A mining central processing unit (CPU) is a small heater that happens to hash.

Watts are how fast electricity goes in. Almost every watt comes back out as heat. The fan does not destroy it. The fan moves it into the room. If the room is warm, the chip starts warmer too.

Temperature is not the electric bill. Temperature is the traffic jam between the chip and the room.

A big cooler and a small cooler can dump the same watts.
The small one just runs hotter, because it cannot move the heat as fast as the chip makes it.
If it falls behind, the CPU slows itself down. You get a high temperature and a disappointing hashrate. That is throttle.
Mining is not a game spike. It sits there for hours. Look at the temperature after a few minutes, not after ten seconds.

Three numbers that are not the same:

Package watts are the chip. That is the heater.
Wall watts are the plug. The rest of the PC, plus waste in the power supply, is in that number. It is always higher.
The volt-ampere (VA) sticker on an uninterruptible power supply (UPS) is not watts.
More threads are not free speed.

Each new thread asks for more watts, which means more heat.
At first, hashrate climbs with them.
Then heat climbs faster than the hashrate. That is the stop.
Sometimes the wall is memory, not heat. A cool chip with a big cooler can want more threads. A small cooler usually does not.
The temperature to watch on a current desktop Ryzen is Tctl (temperature control).

The chip uses that number for the fans and for deciding when to slow down. Stay with Tctl. Do not mix it with a different sensor in the same comparison.

Tccd, the core-complex-die temperature, is a real die sensor. It often reads a little differently. It is useful. It is not the line.
The basic input/output system (BIOS), Ryzen Master, and a motherboard tool can show a different degrees number, or the same heat under a different name. If the label is not Tctl, do not treat it as Tctl.
On Linux, the sensors reading labeled Tctl is the one. On Windows, Libre Hardware Monitor has that same label. A random "CPU" box in the BIOS is not automatically that sensor.
Some first-generation Ryzen chips added an offset, so Tctl sat hotter than the die on purpose. That is not how current desktop Ryzen works. On these, Tctl is the control reading.
Where you change the power matters. These are not the same door.

Windows has a power plan. On one chip it barely moves the watts. On another, Power saver parks the miner and the hashrate falls with the temperature. A cooler number is not a win if the work collapsed.
Linux has a governor. Powersave can do the same trick: the chip looks cool because it slowed down. Performance or ondemand is the mining governor when the cooler can take it. A thread cut is often the cleaner cool dial.
The BIOS is the setting that survives a reboot and an operating-system reinstall. Power limits and fan curves often live there. The name is motherboard specific. One board says PPT, package power tracking, which is AMD's cap on how many watts the chip may pull. Another says Package Power, Eco Mode, or Precision Boost. Same idea, different menu.
Ryzen Master, and the motherboard's own app, can change those same limits from Windows. They can disagree with the BIOS. A slider in Ryzen Master may not stick after reboot unless that tool is set to apply it. If the BIOS and the app both think they own the fan curve, the BIOS usually wins.
A pool dashboard often has a power field made for a graphics processing unit (GPU). On a CPU miner it can read zero. That is not a free miner. It is the wrong sensor.

The check is simple. Add a couple of threads. Wait a few minutes. Read Tctl, the same tool you started with. If the hashrate moved and Tctl barely did, you had room. If Tctl jumped and the hashrate barely did, those threads were mostly a heater.

Pick a Tctl you are willing to sit at all night, for that cooler, that room, and that board. Someone else's number is not your number.