Can we maybe agree on what we expect from 1 core? what’s the baseline?
a core from 5 years ago and a core in 2024 are very different, you can also have a e-core or a p-core (optimise for power or optimise for performance).
There is already consensus on that here Run a Validator (Polkadot) · Polkadot Wiki, all the weights for the runtimes are generated using that.
Another would be to have a benchmark that people can run on their validator and it would tell large enough or not?
Yes, we do have that already, benchmarks are run at polkadot startup and tell all the validators how their single core performance compares with the reference numbers from here: polkadot-sdk/substrate/utils/frame/benchmarking-cli/src/machine/reference_hardware.json at a3eda0a1bdc67fc14efbf64c1f0046d452f55e31 · paritytech/polkadot-sdk · GitHub.
Additionally, part of this I also added a benchmark that would tell validators if their parallel performance does not met the new requirements of 8 cores.