CPU power consumption: What "TDP" currently means for AMD and Intel

– How can it be that two processors with the same Thermal Design Power (TDP) consume so much differently? Or that Core i7-8700K, -8700 and -8700T with 60 watts difference are almost as fast in the TDP? A look at the current definition of TDP at AMD and Intel provides information about it.

Does a CPU use the TDP?

What does Thermal Design Power (TDP) mean? In 2019, this question remains omnipresent in the community. This has also been ensured by the definitions of the major CPU manufacturers AMD and Intel, which have been repeatedly adapted in the past, sometimes congruent and sometimes complementary definitions, as well as the development of the CPU architectures as such: Because there was no turbo before.

BitcoinMinersHashrate therefore explains what TDP means for current CPUs with turbo from AMD and Intel using the currently valid definitions. Examples and a comparison of the two approaches round off the article. Feedback and suggestions for this article are welcome in the comments.

Intel's current definition of the TDP

Intel currently defines the TDP as "the average power consumption (in watts) that the processor derives when operating on base frequency when all cores are active in a highly complex workload defined by Intel".

What is behind the "highly complex workload“Is still a secret. However, BitcoinMinersHashrate has only recently been able to show that, for example, the benchmark Cinebench R15 in the multi-core test obviously comes close to Intel's load, because the consumption corresponds to the TDP at base clock rates. Orientation is therefore possible.

What matters is the basic clock

However, the much more essential statement in Intel's definition is "when operating on base frequency". The TDP corresponds to the consumption of the CPU at the base clock in Intel's test. With the consumption at turbo clock speeds and in other applications, the definition has with it absolutely nothing to do. How much power a processor consumes in everyday life cannot be derived from the TDP at all, and that CPUs in the Turbo consume twice or three times does not contradict the definition.

Nonetheless, Intel's TDP has at least an indirect impact on performance and consumption with turbo. Because Intel continues to set the direction for manufacturers of cooling systems and complete systems with the TDP. Classifying a Core i7-8700K with 15 instead of 95 watts TDP by simply lowering its base clock from 3.7 GHz to 0.7 GHz would simply lead to it being combined with tiny coolers that ultimately gave it too high temperatures would also prevent a high turbo under load. With a clock rate of 0.7 GHz, this processor would really have nothing in common with the prospect on paper.

TDP and base clock are the lower limit

With the specification of 95 watts at a base clock of 3.7 GHz, Intel ensures that the CPU meets coolers in many computers that can also achieve 4.3 GHz under load on all cores. And those who only meet the minimum requirement, namely 95 watts consumption under load at the base cycle, still deliver the promised base cycle. Important: Both cases do not violate Intel's definition.

The Core i7-8700 (K) / (T) example

The example of the three socketed versions of the Core i7-8700 clearly shows what Intel's TDP definition means and what freedom it leaves manufacturers. Core i7-8700K, i7-8700 and i7-8700T all use the same die and offer all six cores as well as Hyper-Threading. While the K model comes with a TDP of 95 watts, the other two models use 65 or only 35 watts. The difference in the base clock is correspondingly large, which is decisive for the TDP classification: 1.3 GHz or 54 percent separate the K processor from the T processor. The difference in turbo, on the other hand, is smaller: the K processor is only allowed to clock 500 MHz or 13 percent faster than the station wagon. The Core i7-8700 is only a maximum of 100 MHz slower.

As a result, the Core i7-8700K and Core i7-8700 are almost on a motherboard that allows the CPU to use as much power as they need, because they both always operate at the maximum and almost identical Turbo clock rates – unless , the temperature rises too high. The consumption of the two CPUs is then almost identical. Despite a 30 watt difference in the TDP.

TDP = consumption is an option

However, OEMs can also use this difference. The Medion Erazer X67020 / X67015 (test) with Core i7-8700 made this very clear at the end of 2017.

In this system, the processor simply cannot consume more power than 65 watts. If Intel's highly complex load were run on this system, the CPU would consume exactly 65 watts and run at the base clock rates. In applications that generate less load than Intel's reference, on the other hand, the CPU clocks higher at 65 watts. Either way, the processor in this system is significantly slower in applications than on a desktop board. The maximum turbo clock speeds are sometimes only achieved in games because games run less on the processor despite the higher clock speeds and therefore still only consume 65 watts.

The game can also be transferred to the T-model with 35 watts. In a PC with a consumption lid at 35 watts, the clock rate of this CPU would drop to 2.4 GHz under high load (basic clock rate). In contrast, in systems without consumption specifications, the turbo clock rates are present. This processor does not reach the performance level of the other CPUs because Intel has also set the maximum clock rates lower in this case. In this case, too, the distance between the processors would be significantly smaller than the TDP suggests – and the consumption would no longer be significantly lower at 112 watts instead of 35 watts.

• Core i7-8700T & i5-8500T in the test: Intel's six-core with 35 watts "consumption"

The example Core i7-8700 to Core i7-8700K is certainly extreme, because although the base clock and thus TDP of the CPUs differ greatly, the turbo clock rates differ less clearly. But the Core i7-8700T shows that almost every turbo processor can have two sides: one, should the OEM limit consumption to the TDP, and one, the system should recognize the maximum turbo clock speeds as an upper limit. PC manufacturers have to decide which strategy to follow.

