Yep. Its easier to think of Vermeer as behaving more like a GPU than a CPU. Power, current and thermal limits are set at bios/firmware level and it will opportunistically boost as high as possible, as often as possible within those limits.
The main limiters are PPT (socket power draw), TDC (peak current), EDC (average current) and temperature. Stock PPT, TDC and EDC for 5900x is 142, 95, 140. Stock temperature limit I think is 90C.
If you do not want your cpu you to exceed a set temperature threshold, you enable PBO2 and set the thermal throttling limit to say 75C. Now your cpu will still opportunistically boost but a limiter will kick in so it never goes above 75C.
If the cpu is miles below its power, current and temperature limits, it treats that as boost headroom. So this is why in very light workloads (or just idling on the desktop with background processes running), it will intermittently shoot a single core up with 1.475V and boost to 4.9ghz+.
Thats why idle and light workload temperatures fluctuate a lot. Mine are between 40 and 65C on a Noctua U12A. If you are upgrading from an Intel xlake cpu, this behaviour might seem strange, but it seems to be what Vermeer chips are designed to do. Don't worry about it.
In games, my temps and vcore are both higher than if I run synthetic all core stress tests. This again seems strange coming from Intel xlake but there are reasons for that.
This behaviour is a result of the limits I set for PBO2 in bios. It can boost all core to about 4.2 ghz at stock PPT, TDC and EDC while staying below 79C (my manual thermal throttling limit) in Cinebench R23. It keeps vcore around 1.1 to 1.3V and never goes above 79C. If you raise PPT, TDC and EDC, it will create more headroom for the cpu to boost to higher clocks and hold them for longer until you hit 79C. Then you run into your temp limiter so if you still want to crank clocks, that has to increase too.
You can effectively remove all the user defined limitations by setting them to absurdly high numbers or letting your motherboard automate PPT, TDC and EDC. It still has all of its hardware level protection mechanisms. It will still thermal shutdown if you go to like 105C.
As long as you have adequate power delivery and cooling, you can relax your user defined limits and let it just do its thing. In exchange for higher temps and/or higher power draw, it will hold higher boost clocks for longer, both single and all core.
There is a point above 1.38V where you get poor voltage/frequency scaling though. If you really don't like your power consumption and temperature numbers while gaming, you just set a limit you don't want it to cross in bios. Say, PPT = 165, TDC = 120 and EDC = 150 and thermal throttle limit = 75C. These are just aribtrary numbers for illustrative purposes.
Will it hold its highest clock frequency for as long and as often as if those limit were relaxed or removed? No, but this happens at the tail end of the voltage/frequency curve where the scaling goes in the toilet. On non exotic cooling it amounts to smaller increases in clock frequency for disproportionately more power consumption and heat. Physics will win in the end.
