Awesome Client-Side GPU

GPU-accelerated cryptography and zero-knowledge proof projects that run on user devices — browsers, phones, and desktops — not remote servers.

Contents

• What Is Client-Side GPU Proving
• GPU API Quick-Reference
• Projects by GPU Technology
  • WebGPU (Browser-Native)
  • Metal (Apple Silicon / iOS / macOS)
  • Vulkan / ROCm / HIP (Cross-Platform Native)
  • Cross-Platform Frameworks
• Projects by Cryptographic Operation
  • Multi-Scalar Multiplication (MSM)
  • Number Theoretic Transform (NTT / FFT)
  • Field Arithmetic
  • Hash Functions
  • Full Proof Systems
• Mobile and Edge Proving
• Competitions and Benchmarks
  • ZPrize Entries
  • Benchmarks and Metrics
• Learning Resources
  • Blog Posts and Articles
  • Research Papers

---

What Is Client-Side GPU Proving

Zero-knowledge proofs require heavy finite-field and elliptic-curve arithmetic that maps naturally onto GPU parallelism. Client-side GPU proving brings these speedups to ordinary user hardware — a browser tab, a smartphone, a laptop — so privacy-preserving applications never need to send sensitive inputs to a remote server.

Six classes of operation benefit most from GPU acceleration:

• MSM (Multi-Scalar Multiplication) — dominant cost in pairing-based systems (Groth16, PLONK).
• NTT (Number Theoretic Transform) — core of STARK and FRI-based systems.
• Polynomial evaluation and matrix-vector multiplication — high-throughput linear algebra.
• Merkle tree commitments — batch hashing for hash-based proof schemes.
• Sumcheck protocol — multivariate polynomial reductions (GKR, WHIR).
• Linear encoding and hashing — used in Basefold and Brakedown commitment schemes.

MSM and NTT yield 10–100× speedups over CPU on modern GPU hardware and are the primary focus of most projects in this list.

GPU API Quick-Reference

API	Platforms	Primary Languages	Browser Support	Notes
WebGPU	Browser, native (via wgpu)	WGSL, Rust, JS/TS	Chrome 113+, Firefox (flag), Safari TP	W3C standard; replaces WebGL for compute
Metal	iOS, macOS, Apple Silicon	MSL, Rust (metal-rs), Swift	No	2–3× faster than OpenGL on Apple hardware
Vulkan Compute	Windows, Linux, Android	GLSL, SPIR-V, Rust	No	Cross-vendor; lowest-level portable API
ROCm / HIP	AMD GPUs (Linux)	HIP C++	No	CUDA-compatible syntax; growing mobile support
CUDA	NVIDIA GPUs	CUDA C++	No	Highest throughput; server/desktop only
OpenCL	Most GPUs	OpenCL C	No	Legacy; being superseded by Vulkan/WebGPU

Projects by GPU Technology

WebGPU (Browser-Native)

• ICME-Lab/msm-webgpu - MSM over BN254 using the cuZK algorithm, compiled to WGSL shaders. GPU: WebGPU Curve: BN254 Op: MSM Lang: Rust.

• td-kwj-zp2023/webgpu-msm-bls12-377 - Winning ZPrize 2023 entry for MSM over BLS12-377. GPU: WebGPU Curve: BLS12-377 Op: MSM Lang: TypeScript.

• td-kwj-zp2023/webgpu-msm-twisted-edwards - ZPrize 2023 MSM entry targeting Twisted Edwards curves with extended-coordinates arithmetic in WGSL. GPU: WebGPU Curve: BLS12-377 Op: MSM Lang: TypeScript.

• demox-labs/webgpu-crypto - Research-stage library covering MSM, NTT, and hash functions over BLS12-377 and BN254. GPU: WebGPU Curve: BLS12-377, BN254 Op: MSM, NTT, Hash Lang: TypeScript.

• demox-labs/webgpu-msm - Baseline ZPrize 2023 reference for browser-native MSM over BLS12-377. GPU: WebGPU Curve: BLS12-377 Op: MSM Lang: TypeScript.

• penumbra-zone/webgpu - Pioneer Groth16 prover for Penumbra's shielded transactions over BLS12-377. GPU: WebGPU Curve: BLS12-377 Op: Groth16 Lang: Rust Status: Archived.

• ligeroinc/ligero-prover - Hash-based ZK prover using FFT-based polynomial IOPs, running in browsers via Dawn/Emscripten. GPU: WebGPU Op: NTT, Hash Lang: C++.

Metal (Apple Silicon / iOS / macOS)

• zkmopro/gpu-acceleration - Metal MSM library for Apple Silicon achieving 40–100× speedup over CPU, integrated into the mopro toolkit. GPU: Metal Curve: BN254 Op: MSM Lang: Rust.

• geometryxyz/msl-secp256k1 - Metal Shading Language implementation of secp256k1 elliptic-curve arithmetic using Jacobian coordinates, targeting Apple Silicon GPUs. GPU: Metal Curve: secp256k1 Op: EC Arithmetic Lang: MSL.

Vulkan / ROCm / HIP (Cross-Platform Native)

The Cross-Platform Frameworks listed below all support Vulkan or ROCm as one of their backends.

Cross-Platform Frameworks

• ingonyama-zk/icicle - Production-ready ZK acceleration library with CUDA, Metal, and ROCm backends and iOS/Android support. GPU: Multi Op: MSM, NTT Lang: C++, Rust.

• tracel-ai/cubecl - GPU compute framework for Rust using a proc-macro kernel language that compiles to WebGPU, CUDA, ROCm, and Metal. GPU: Multi Op: Field Arithmetic Lang: Rust.

