Putting aside having to know the quirks behind one's CPU, we have:
AMD
- AMD Overdrive shouldn't be used on >K10 CPUs.
- BCLK overclock is known to screw up with the SATA controller, requiring to disable AHCI and enable IDE mode for all but the most timid adjustments. This doesn't seem to happen when using AMD's own SATA controller driver in place of Windows's.
- BCLK cannot really be touched on AM4 boards, unless using a high-end motherboard with an external clock generator
- IGP overclocking may even be pointless for performance, at least with regard to some games on slower DDR3 platforms
- Forcing "directly" any IGP frequency will nuke any kind of idle state and power saving measure. This is not necessarily a deal-breaker, but it probably adds up to the number of reasons "practical overclock" should pass through the SMU-hacking tools.
Intel
- XTU (version notes, caveat) and ThrottleStop, end.
- Well, not so fast. The Plundervolt attack made Intel recommend vendors to enable the "Overclocking Lock" flag in the bios (which, despite the name, at least on normal locked cpus just allowed for voltage control). It wasn't a hard requirement.. but this was just the path of least resistance for most. It's not really difficult to unlock with the usual bios tools, but still.
- BD PROCHOT misbehaving or errata can explain "cold temperature" throttling
- For Turbo Boost to consider a core inactive this has to be at least in C3 sleep, so (at least on non-K cpus normally?) the most extreme quests for minimum latency may end up hurting maximum few-threaded performance without some planning, core pinning and manual fine-tuning of individual per-CPU-registers.
General
- HPET.
- Operating systems might manage to override some BIOS settings, always double check from the live system.
- If for some reason you need to maximize memory frequency/bandwidth (whatever the crazy benchmark or the IGP troubles above), but memory multipliers aren't granular enough, one crafty expedient is to play with the BCLK
- Undervolting is probably the most no-brainer free-lunch improvement of them all, at least in thermally cramped systems (i.e. about every one that isn't a desktop with a <100W cpu)
- Wherever exhaust heat of CPU and GPU is shared on a single heatsink (e.g. most laptops with a common heat pipe and fan) slowing down the CPU to afford more thermal boost headroom for the GPU
or viceversa,can be a winning strategy depending on the application/game/environment. When graphics is integrated on the same chip, I'm afraid only cpu underclocking is gonna be feasible ("Maximum Processor State" may be lowered if lacking specialized tools, even though that's not exactly the most of granularity).. Otherwise with a dedicated gpu there should be plenty of utilities to lower directly the TDP itself.