How Turbo modes work on the CPU and GPU

As you well know, current CPUs and GPUs use what is called Turbo modes to speed up the clock frequency at given times when it is needed and the conditions for it to do so exist. Well, in this article we will explain how these modes work.

Index of contents

• What is DVFS?
  • DVFS: registered trademarks
• Turbo mode and TDP ratio
• What is Intel Turbo Boost and how does it work?
  • Definition of PL1, PL2 and Tau
• What is AMD Turbo Core and how does it work?
  • Definition of PPT, TOC and EDC
• The case for GPUs

What is DVFS?

DVFS stands for Dynamic Voltage and Frequency Scaling , that is, dynamic voltage and frequency scaling. This is a combination of DFS and DVS, which correspond to dynamic frequency and voltage scaling respectively. Thanks to this throttling technology, energy, temperature, and performance can be better managed according to the load demand at each moment.

Basically, DVFS techniques consist of performing underclocking/undervolting or overclocking/overvolting according to the needs of the system at all times. And all this will be controlled by the operating system through governors that interact with the ACPI system.

As you know, and as I will discuss in more detail later, you should know that the Power dissipated by a chip is :

P = C V ² A F

Where C is the switching capacitance of the chip, V is the voltage at which it works, A is the activity factor that will indicate the average number of transistor switching per clock cycle, and F is the clock frequency at which the chip works. . Therefore, altering the frequency and voltage values, since the others cannot be altered, allows the CPU or GPU to consume and dissipate more or less power. For example, a CPU might be clocked at 500 Mhz when running a word processor, or clocked at 3.2 Ghz when running a video game.

However, as you well know, in current chips, power and complexity have been increased so much that not all parts are always working at the same time, this is what is known as dark silicon . But it may be that some point cores can reach frequencies much higher than stock or nominal, that’s what Turbo mode is all about.

DVFS: registered trademarks

Many manufacturers often use patent trademarks to name their DVFS implementations for marketing purposes, even though they refer to the same thing. This can cause confusion for the user, but here I summarize some of them:

• Intel:
  • SpeedStep: is a technology for dynamically scaling frequency and voltage for the CPU.
  • Turbo Boost:similar to the previous one, but the voltage and clock frequency values ​​are modified to reach very high frequencies at specific times for the CPU.
  • ABT:an exclusive turbo technology for the Core i9.
• AMD:
  • PowerNow! o Cool’n’Quiet: it is the dynamic scaling technology of frequency and voltage in the CPU, equivalent to Intel’s SpeedStep.
  • Turbo Core:to increase the frequency at specific moments in the CPU, similar to Intel’s Turbo Boost.
  • PowerPlay, PowerTune, or ZeroCore Power:Dynamic frequency and voltage scaling for Radeon GPUs.
• NVIDIA:
  • Dynamic Boost– A technology capable of dynamically changing CPU/GPU power.
  • Max-Q: Another dynamic control technology powered by AI and for Mobile GPUs, that is, for laptops.
• VIA Technologies:
  • LongHaul or PowerSaver: Same as SpeedStep or PowerNow!, ie for dynamic frequency and voltage scaling for the CPU.
• Transmit:
  • LongRun:Yet another dynamic scaling of frequency and voltage on the CPU.
• Others…
  • Of course, others also use similar DVFS systems, as is the case with Arm SoCs, Apple, etc.

Turbo mode and TDP ratio

As you well know, the dynamic power or switching power of a chip that I have mentioned above depends on the capacitance, the voltage, the activity factor and the frequency. This power is not only related to the consumption of electrical energy (although the static power would also have to be added to determine the consumption or total power of the chip, since consumption also occurs statically due to leakage currents), also with the emitted temperature that will have to be dissipated.

The voltage and the clock frequency have a direct impact with the dissipated temperature or thermal power, and that is directly related to the TDP (Thermal Design Power) . It is measured in watts (w) and calculated by the chip designer to refer to a heat dissipation parameter that allows you to choose the right cooling system, but it can also be controlled to interfere with Turbo modes as you will see…

What is Intel Turbo Boost and how does it work?

Intel introduced its Turbo Boost technology a long time ago. With it, the clock frequency could be scaled in a timely manner, according to some parameters that we will analyze later, such as the distance to the maximum TDP in which the CPU is at the moment, or the temperature.

This is one of the Turbo modes, and it does not have a fixed clock rate , as many think. What is given in the technical specifications of the processor is the limit frequency to which it can reach. For example, if a processor works at a frequency of 3.7 Ghz and reaches 4.6 Ghz in Turbo Boost mode, this means an increase of up to 900 Mhz, but it will not reach that speed in any case. It will do it by progressive increments, until it is possible.

