A self-hosted CPU benchmark would be more stable and would allow better monitoring of performance over time.
Configurations of interest:
Linux gcc/clang
MacOS clang
Windows MSVC
Various thread counts (though single thread performance is fine for now)
It would probably be fine to start with a linux performance test only and monitor the other performance numbers in line with that until more resources are available.
A self-hosted CPU benchmark would be more stable and would allow better monitoring of performance over time.
Configurations of interest:
It would probably be fine to start with a linux performance test only and monitor the other performance numbers in line with that until more resources are available.