Abstract

This tutorial provides a systematic and practical introduction to building high-performance privacy-preserving applications using GPU-accelerated fully homomorphic encryption (FHE). We focus on bridging the gap between cryptographic theory and real-world system implementation by examining how modern FHE schemes are mapped onto contemporary GPU architectures.

Fully homomorphic encryption enables computation directly on encrypted data, providing strong end-to-end data security without requiring decryption during computation. Recent advances in hardware accelerators, especially GPUs, have significantly improved the practicality of FHE.

Cheddar is the fastest open-source GPU library for accelerating the CKKS homomorphic encryption scheme. It provides high-level primitives such as linear transformations and enables developers to build secure computation pipelines efficiently.

Organizers

Requirement

Source code is available at Here

To run Cheddar, you will need the following hardware and software:

• NVIDIA server/consumer GPU (Pascal or later) with at least 16GB of DRAM
• CUDA 11.8 or later
• Docker and nvidia-container-toolkit (install guide)

About Cheddar

Cheddar is a high-performance GPU library for accelerating the CKKS fully homomorphic encryption scheme. It provides optimized primitives for encrypted computation and enables practical privacy-preserving applications on modern GPU architectures.

For more details about the system design and implementation, please refer to the original paper:Cheddar: A Swift Fully Homomorphic Encryption Library Designed for GPU Architectures.

Citation

If you use Cheddar or refer to its performance analysis, Theodosian, in your research, please consider citing the following papers:

Cheddar

@inproceedings{asplos-2026-cheddar,
  author = {Choi, Wonseok and Kim, Jongmin and Ahn, {Jung Ho}},
  title = {Cheddar: {A} Swift Fully Homomorphic Encryption Library Designed for {GPU} Architectures},
  booktitle = {Proceedings of the ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)},
  year = {2026},
  url = {https://doi.org/10.1145/3760250.3762223},
  doi = {10.1145/3760250.3762223}
}

Theodosian

@inproceedings{ispass-2026-theodosian,
  title = {Theodosian: {A} Deep Dive into Memory-Hierarchy-Centric {FHE} Acceleration},
  author = {Choi, Wonseok and Yu, Hyunah and Kim, Jongmin and Ji, Hyesung and Park, Jaiyoung and Ahn, {Jung Ho}},
  year = {2026},
  booktitle = {IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)}
}

Schedule

This half-day tutorial will be held from 8:00 AM to 11:50 AM on Sunday, June 28, 2026, in Room 301A.