Vitalik calls on ZK and FHE developers to "show the encryption ratio directly": you can see the difference at a glance and then talk about optimization

Privacy technology needs to be clear at a glance. Vitalik Buterin called on developers to directly show the "efficiency ratio" when talking about the performance of ZK and FHE.
(Preliminary summary: The Ethereum Foundation established a "Privacy Research Group" to promote six major roadmaps and fully launch competition on the privacy track)
(Background supplement: The Ethereum Foundation released an end-to-end privacy blueprint, a three-pronged approach to strengthen the foundation of DeFi and compliance)

Contents of this article

Ethereum co-founder Vitalik Buterin recently published an article on the X platform, recommending developers to evaluate zero-knowledge proofs (ZK) and fully homomorphic encryption (FHE), we should abandon the usual indicator of "N operations per second" and instead focus on the efficiency ratio of "encryption calculation time/original calculation time". The intention is to propose a more direct test standard for the feasibility of Web3 privacy technology.

Vitalik Focus on Efficiency Ratio

Traditional throughput metrics are extremely dependent on the hardware environment and cannot reveal the true burden caused by the encryption layer. Vitalik points out that if developers know that the original calculation only takes 1 millisecond, they can directly deduce from the efficiency ratio how long encryption will amplify.

Vitalik's tweet translation:

I hope more people doing ZK (zero-knowledge) and FHE (fully homomorphic encryption) can use ratio values to express the additional overhead ("computation time under cryptographic protection" vs. "original computing time"), instead of just saying "we can do N operations per second."

This is more independent of hardware and can provide a very informative number: If my application is protected by cryptography instead of relying on trust, how much efficiency will I sacrifice?

This is also generally better for estimating, because as a developer Ialready know how much time the raw calculation takes, and I just take that time and multiply it by the multiplier.

(Yes, I know this is difficult, because the operations required between "execution" and "generating a proof" are of a different nature, especially involving SIMD/parallelization and memory access patterns, so even the ratio is still affected by the hardware to some extent. But even so, I still think that expressing the overhead as a multiple, although not perfect, is still a good indicator.)

I wish more ZK and FHE people would give their overhead as a ratio (time to compute in-cryptography vs time to compute raw), rather than just saying "we can do N ops per second"

It's more hardware-independent, and it gives a very informative number: how much efficiency am I…

— vitalik.eth (@VitalikButerin) October 18, 2025

Vitalik emphasized that although this ratio will still be affected by memory layout, degree of parallelization and instruction set differences, it at least allows the community to "use the same ruler" to measure different solutions.

Performance bottlenecks of ZK and FHE

ZK and FHE have very different functions in protecting user privacy, but they also face high overhead. Once the ZK circuit complexity increases, the proof generation time may take hundreds of times. The bottleneck of FHE is even more obvious. The FHE version of machine learning inference is 20,000 times slower than plain text.

These delays make it difficult to implement scenarios such as DeFi, decentralized identity (DID), and AI on-chain, and also highlight the importance of the efficiency ratio framework. Therefore, Vitalik calls on everyone to see the burden of each solution before we can talk about optimization.

Optimization path and ecological cooperation

Vitalik’s initiative encourages the community to reallocate R&D resources. It can be seen that in the short term, innovation at the algorithm level is still the primary means to reduce the ratio. The next mid-term is the upgrade of GPU or ASIC computing equipment, which is expected to reduce the absolute time calculation time to a range acceptable to users.

In the long term, selective encryption and cross-layer collaboration will be key to driving mass adoption. Currently, the encryption industry is promoting ZK circuit standardization, FHE compiler optimization and off-chain proof sharing. The goal is to reduce the efficiency ratio without weakening privacy and win more scenarios for decentralized applications.