Typically 2 per core.
https://en.wikipedia.org/wiki/Nice_(Unix)
The exact mathematical effect of setting a particular niceness value for a process depends on the details of how the scheduler is designed on that implementation of Unix. A particular operating system's scheduler will also have various heuristics built into it (e.g. to favor processes that are mostly I/O-bound over processes that are CPU-bound). As a simple example, when two otherwise identical CPU-bound processes are running simultaneously on a single-CPU Linux system, each one's share of the CPU time will be proportional to 20 − p, where p is the process' priority. Thus a process, run with nice +15, will receive 25% of the CPU time allocated to a normal-priority process: (20 − 15)/(20 − 0) = 0.25.[2] On the BSD 4.x scheduler, on the other hand, the ratio in the same example is about ten to one.[citation needed]