• lambdaclass/lambdaworks - Full ZK cryptography library with GPU-accelerated FFT, MSM, and Merkle tree backends for CUDA and Metal. GPU: Multi Op: FFT, MSM Lang: Rust.

• mratsim/constantine - High-performance elliptic-curve and pairing library in Nim with experimental WebGPU and CUDA backends. GPU: Multi Op: EC Arithmetic, MSM Lang: Nim.

• spaceandtimefdn/blitzar - MSM and Pedersen commitment library with CUDA and CPU backends. GPU: CUDA Op: MSM Lang: C++, Rust.

Projects by Cryptographic Operation

Multi-Scalar Multiplication (MSM)

Primary entries are in the GPU Technology sections above.

Project	GPU	Curve
ICME-Lab/msm-webgpu	WebGPU	BN254
td-kwj-zp2023/webgpu-msm-bls12-377	WebGPU	BLS12-377
td-kwj-zp2023/webgpu-msm-twisted-edwards	WebGPU	BLS12-377 (Twisted Edwards)
demox-labs/webgpu-msm	WebGPU	BLS12-377
zkmopro/gpu-acceleration	Metal	BN254
ingonyama-zk/icicle	Multi	Multi-curve
lambdaclass/lambdaworks	Multi	Multi-curve
spaceandtimefdn/blitzar	CUDA	Multi-curve

Number Theoretic Transform (NTT / FFT)

Primary entries are in the GPU Technology sections above.

Project	GPU	Notes
demox-labs/webgpu-crypto	WebGPU	Research-stage
ligeroinc/ligero-prover	WebGPU	FFT-based polynomial IOPs
ingonyama-zk/icicle	Multi	Production-ready
lambdaclass/lambdaworks	Multi	Full ZK library

Field Arithmetic

Project	GPU	Notes
tracel-ai/cubecl	Multi	Kernel language targets any backend
mratsim/constantine	Multi	Nim runtime JIT

Hash Functions

Project	GPU	Notes
demox-labs/webgpu-crypto	WebGPU	Poseidon, Keccak research
ligeroinc/ligero-prover	WebGPU	Hash-based proof system

Full Proof Systems

Project	GPU	System	Notes
penumbra-zone/webgpu	WebGPU	Groth16	Archived pioneer
ligeroinc/ligero-prover	WebGPU	Ligero	Browser via Dawn/Emscripten
zkonduit/ezkl	Metal	PLONK / ZKML	iOS/macOS
zkmopro/mopro	Metal	Multi-prover	iOS mobile toolkit

Mobile and Edge Proving

• zkmopro/mopro - Multi-prover mobile toolkit exposing a unified Swift/Kotlin API over Halo2, Noir, and Circom backends. GPU: Metal Op: Groth16, PLONK Lang: Rust.

• zkonduit/ezkl - ZK machine-learning inference library on iOS and macOS, delivering ~2× speedup over the CPU path. GPU: Metal Op: ZKML, PLONK Lang: Rust.

Competitions and Benchmarks

ZPrize Entries

ZPrize is an open competition that accelerates ZK hardware and software. The 2023 WebGPU track (Prize #2) produced the fastest known browser-native MSM implementations;

all entries are listed in the WebGPU section above.

Benchmarks and Metrics

• moven0831/field-ops-benchmarks - Benchmarks for M31 and BN254 field operations across Metal and WebGPU backends. GPU: Metal, WebGPU Op: Field Arithmetic.

• ethproofs.org — client-side proving dashboard - Live leaderboard tracking proof generation times and costs for client-side Ethereum proof systems across various hardware targets.

Learning Resources

Blog Posts and Articles

• Mopro: "Metal MSM v2" - Technical walkthrough of the second-generation Metal MSM implementation in zkmopro/gpu-acceleration, explaining the 40–100× CPU speedup.

• zkSecurity: "Accelerating ZK Proving with WebGPU" - Explores how WebGPU NTT shaders accelerate Stwo STARK proving in the browser.

• Geometry Research: "Accelerating Client-Side ZK with WebGPU" - Introduces the MSL secp256k1 work and frames the broader opportunity for WebGPU in client-side ZK proving.

• PSE: "Client-Side GPU Acceleration for ZK" - Frames why client-side GPU proving matters for Ethereum privacy and surveys six key GPU-accelerable ZK primitives.

Research Papers

• cuZK: Accelerating Zero-Knowledge Proof with a Faster Parallel Multi-Scalar Multiplication Algorithm on GPUs - IACR 2022/1321; introduces the cuZK MSM algorithm that underpins most WebGPU and Metal MSM implementations in this list. Op: MSM.

Related Lists

WebGPU and GPU Compute

• mikbry/awesome-webgpu - Comprehensive WebGPU ecosystem list covering specs, demos, shader libraries, and tutorials.

• rofrol/awesome-wgpu - Curated resources for wgpu, the primary Rust WebGPU implementation used by many projects in this list.

• jslee02/awesome-gpgpu - General-purpose GPU compute frameworks including CUDA, OpenCL, and Vulkan Compute.

• goabiaryan/awesome-gpu-engineering - GPU architecture, optimization techniques, and Metal Performance Shaders resources.

Zero-Knowledge Proofs

• matter-labs/awesome-zero-knowledge-proofs - Comprehensive ZKP educational resources covering theory, tools, and applications.

• ventali/awesome-zk - ZK resources spanning zkVMs, L1/L2 protocols, developer tools, and benchmarks.

• odradev/awesome-zero-knowledge - Blockchain-focused ZK resources with a companion weekly newsletter at zknewsletter.com.

---

Contributing

See CONTRIBUTING.md for the scope rules, tag vocabulary, and submission checklist.