NVIDIA cuPQC
NVIDIA cuPQC is an SDK of optimized libraries for accelerating post-quantum cryptography (PQC) workflows. Ensuring the security and authenticity of sensitive data is of critical importance. Quantum computers threaten to break today’s public-key encryption mechanisms, and migration to algorithms considered safe against a quantum computer is needed to protect the world’s sensitive data. NVIDIA cuPQC accelerates the leading PQC algorithms, advancing data security against quantum computer threats.
Join our early interest list to stay informed about the latest updates
Worldwide Effort to Migrate to PQC
Though a quantum computer powerful enough to break today’s encryption of sensitive data is many years away, harvest-now decrypt-later attacks mean data that needs to be secured for many years could be vulnerable now. To meet that challenge, efforts are underway worldwide to develop, standardize, and implement new PQC algorithms. NVIDIA works with the entire global cryptographic community to enable this transition, and cuPQC can be integrated into any encryption software to ensure maximum performance.
Accelerating Leading Algorithms
cuPQC achieves breakthrough performance on the NIST finalist algorithms, accelerating Kyber768 by up to 500x.
PQC Alliance
NVIDIA is a founding member of the PQC Alliance, a Linux Foundation consortium which aims to advance the adoption of PQC by producing high-assurance software implementations of standardized algorithms.
Adopted by the Global Cryptography Ecosystem
cuPQC is being adopted by leading cryptography developers, from cloud service providers to leading startups.
Performance
Accelerating Leading PQC Algorithms
ML-KEM, or Kyber, is the finalist for key exchange expected to be standardized by NIST in 2024. cuPQC achieves throughputs of up to 27 million keygen/s, 22 million encapsulations/s, and 20 million decapsulations/s for batched Kyber768 on a single H100 GPU, increases of 515x, 486x, and 340x, respectively over a state-of-the-art CPU.
Performed on an NVIDIA H100, compared to state-of-the-art, single-threaded CPU results from the liboqs benchmark suite. Measured in terms of throughput and operations per second, using a batch size of 1,000,000.
Partners Adopting NVIDIA cuPQC
