Thermal design power

from Wikipedia, the free encyclopedia

With thermal design power (abbreviation: TDP , occasional incorrect: Thermal Design Point ) is used in the electronics industry , a maximum value (some manufacturers also the average value) for the thermal dissipation of electrical components, or a processor referred to on the basis of the cooling ( if necessary) and the power supply. The TDP is usually less than the real maximum power loss. Depending on the type of component, cooling system and the ambient temperature (mostly air temperature inside a housing), it must be ensured that the waste heat generated is dissipated even in special conditions (high ambient temperature and high processor load). This creates a conflict of objectives between performance, costs, noise pollution and indoor climate. The TDP was introduced in order to be able to plan the thermal dimensioning of a system in advance. To determine the TDP, load cases are used that occur with typical maximum stress in real use, for example with x86 processors when coding videos or operating at maximum core voltage with maximum permissible core temperature.

Application (x86)

Manufacturers define the TDP differently. Often it corresponds to the maximum possible power loss, so that the cooling must be designed for the corresponding waste heat. For their calculation, the manufacturer does not use standardized methods that can also change at the manufacturer. A uniform method that spans manufacturer and processor generations would be very difficult to implement, however, since different microarchitectures and different processor families or processor generations require a different distribution of the load in order to be fully utilized. In addition, more and more parts are added to the actual CPU in modern processors (hence the correct use of the term "processor") such as caches , integrated graphics units , controllers for RAM , PCIe or HyperTransport interfaces and voltage converters that deal with Do not allow the generation of pure processor load to be used, but have a share of the total TDP in the data sheet. The TDP specified in the data sheet can be exceeded with non-application processes that are used more for test purposes, such as burn-in .

With x86-compatible processors from 2004, the TDP came increasingly into the focus of chip development with regard to mobile devices. There is a tendency to optimize the amount of waste heat both in idle mode and under full load. Instruction set extensions and optimizations to the micro-architectures as well as dynamic energy management such as PowerNow! , Enhanced PowerNow! and Cool'n'Quiet from AMD as well as the Intel SpeedStep technology have achieved an increase in performance with a simultaneous reduction in energy consumption. Another development direction is the maximum exploitation of the TDP. In new processors that support the automatic self- overclocking functions AMD Turbo Core or Intel Turbo Boost , individual, unused cores can be clocked down and switched off so that their share of the TDP is available to another core and this can thus achieve a higher clock frequency, which is at single-threaded applications offer a decisive advantage. If the internal control of the processor determines that there are still reserves for the current power consumption and thermal budget, the processor can dynamically overclock all cores and thus move closer to the TDP. This possible overclocking of all cores is sometimes given in data sheets as the maximum turbo frequency.

While different processors used to have different TDP values, today there are more and more TDP classes into which processors are sorted. For desktop processors, typical classes would be 25, 35, 45, 65, 73, 80, 95, 125 and 130 watts. For mobile processors in notebooks, the classes would be 17, 25, 35 and 45 watts.

Due to the increased attention to the TDP, however, it is also increasingly becoming a marketing tool for manufacturers. So there are For example, Nvidia offers a lower TDP for the Geforce 8800 Ultra graphics card than for a Geforce 8800 GTX , although the same graphics processor with a higher voltage and higher clock rates is used on the 8800 Ultra and - when the power consumption was checked - yielded a higher value as expected . The lower specification than for the 8800 GTX was possible because it had been given an unnecessarily high TDP value.

Significance of the TDP (x86)

Although the TDP is an important property of a component, it only indicates the maximum expected heat to be dissipated and is therefore not suitable for determining the typical power consumption of a processor or even an entire system when idling or under a certain load. Benchmark programs that are supposed to fully utilize an x86 processor (e.g. Cinebench, Core2MaxPerf, CPU Burn-in, IntelBurnTest or Prime95 ) are sometimes below the specified TDP. One reason for this can be that certain components of the arithmetic logic unit are not fully utilized at the same time or only with certain command combinations or command sets (e.g. AVX ). In addition, depending on the application, the capacity utilization of a processor can be limited by the data transmission rate, namely if a large amount of data is to be processed at which the required transmission speed can only be provided by the comparatively small caches. Even with speed-optimized program libraries (e.g. Linpack ), only 80 percent of the theoretical maximum is used on average. Processors outside of the embedded area with a cache that is too small or missing can therefore no longer be found.

The TDP (even in combination with the clock frequency) also says little about the efficiency of a processor. For example, an Intel i7-980X with 6 cores, 3333 MHz maximum clock frequency and a TDP of 130 watts, under full load, requires 19 times more power than an Intel Atom N450 with one core, 1667 MHz maximum clock frequency and a TDP of 5 , 5 watts, but the Intel i7-980X has a 30 times higher final score in the “Cinebench R10” test.

Due to production-related variances, the actual power consumption and other parameters can vary between identical models. Also can Stepping have an impact.

Despite the low informative value of the typical power consumption, the manufacturers' data sheets hardly specify more than the TDP. This is due, among other things, to the fact that it is even more difficult to determine a typical than a maximum waste heat, since this depends very much on the individual application (e.g. on the software used and components built into the system) and hardly allows generalizations. Solutions from the manufacturers of x86-compatible processors are the Average CPU Power (ACP) from AMD and the Scenario Design Power (SDP) from Intel, which are intended to describe the power consumption or the waste heat output in typical use.

With the introduction of functions such as Turbo Boost , the informative value of the TDP is further limited, because at Intel, the TDP is "the average power consumption (in watts) that the processor derives when operating at the base frequency when all cores are under a highly complex workload defined by Intel are active ". There is no information from any CPU manufacturer about the maximum power consumption when the boost clock is applied, which can go far beyond the base frequency.

Web links

Individual evidence

  1. ↑ a b Christof Windeck: "Hitzewelle", c't - magazine for computer technology - issue 15 from 2010 p. 174
  2. ↑ a b Florian Müssig: "Auf allen vieren", c't - magazine for computer technology - issue 10 from 2012 p. 94
  3. ↑ Jan-Frederik Timm: CPU power consumption: What "TDP" currently means at AMD and Intel. Retrieved January 10, 2020 .