It looks like the finest I get to select whether through the API or through the Google Compute's OWA is a whole family/architecture (Broadwell, Skylake, etc.) However, performance and instruction set support varies greatly within those architectures. For example, one look at the Skylake family tells you that CPU speed varies from 2.2 to over 3 GHz and there is also a wide variation in the instruction sets and extensions supported by the members of the family:
In light of that, I have the following question: does Google provide any guarantees that the two VM's selected to be within the same family would perform comparably? Is there any way to select the spec for your cores with more granularity?