[Question] Performance comparison with GPU approaches

[ajx42]

Hey all, I was just going through the repository and associated blog: https://www.ingonyama.com/blog/hardware-review-gpus-fpgas-and-zero-knowledge-proofs.

The blog compares your work @ 128 MHz with Sppark's MSM. I wonder if you have also closed timing with a higher frequency (350+ MHz) and have results just to see what kind of peak performance is possible and if that gets close to the state-of-the-art GPU performance.

Thanks a lot in advance!

[omershlo]

good question. The blog you mentioned is referring to results from our PipeMSM paper. We didn't publish a follow-up paper but since the time of publication, we've managed to improve our algorithms and engineering (increased frequency) and collect many more data points for comparison. Some other teams announced results as well. In my latest blog post, I am trying to converge to a canonical way of comparing between zk hw. We will soon write more about that topic and release our database with all the different benchmarks

[omershlo]

https://medium.com/@omershlomovits/revisiting-paradigm-hardware-acceleration-for-zero-knowledge-proofs-16f717a49555

[ajx42]

Thanks a lot for sharing, I will check this out! Best wishes