Why do CPUs downclock when in full load?

[Gat Pelsinger]

First of all, if a CPU is rated to reach a maximum boost clock of X GHz, then I've never ever seen it reach the limit. For example, my older laptop's CPU was designed for a max turbo of 2.70 GHz, but I have never seen it clock that high. The highest I have seen is 2.67, and that too when I was running a specific workload which demanding highest clocks or highest single threaded performance with not too much load so that it won't downclock.

 

But why do CPUs downclock when in load? No, I am not talking about the thermal and power limits. As a matter of fact, I've never seen an actual CPU downclock because of just load, because my laptop CPUs always get downclocked my power limits.

[Helpful Tech Witch]

4 minutes ago, Hensen Juang said:

But why do CPUs downclock when in load? No, I am not talking about the thermal and power limits. As a matter of fact, I've never seen an actual CPU downclock because of just load, because my laptop CPUs always get downclocked my power limits.

Usually, its because the boost clock is for a single core boost. If a load is multicore, itll never reach that boost.

Also, most the time, the power or temperature is why cpus downclock.

[igormp]

The boost clocks are not for all-core, but rather single-core, and a CPU has a limited energy and thermal capacity, so when going for all load and having all cores asking for power, you now have to distribute this energy and thermal capacity among all cores instead of just a single one, thus reducing the all-core clocks.

 

So yes, it is related to thermal and power. If you improve cooling and give it more power, you're able to reach higher all-core clocks.

 

Also, laptops have a limit on the amount of time they can boost, so it won't run at boost at all times in order to respect its thermal envelope.

[Gat Pelsinger]

@igormp So, if I remove all power limits in BIOS and make sure I have enough thermal dissipation, I can get EXACTLY the same frequency on all cores that I would get on a single core, or is there going to be something hold be a little behind?

[igormp]

14 minutes ago, Hensen Juang said:

or is there going to be something hold be a little behind?

Not all cores are equal and there are specific cores than can actually achieve those higher frequencies. If you manage to get a really golden chip and remove all limits, then yes, you'd be able to reach all core boosts, and that's what most OCers do.

[Mister Woof]

On a laptop I would not expect to regularly see high boost clocks. It's a combination of firmware power limits, thermal throttling, and general locked nature of intel CPUs

[YoungBlade]

33 minutes ago, Hensen Juang said:

@igormp So, if I remove all power limits in BIOS and make sure I have enough thermal dissipation, I can get EXACTLY the same frequency on all cores that I would get on a single core, or is there going to be something hold be a little behind?

Without an overclock, you won't see the max boost frequency on all cores. Intel used to actually be open about this and publish what were called "Boost Tables" that told you exactly what clockspeed to expect when having 1, 2, 3, etc. cores under load. Here's an old example from the first gen core series:

 

 

So the i7 870 will only boost to its rated 3.6GHz max if a single core is in use. If 3 or 4 cores are in use, it will only boost to 3.2GHz. All Intel CPUs still have a boost table to this day, and you can actually find it via software like Intel XTU, but Intel no longer makes the information easily publicly available.

 

For AMD, it's more complicated, and the exact way they determine this is not publicly known - at least, I've never seen it published anyware. They don't have a basic Boost Table like Intel does, and instead use an algorithm based on temperature, power, and current. But the result is similar: In practice, you're ever going to see the max rated boost clock on an all core load, even if you remove power limits.

[Gat Pelsinger]

@YoungBlade The question is, why though?

[YoungBlade]

2 hours ago, Hensen Juang said:

@YoungBlade The question is, why though?

There's a number of reasons for this.

 

First is power efficiency. It takes a lot more power to get a CPU to clock beyond its efficiency window. So limiting the max clockspeed boost to fewer cores makes it easier to stay within the rated max power draw of the processor.

 

Second is heat dissipation. If you just unlocked the power limit and threw efficiency out the window, you're going to find that the CPU is much harder to cool. This means that default cooling solutions - the box cooler or the cooler in OEM systems - would need to be a lot beefier (i.e. more expensive) like when AMD included a box AIO with the power hungry FX-9590.

 

Third is silicon fitness. The max boosting core is not random. Software today can tell you which cores are flagged as the best and second best. Some cores may not actually be able to reach the max boost speed at stock voltages. If they were all required to, the CPU would have failed validation. By limiting the all-core boost, more chips are able to pass validation and be sold as higher core count parts.

[freeagent]

Usually when clocks get pulled back its for a couple of different reasons. Self preservation- heat is still the enemy, power target could have been reached, the time it is allowed to spend at that clock may have been reached.

 

If you have never seen full boost, then there is a very good chance you would not be able to control it. It is called thermal runaway, and it can happen with ambient cooling if the conditions are met.