AMD's current definition of TDP

AMD's TDP also has the goal of prescribing manufacturers what heat dissipation a cooling system must dissipate at least permanently in order not to slow down the processor. However, AMD does not approach the TDP by means of a concrete measurement of CPU consumption, but rather a theoretical view of the cooling system. At the TDP, AMD therefore also speaks of “thermal watt" Not "electric watts".

AMD says "The TDP is a strict measure of the thermal power loss of an ASIC, which defines the minimum cooling system to achieve the specified performance". The TDP is calculated using a formula that takes into account two properties of the CPU and the cooler and the environmental conditions: the maximum permitted CPU temperature, the heat resistance of the cooler and the temperature of the cooling air.

TDP (Watt) = (tCase (° C) - tAmbient (° C)) / (HSF (° C / W))

The following applies:

• tCase ° C: The maximum allowed temperature at the interface between die and heat spreader to achieve the specified performance
• tAmbient ° C: the maximum allowed temperature of the air at the cooler inlet to achieve the specified performance
• HSF Θca (° C / W): The maximum thermal resistance to be achieved by the cooler in order to achieve the specified performance

On the one hand clear, on the other hand not tangible

The train of thought makes sense: The CPU can only achieve its specified performance if it is not throttled by excessive temperatures. The minimum power dissipation to be dissipated by the cooler is therefore the one that keeps the temperature of the CPU exactly at its critical limit at the maximum permitted ambient temperature.

In contrast to Intel's definition, AMD's definition no longer directly relates to a specific consumption at a specific clock. At first glance, the TDP of a CPU could be chosen as desired by repeatedly choosing the maximum permissible temperature, the assumed air temperature or the performance of the cooling system.

In practice, however, there are of course understandable orders of magnitude: The maximum CPU temperature determines the clock, the cooler the maximum CPU temperature and thus the clock and the air temperature the performance of the cooler. Basically, the formula has a very specific and CPU-specific reference, it is just no longer accessible to the user and the information about the operating state in which the TDP may be equivalent to consumption is not recognizable.

An example: Ryzen 7 2700X

The example of the Ryzen 7 2700X clearly shows how AMD's approach works. The processor may heat up to a maximum of 61.8 ° C at the interface between die and heat spreader and the ambient temperature may not exceed 42.0 ° C – otherwise the performance values ​​provided by AMD will obviously not be achieved. The CPU can therefore be a maximum of 19.8 ° C warmer than the air at the cooler inlet.

For the Ryzen 7 2700X, AMD has specified a cooler with a thermal resistance of at most 0.189 ° C / W. That for each watt of power consumed, the weakest cooler still released heats itself up by 0.189 ° C. So that its temperature and that of the CPU does not rise by more than 19.8 ° C, the CPU must not consume more than 105 watts.

104.76 watts = (61.8 ° C - 42.0 ° C) / (0.189 ° C / W)

A mind game

If AMD had released the Ryzen 7 2700X up to a maximum of 71.8 ° C and would basically assume 32.0 ° C cool air in front of the cooler, the CPU would have a TDP of 210 watts. However, this information would only make sense for the design of the cooling system if the processor also reached such dimensions under load. Significantly higher clock rates and / or higher voltages would be required. Neither that nor the 32 ° C ambient temperature are realistic – this shows that the variables set by AMD are of course firmly related to reality.

Users can grasp the "specified performance“Or ultimately not the consumption at AMD. Measurements of the power consumption do suggest that the TDP also aims at a load state beyond the maximum turbo clock rates. Specifically, the definition of the TDP does not allow this conclusion, as AMD itself has publicly affirmed.

The consumption will usually be higher

However, AMD indicates or indicates that consumption under full load will generally be higher. For example, the manufacturer explains for definition what has to happen so that the processor can work even faster than "specified": The maximum allowed CPU temperature would have to be higher, the ambient temperature lower or the cooler stronger – so there is still scope for the temperature prevalence. And this continuous sounding is exactly what the Turbo (Precision Boost (2), XFR (2)) does at AMD.

At AMD, too, there is another limit in addition to the temperature that can limit the cycle: the maximum permissible consumption. So the thermal consumption "in massively parallelized all-core turbo scenarios“May be exceeded, but the maximum is then defined via the maximum current strength permitted by the base (AM4 = 95 amperes). At 1.35 volts, the consumption is almost 130 watts. That fits more with the consumption for the Ryzen 7 2700X determined by BitcoinMinersHashrate under load.

For the market launch of AMD Ryzen 3000 (test), AMD again defined limit values ​​for short-term loads above these limits in July 2019.

AMD's new processors of the 3000 series are again classified in the TDP classes 65 and 105 watts. However, you can consume more if the temperature is not too high: 88 watts in the 65-watt class and 142 watts in the 105-watt class are the firmly defined limits that AMD Package Power Target (PPT) calls. AMD also defines limit values ​​for the maximum current, which in the case of the 105-watt CPU are once again significantly higher than the 95 amperes for the AM4 socket.

