Reproduce “Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wonseok Choi, Jongmin Kim, Jung Ho Ahn
Seoul National University researchers developed Cheddar, a Swift Fully Homomorphic Encryption library that accelerates CKKS scheme operations on GPU architectures. It achieves up to 19.6x speedups for FHE workloads over existing GPU implementations and reduces encrypted ResNet inference latency to sub-second levels on modern NVIDIA GPUs.
Key insights
- Efficiently leveraging the native 32-bit integer datapath of GPUs requires a systematic 32-bit RNS design, which can avoid the quadratic overhead of emulating 64-bit operations.
- Micro-architectural optimizations for core FHE kernels, such as signed Montgomery reduction and on-the-fly twiddle factor generation, are critical for maximizing computational efficiency.
- Extensive kernel fusion, both sequential and parallel, significantly reduces memory bandwidth bottlenecks and data movement, contributing substantially to overall FHE workload acceleration.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.