What's the best machine choice for a very large self-managed Postgres cluster on GCP – c3, c3d, n2 ?

The purpose of this issue is to consolidate all the info we have so far to compare the biggest instances on GCP for the cases when we need to operate the largest possible instances for self-managed Postgres:

• c3 (4th-gen Intel "Sapphire Rapids") with max 176 vCPUs and 1408 GiB RAM
• c3d (4th-gen AMD EPYC) with max 360 vCPUs and 2880 GiB RAM
• n2 (Intel Cascade Lake or Ice Lake) with max 128 vCPUs and 864 GiB RAM

Comparison should include:

• price vs compute power
• max possible numbers observed for basic benchmarks
• known issues (e.g. lwlock; pgss overhead)
• two lines of considerations:
  • the room for growth "no matter what" (not taking the price into account)
  • compute/price ratio compared to n2-highmem-128 as a baseline (in other words: if we want to replace the biggest n2, is it reasonable in terms of power/money?)

@fomin.list – please consolidate the knowledge in this issue and describe the planned steps