our @eth_proofs Summit at @BerBlockWeek a great success.
Special thanks to the @ethereumfndn for their support in organizing, and to @vanishree_rao of @fermah_xyz for her help in programming.
And the biggest thanks goes to our sponsors: @zksync, @LineaBuild, @ProjectZKM, and @thezkcloud. This event wouldn't have been possible without your support.
