Closed thomaslavaur closed 1 week ago
Please run with maxopt like this:
RUN bazel build --config maxopt //:prover_main && \
cp ./bazel-bin/prover_main /usr/local/sbin/rapidsnark
It's better but Tachyon is still slower. I changed my computer so here are the new performances:
Succint proof (without wrapping): 20s Complete proof (with rapidsnark): 50s Complete proof (with Tachyon): 104s
Can it be because the circuit is too large?
@thomaslavaur Thanks for reporting the issue. We'll take a look at it and get back to you
bazel build --@kroma_network_tachyon//:has_openmp --config maxopt //:prover_main
@thomaslavaur Sorry you didn't enable multithreading as well. Please try like this.
In vendors/circom
there is a README with guides on how to build and info on docker as well. I think you can refer to it when building your docker.
Thanks for your help. Now times are close to each other: Succint proof (without wrapping): 20s Complete proof (with rapidsnark): 52s Complete proof (with Tachyon): 50s
Closing the issue! Btw, we also have docker for CUDA build :)
Issue type
Performance
OS platform and distribution
Ubuntu 22
Current behavior?
Machine: 13th Gen Intel(R) Core(TM) i9-13980HX I benchmark tachyon circom vendor and rapidsnark, and compare their performance over the RISC0 zkVM wrapper.
I obtained that: Succint proof (without wrapping): 74s Complete proof (with rapidsnark): 156s Complete proof (with Tachyon): 400s
Expected Behavior?
I want to figure out the reason.
Standalone code or description to reproduce the issue
I used their own script for rapidsnark and modified it to build the groth16 wrapper of RISC0 in docker:
Additional context
No response