April 19, 2024

What Is TDP for CPUs and GPUs?

What are we talking about ?

To explain what TDP is, one must already deconstruct certain prejudices. and that we aren’t getting to go four ways: the TDP isn’t the consumption of the chip and has nothing to try to to with the latter.

The TDP stands for Thermal Design Power  and may be a value expressed in Watt. this is often why people confuse TDP with “consuming” a chip. Note the quotes, because normally this is often the facility of a chip, because consumption is power times times, Watt.hours… the amalgamation of power and consumption is common among people. Watts means power, but not necessarily electric power . Here it’s an issue of thermal power.

Before going into the small print of how this figure comes out, we are getting to explain the most points to you.


The base

A TDP may be a value that’s calculated from an experiment during a laboratory. It can’t be measured, but is calculated with a formula that we’ll detail below.

The TDP indicates the thermal envelope and thus which cooling system is required to chill the chip. No more no less. an outsized TDP indicates a requirement for giant cooling, alittle TDP indicates a requirement for fewer cooling. Two processors with identical power consumption can in absolute terms have different TDP.

Okay now that the bottom is laid, we’ll be ready to explain the detail.

Crazy scientists’ experiments

First of all, the manufacturer chooses a temperature named tCase. this is often the utmost recommended temperature of the processor heatspreader. It shouldn’t be confused with the tjmax which is that the maximum temperature of the processor. If the tjmax is reached, the CPU throttle to ascertain is cut. for instance , the manufacturer can say “go i will be able to attend tCase = 70 ° C”.
From there, they go to run an entire series of lab experiments. The principle of this series of experiments is, for a given ambient temperature (tAmbiant), to work out what’s the minimum cooling (Θca) which allows the processor to remain below the edge tCase chosen previously. This tAmbiant value corresponds to the temperature in real world at the temperature inside our enclosure.

Each test cooling system features a thermal resistance value, in ° C / W. The lower this value, the higher the cooling.

To come back to the experiment, they’re going to therefore determine an ambient temperature, for instance 40 ° C. Indeed, they often wish to take fairly high values just in case the case is poorly ventilated, hello laptops!


A small example

So lets say our fictitious processor where the manufacturer had chosen a tCase at 70 ° C and a tAmbiant at 40 ° C. They then do their experiment, by decreasing the dimensions of the radiator whenever , to ascertain when the processor will exceed the limit of 70 ° C within the case of very heavy use.

radiator temperature TDP First experiment with an outsized radiator with a thermal resistance of 0.15 ° C / W. No problem, the processor stays below 70 ° C. They use smaller coolers whenever (for example 0.20 ° C / W, then 0.22 ° C / W, then 0.25 ° C / W then 0.27 ° C / W). and that they see that with 0.25 ° C / W, they manage to stay the processor below the 70 ° C of the tCase, while thereupon at 0.27 ° C / W the processor exceeds the limit. it’s then considered that the radiator at 0.25 ° C / W is that the minimum cooling.


The Magic Formula

And this is often where the TDP formula comes in. It only uses these 3 values that we’ve explained: tCase, tAmbiant and Θca (the thermal resistance of the minimum heatsink which allows the processor to be kept under tCase).

Here is that the formula: TDP = (tCase – tAmbiant) / Θca

In our case, this makes: TDP = (70 – 40) / 0.25 = 120 W

Here the TDP is now calculated. As you’ll have noticed, at no time was CPU consumption taken under consideration . it had been all about temperature and warmth dissipation here.

Cheating the TDP

And this value, what’s it used for? it’s mainly employed by OEMs to size their cooling to a variety of processors. But ultimately, what is going on to form one CPU (or GPU) have a robust TDP and another not, or how can we influence the TDP?

Already, the primary point: the tCase. it had been said at the start that the manufacturer chose a threshold temperature. But what would prevent them from choosing a better maximum temperature (and therefore much less restrictive)? Nothing in the least .

