IaaS vendors like AWS sell cloud instances based on vCPUs. In principle, vCPUs within the same type of cloud instance should have similar single-core performance (excluding the issue of noisy neighbors). However, for the vendor, the CPUs in their data centers are bound to be diverse, with each CPU model having different single-core performance capabilities.
So, how do they establish the correspondence between vCPUs on the cloud instance and the logical cores on the physical machine?
For instance, if I purchase a cloud instance with 16 vCPUs, it might correspond to 14 CPU quotas on a newer machine with higher single-core performance, such as Icelake or SPR. On the other hand, it might require 18 CPU quotas on an older machine with lower single-core performance, like Cascade Lake. How is this ratio, such as 16-14 or 16-18, determined? Please note that the specific numbers mentioned here are just examples for illustration purposes.