mhmughees
mhmughees
I added more detail to my above question for clarity
okay thanks for explanation, so does around 100x time more time for ciphertext-ciphertext mul sounds right in ball park to you?
very interesting but I didnt get your parameters, I actually updated my first comment to add details of coefficient modulus, I mention it's value without last special parameter. Which updated...
Yeah I think processor difference may be playing a role here as well. Also as you are developer of Lattigo , can you confirm if Lattigo use 32 word size...
Okay I just reran the experiment using aws c5.2xlarge with avx512 enabled Coefficient modulli = {58,60,40} without special mod. N=8192 and Now CT*CT ~ 9 ms which is still higher...