The metric we are really interested in is proving time. So far the number of qemu steps has been measured. That was easy to set up and was independent of the zkVM vendor tooling. Now lets add proving time. That requires making some further decisions: choosing zkVM (or zkVMs), hardware for benchmarking, plugging it into CI and making sure the benchmarking environment is stable in a sense for being able to reproduce past benchmark results.
If precompiles are to be accelerated by a zkVM then vkVM's emulator has to be used instead of qemu.
The metric we are really interested in is proving time. So far the number of qemu steps has been measured. That was easy to set up and was independent of the zkVM vendor tooling. Now lets add proving time. That requires making some further decisions: choosing zkVM (or zkVMs), hardware for benchmarking, plugging it into CI and making sure the benchmarking environment is stable in a sense for being able to reproduce past benchmark results.
If precompiles are to be accelerated by a zkVM then vkVM's emulator has to be used instead of qemu.