If the processors in Prime95 are then tormented to the maximum using AVX, HWiNFO spits out these limits quite exactly: The two 65-watt processors Ryzen 5 3600 and Ryzen 7 3700X test 90 watts, while the Ryzen 9 3900X with 130 Watt Package Power is measured. The PPT is therefore a maximum consumption defined by AMD. Intel currently does not provide such an indication.

Because the cooling is good enough, this consumption is also permanent – AMD's specifications are not linked to a period of time, but solely to the temperature.

This is how 105, 95 and 65 watts TDP come about

AMD's approach is interesting to define the TDP of different classes. The 65 watt TDP of the Ryzen 7 1700 compared to the Ryzen 7 1700X with 95 watt TDP results, for example, from the combination of a poorer specified cooling system and a higher allowed CPU temperature. However, the idea behind it is clear: the CPU gets the (basic) clock with the consumption, which a much worse cooling system can still dissipate despite the higher permissible temperature. In the specific case, this is possible due to a base clock that clearly drops by 400 MHz and falling multi-core turbo clock rates (not public).

AMD Ryzen is more dependent on the TDP class than Intel Core

Core i7-8700K, -8700 and -8700T with 95, 65 and 35 watts TDP impressively show at Intel that the TDP class alone says nothing about the conductivity of a CPU if the system does not use it and the turbo clock rates from Intel were chosen so close together. At AMD Ryzen it is different. Ryzen 7 2700X (105 watt TDP) and Ryzen 7 2700 (65 watt TDP) only separate 100 MHz at maximum turbo (single-core load), but in everyday use, the load on all cores is clearly different. In this case, AMD has defined either maximum permissible currents or maximum permissible clock rates with larger intervals between the TDP classes, so that CPUs of two TDP classes differ more from one another. However, this information is no longer publicly available.

Further limits

At AMD and Intel, the TDP is the consumption that a cooling system must be able to discharge permanently so that the specified performance (officially the base clock at Intel) can be maintained. It can also be used as a border, but it doesn't have to.

Allow manufacturers of mainboards or PCs of the CPU to run free, but do not necessarily have to achieve or keep the maximum promised turbo clock speeds. As already mentioned, the CPU temperature can speak against it. But consumption can also come into play as a limiting factor. Because CPU manufacturers usually also give processors upper limits for the current intensity to be consumed. In the case of CPUs for which there were turbo clock speeds under load on one, two or any other number of cores, this limit generally never came into play. In the case of the very latest processors, for which there are only basic and single-core turbo clock rates, on the other hand, it is normally the factor that determines the clock – see example AMD Ryzen.

AMD Ryzen Threadripper 2000 (test) has made this clear again these days: The CPU is usually regulated by a consumption limit that is above the TDP under load on x cores.

At first glance, this procedure sounds like a step backwards, but exactly the opposite is the case: Instead of defining fixed maximum turbo clock rates for loads on several cores, which are always achieved but never exceeded, the new approach of the CPU enables Depending on the specific load, clocking sometimes higher and sometimes lower – and not depending on the number of cores used in parallel, but on the very specific requirement measured in terms of power consumption.

Conclusion

In terms of content, AMD and Intel now speak more or less the same language in terms of content: Both relate to the consumption to be dissipated by the cooling system at a specified performance, which is, however, only reasonably tangible at Intel: here it corresponds to the performance with a highly complex load basic clock. AMD's definition, however, does not contain this specific point of contact between TDP, consumption and performance. However, the parallels between the TDP definitions that are relevant in practice cannot be dismissed out of hand.

Users of both manufacturers must be aware that the TDP does not allow any direct statements about consumption and performance in everyday life. Processors from both manufacturers with high maximum turbo clock speeds will also be able to consume significantly more in mainboards with no upper consumption limit. In OEM systems, which in the worst case limit the consumption hard to the TDP, the CPU will be slower and only consume as much as the TDP. The comparison of the three CPUs from the Core i7-8700 series is a very striking current example in this case.

However, the behavior can also come across developments that are not understandable at first glance: for example, louder cooling systems or shorter battery runtimes in the notebook.

A prominent example is Intel's mobile 15-watt CPUs for notebooks. They have offered more cores since 2018 and still higher maximum turbo clock speeds, although the TDP has remained the same. On paper, the CPUs are faster with the same consumption, but they consume more under load apart from the basic clock relevant for the TDP.

How fast a processor is in your own system is not solely a question of TDP and the (turbo) clock rates specified by the manufacturer. On the other hand, it depends on whether the manufacturer sets the TDP as an upper consumption limit or whether the CPU approves as much power consumption as it needs for the maximum turbo. Only AMD at Ryzen currently seems to interpret this more clearly according to the TDP class, so that a Ryzen 7 2700 (65 watts) operates significantly slower than a Ryzen 7 2700X (105 watts). At Intel, on the other hand, maximum turbo and TDP have very little to do with each other in many cases.

This article was interesting, helpful, or both? The editors appreciate any support in the form of deactivated ad blockers or a subscription to BitcoinMinersHashrate. More on the topic of ads on BitcoinMinersHashrate.