The first time this technology was implemented, there were 133 Mhz increments in Nehalem and Westmere, now, starting at Sandy Bridge, the jumps are 100 Mhz in 100 Mhz. The Turbo Boost Max version, which has had several revisions since its release, also allows you to increase the clock frequency in steps of 200 Mhz, but will increase it more extreme only on those cores that are better, since they are not all exactly the same. due to differences or manufacturing faults.

What Intel has done is to use Turbo Boost Max for P-Cores, or high performance cores, while for E-Cores, or efficient cores, conventional Turbo Boost is used.

Definition of PL1, PL2 and Tau

To determine how Intel Turbo modes work , you also need to know other important parameters. These concepts are somewhat confusing, and puzzle users. For example, we can find an Intel Core chip that has a TDP of 95W, but can actually reach a maximum TDP of 225W. In this case, the 95W refers to the TDP for the stock frequency, while the maximum TDP refers to the one achieved in Turbo mode.

The concepts that must be highlighted to tend this are:

• PL2 (Power Level 2): refers to the Turbo mode of the CPU, in which a maximum power dissipation is reached.
• PL1 (Power Level 1):is the power when the CPU is working at the stock frequency, that is, without activating Turbo mode.
• Tau: refers to the time that the chip can remain in the PL2 state, that is, in Turbo modes, without taking damage. Therefore, it will be a limited time.

That being said, when a processor goes from PL1 to PL2, it will stay in PL2 for a Tau time. Once that moment arrives, the PL1 state will return to avoid overheating problems, or damage due to a high TDP for a long time. However, it must be said that Intel does not impose a specific PL2, nor a specific Tau. It all depends on the motherboard firmware , which manufacturers can optimize as they prefer for higher PL2 and shorter Tau, or lower PL2 and longer Tau.

For example , an Intel Core i9-10900K has a PL1 of 125W, while the PL2 is of 250W. The Tau is 56 seconds. Instead, the i7-10700K has a 125w PL1, a 229W PL2, and a 56s Tau. While the i7-10700, it is 65W, 224W and 29 seconds for the PL1, PL2 and Tau respectively.

You may also be interested in our guide on the best processors on the market .

What is AMD Turbo Core and how does it work?

AMD has also implemented its own technology, its Turbo modes like Intel. In this case it is known as AMD Turbo Core . The way of working is quite similar, allowing the CPU to go from a certain stock clock frequency to a maximum frequency for a limited time.

AMD Turbo Core can also adapt to heavy loads, increasing performance in a timely manner by increasing the clock frequency. Specifically, it can scale up to 900 Mhz, but in this case the steps will be “shorter” than Intel’s, since it increases by +25 Mhz at each step .

In the same way that it happens with the Intel Turbo Boost, AMD has also improved its versions , with technologies such as:

• Precision Boost: Allows you to increase the clock frequency on two cores or on all cores. If it is done only on two cores, it will be a bigger boost, while if it is applied to all it will be less so as not to generate a very high TDP, or problems of damage due to high temperatures.
• Precision Boost 2:This second generation allows all cores to run at maximum frequency when consumption and temperature allow.
• Precision Boost Overdrive (PBO)– This is similar to Intel Turbo Boost Max, where the forward voltage of the CPU cores is altered and allows for higher performance gains.
• XFR2 (Xtended Frequency Range)– Works together with Turbo Precision Boost 2. It will constantly evaluate how far the clock frequency can be extended based on cooling. So the better the cooling, the more boost you’ll get.

Definition of PPT, TOC and EDC

As was the case with Intel chips, AMD also has its own concepts or terms that can be confusing. Instead of PL2, PL1 or Tau, AMD uses PPT, TDC and EDC. For example, in the case of a Ryzen 7 5800X3D it is:

Currently, AMD has also introduced an ECO mode, in this case it will drop below the default TDP.

If you’re wondering what those AMD terms mean, basically we have:

• TDC (Thermal Design Current): refers to the maximum current (A) that the motherboard can deliver with thermal constraint scenarios.
• EDC (Electrical Design Current)– Electrical Design Current (A), a value indicated to the CPU by the motherboard and representing the maximum current that the VRMs can handle in the short term.
• PPT (Package Power Tracing): is AMD’s way of referring to the maximum consumption that a processor can have, that is, when Turbo modes are active.

The case for GPUs

You may also be interested in the best graphics cards on the market .

Everything said above for the CPU can also be applied to Intel, NVIDIA and AMD GPUs , since they all also implement their own Turbo modes similar to the way CPUs work. Therefore, the theory that you can apply would be the same…