If they choose a better max temperature, say 80 ° C for instance , as a results of their experiment, they might see for instance that a radiator with resistance of 0.4 ° C / W would be enough to stay it below 80 ° C. during which case the TDP would become (80-40) /0.40 = 100 W. Hop, by arbitrarily increasing the utmost temperature, they obtain a lower TDP. And it makes sense: if you’re more lax with reference to temperature, the required cooling is a smaller amount .

The same goes for tAmbiant: they might lower the ambient temperature during their lab tests, and this is able to have the consequence of lowering the TDP. on the other hand they take the danger that the worth they get at the top doesn’t take under consideration poorly ventilated cases, for instance . and therefore the last item a manufacturer wants is service returns or dissatisfied OEMs.


Indium seal vs Nutella

Finally, consumption necessarily influences the TDP since a CPU that heats up will mechanically have a higher temperature on a given cooling. But the influence is not direct on the TDP, as we saw above. Now imagine our previous hypothetical processor with a TDP of 120W (with the following calculation: (70-40) /0.25).

Imagine that the manufacturer decides to exchange the indium seal with an indoor thermal paste. The direct consequence is that the processor will heat up more, albeit the consumption remains an equivalent . But that this heating are going to be transmitted less to the heatspreader (IHS) thus giving a drop by the temperature of the latter. But on the opposite hand, CPU calories are harder to release (since there’s some quite bottleneck at the thermal seal), which increases the interior temperature of the processor. And so, counting on the case, the TDP can increase, or decrease, while the processor remains an equivalent , if it’s not for the thermal seal.

For example, Intel had changed its thermal seal type between the i7 2600K and therefore the i7 3770K. Remember the switch from indium solder to thermal paste. And if we’ll see the specifications of the latter, the i7 3770K had a lower maximum tCase (it goes from 72.6 ° C to 67.4 ° C), but a better Tjmax (it goes from 98 ° C to 105 ° C) . This resulted during a TDP which remained stable compared to the 2600K (it drops by a couple of Watts while the consumption has also fallen by a couple of Watts).

Finally, the last issue is: “Under what CPU load conditions exactly are these measurements made?” “. there’s a transparent difference in CPU load between computer game , 3D creation, AVX encoding, or torture test.

The same goes for tAmbiant: they might lower the ambient temperature during their lab tests, and this is able to have the consequence of lowering the TDP. on the other hand they take the danger that the worth they get at the top doesn’t take under consideration poorly ventilated cases, for instance . and therefore the last item a manufacturer wants is service returns or dissatisfied OEMs.

TDP and consumption

This is what TDP is. But finally, we’ll briefly tell you about other uses of this value. Some manufacturers use, within certain limits in fact , the worth of the TDP as a consumption limit. The processor can throttle if it exceeds it, this is often the famous power throttle.

Consumption This principle, widely used, may be a bit particular in logic. Indeed, during this case we normally calculate a TDP supported experience, then we use it as a limitation of consumption, which affects the previously performed experiment. In fact, in practice, we will think that the manufacturer chooses a TDP before the experiment, applies the consumption limit within the processor thereto , and performs the experiment, which makes it possible to reach the ultimate value of the TDP already chosen. it’s a touch of a self-fulfilling effect, a touch sort of a feedback circuit .

This is particularly the case with AMD A-series which have “variable” TDP (you can choose from 95W and 65 W for instance , during which case the operating frequency differs, also because the consumption within the end). And without even talking about variable TDP, the straightforward fact of introducing a Turbo frequency which differs consistent with the amount of cores used allows the processor to remain within the nails even with an outsized multithreaded load, while having an honest frequency when few threads. are used.


The moral of all of this is that you shouldn’t pass off TDP for what it isn’t. It is not a measurement of consumption, but a calculated value of heat release (and therefore adequate cooling). This value can be used as a consumption limit, or not, but that is not necessarily the reason for this value. This was hijacked with the advent of Turbo et al. This is the reason why you will often see “such processor has exceeded its TDP” written, which does not make the TDP value wrong.

Leave a Reply

Your email address will not be published. Required fields are marked *