18 Nov 2026, Wed

TUTTI: Researchers Achieve Parallel CKKS Bootstrapping in Under 3 Milliseconds on CPU, Rivaling Lightweight Schemes

In a significant milestone for cryptographic research, a new breakthrough in Fully Homomorphic Encryption (FHE) has effectively bridged a long-standing performance gap between approximate homomorphic schemes and lightweight alternatives. Researchers Robin Köstler from the University of Luxembourg and Antonio Guimarães from the IMDEA Software Institute have introduced TUTTI, an advanced CKKS bootstrapping procedure designed to optimize multi-threaded latency across varying message counts. Most notably, the new method achieves faster bootstrapping times for a single message than TFHE, a scheme traditionally favored for its lower latency profiles.

The findings are detailed in a new research paper titled "TUTTI: Parallel CKKS Bootstrapping in under 3 ms on CPU," published on the Cryptology ePrint Archive as Paper 2026/2848. The work addresses one of the most persistent bottlenecks in practical FHE implementations: the computational overhead required to refresh ciphertexts and maintain multiplicative depth without compromising data privacy.

Advancing State-of-the-Art CKKS Bootstrapping

Fully Homomorphic Encryption allows computations to be performed directly on encrypted data without requiring decryption keys. Among various schemes, CKKS (Cheon-Kim-Kim-Song) is widely utilized for privacy-preserving machine learning and data science because it supports arithmetic operations on approximate real numbers and vectors. However, every homomorphic multiplication consumes a portion of the ciphertext’s "multiplicative level," eventually necessitating a costly operation known as bootstrapping to clean accumulated noise and restore the ciphertext’s capacity for further computation.

TUTTI builds directly upon SPRU, a previous CKKS bootstrapping advancement introduced at TCHES 2027. The authors refined and improved the foundational SPRU architecture by developing innovative methods to exploit high levels of hardware parallelism and scale the number of processed messages efficiently. Furthermore, the work incorporates a refined security analysis of the block-binary secret keys utilized in LMSS, originally presented at AsiaCCS 2023. This security analysis holds independent interest for cryptographers studying the hardness assumptions underpinning modern lattice-based cryptographic schemes.

Unprecedented Multi-Threaded Performance and Scaling

According to the paper’s benchmark evaluations, TUTTI delivers exceptional performance metrics across multi-threaded CPU environments. When operating on a high-core-count machine utilizing 192 threads, TUTTI successfully bootstraps 4096 real-number messages in a timeframe ranging from 46 to 83.4 milliseconds. This process restores anywhere from 1 to 9 multiplicative levels, outperforming SHIP—a prominent previous solution from Eurocrypt 2025—by a factor of up to three on identical hardware configurations.

Unlike SHIP, which is optimized primarily for massive batch operations, TUTTI demonstrates immediate efficiency gains when scaling down to smaller workloads. Processing 256 messages requires just 15 to 34 milliseconds on 32 threads. When narrowing the scope down to a single message, TUTTI completes the bootstrapping process in 4.6 milliseconds using 32 threads. For comparison, the established TFHE-rs library requires 7.3 milliseconds to execute a 4-bit bootstrapping procedure on the exact same machine and thread count.

The performance advantages extend beyond traditional high-core server architectures to consumer-grade hardware. Testing on an Apple M5 Pro laptop featuring ARM architecture, TUTTI achieves a single-message bootstrapping latency of just 2.4 milliseconds across 14 threads. In contrast, TFHE-rs requires 4.5 milliseconds on the same Apple hardware.

Closing the Gap in Practical FHE Literature

The introduction of TUTTI marks a turning point in the ongoing dialogue surrounding the practical deployment of homomorphic encryption. While structural differences between CKKS and lightweight schemes like TFHE continue to persist across several dimensions—including application functionality, ciphertext expansion, key sizes, and underlying security assumptions—the research demonstrates that multi-threaded execution latency is no longer a distinct differentiator.

By bringing CKKS bootstrapping speeds down to under three milliseconds on consumer and enterprise CPUs, the authors have effectively closed a long-standing performance gap that has constrained practical FHE literature for years. The complete research paper, along with formal academic citations and BibTeX references for integration into cryptographic libraries and scholarly works, remains publicly accessible via the Cryptology ePrint Archive under identifier 2026/2848.

Leave a Reply

Your email address will not be published. Required fields are marked *