In TFHE.rs, all the ciphertexts of integers are composed of many boolean ciphertexts?

[GuLu_GuLu]

I read the answer of this question:

Why does FheUint16 addition take more than 20 ms with GPU?

Hi, I wonder why TFHE addition (FheUint16) takes as much as 25 ms when using GPU: https://docs.zama.ai/tfhe-rs/guides/run_on_gpu#benchmarks [image] Basicaly, a TFHE ciphertext is (vector(a), b). When we add two TFHE ciphertexts as follows: C1 = (vector(a1), b1) C2 = (vector(a2), b2) We only need to compute C1 + C2 = (vector(a1) + vector(a2), b1 + b2) Provided the vector is around 1000 dimension (for 128 bit security?), this computation can be done almost instantly. But the benchmark page…

So, take the 8-bit integer as an example, is it implemented by using 8 boolean ciphertexts?

[Agnes]

Hello! With the default TFHE-rs parameters, an 8-bit integer is represented using 4 LWE ciphertexts, each encrypting 2 bits of message and 2 bits of carry. There are parameters in TFHE-rs to represent an 8-bit integer with 8 boolean ciphertexts but it’s not as efficient.