Hardware Guides stub

AMD Ryzen TDP Explained: Deep-Dive on TDP Definitions & What Cooler Manufacturers Think

Posted on October 20, 2019

Thermal Design Power, or TDP, is a term used by AMD and Intel to refer in an extremely broad sense to the rate at which a CPU cooler must dissipate heat from the chip to allow it to perform as advertised. Sort of. Depending on the specific formula and product, this number often ends up a combination of science-y variables and voodoo mysticism, ultimately culminating in a figure that’s used to beat-down forum users over which processor has a lower advertised “TDP”. With the push of Ryzen 3000, we’re focusing today on how AMD defines TDP and what its formula actually breaks into, and how that differs from the way cooler manufacturers define it. Buying a 95W TDP processor and a 95W TDP CPU cooler doesn’t mean they’re perfectly matched, and TDP is a much looser calculation than most would expect. There’s also contention between cooler manufacturers and CPU manufacturers over how this should be accurately calculated versus calculated for marketing, something we’ll explore in today’s content.

This content comes from an earlier-published feature-length video we made. We don’t really make any profit on the articles, but maintain them anyway as a point of reference. If you’d like to support deep-dive, long-form content like this, please consider supporting us the following ways:

  1. Watching the video is a great way, but we know that you’re here because you prefer reading! It’s faster, after all
  2. Grabbing a GamersNexus GPU Disassembly Toolkit, an anti-static GN Modmat (available in medium & classic large), T-shirt, GPU anatomy poster, glassware, or other merch to support us
  3. Contributing to our efforts on Patreon

The article continues after the embedded video. Please note that some off-the-cuff/unscripted commentary will not be ported to the article, so you may miss on some commentary, but most of it is here.

 

TDP is a number chosen by AMD and Intel, not calculated. The calculation is secondary to the chosen number, and is used as a means to “show your work.” Ultimately, TDP is picked for some useful purposes and some marketing purposes, and the formula is built around the number. If you don’t believe us and are about to paste the formula for TDP in the comments, stick around for the video and we’ll explain why that is, including some math of our own to show our own work.

TDP has been a pretty heavily mocked number by a subset of the enthusiast userbase, both from Intel and AMD, but it’s also frequently leaned upon in the world’s most important discussions: Reddit arguments where one’s goal is only to assert intelligence over the next user. TDP is confusing in part because it’s assigned a “Watt” unit of measurement, but doesn’t actually correlate to real power consumption in some of the formulas the manufacturers use. Intel i7 CPUs, for instance, can sustain about a 90-95W power consumption under full stock conditions, but only after the turbo boost duration is expired and if MCE is off; otherwise, it bursts much higher, then comes down when the turbo duration limit dictates.

article tdp silverstone

This is really the source of the confusion: You have a number assigned a unit of measurement called “Watts,” but not once in the derivation of that number does a unit of power get calculated. AMD doesn’t even have power, as in “electrical power” (to use AMD’s terminology), anywhere in its formula to define TDP, but people still use “TDP” figures to talk about how much power a CPU consumes. AMD itself uses this figure in press events to knock Intel, even though the two companies have different ways to reverse create whatever TDP value they want. TDP ends up being useless for just about anything that the user might want to do. This requires some more discussion to better explain, and our goal today is to answer a few core questions:

1 - How is TDP calculated and what is it, exactly?

2 - Can you use TDP to determine cooler choice, and if not, what’s the next best means of picking a cooler?

3 - Does TDP correlate with power consumption?

For the last point, we’ll talk about how power consumption of an R9 3900X might be 120W core and 24W uncore under a 24T workload, often running over 140W, but how its TDP is listed as 105W, and then we’ll look into if that’s actually useful for anything.

TDP (Watts) = (tCase°C - tAmbient°C)/(HSF θca)

AMD’s definition of TDP will be dissected over the entirety of this piece, but we’ll start with the formula: TDP (Watts) = (tCase°C - tAmbient°C)/(HSF θca). Critically, and more on this later, we should note that each of these values changes based on the processor, so there’s no fixed set of values and each remains a variable for all processors. This means that TDP can be configured to equal other desirable numbers just by arbitrarily redefining values like tAmbient, or soft numbers like tCase. These can be manipulated for the desired TDP value.

We’ll define each part of this more in a moment, but we should start with a table with the parameters AMD used to calculate each of its TDP values.

AMD’s Math

CPU

tCase°C

tAmbient°C

HSF θca (°C/W)

TDP (W)

Rated P0 Power (W)

Ryzen 9 3950X

?

?

?

~105

?

Ryzen 9 3900X

61.8

42

0.189

104.76

[127W listed, may have been lowered before release]

Ryzen 7 3800X

61.8

42

0.189

104.76

[listed number outdated]

Ryzen 7 3700X

69.3

42

0.420

65

87.8

Ryzen 7 PRO 3700

?

?

?

~65

87.8

Ryzen 5 3600X

69.3

42

0.287

95.12

87

Ryzen 5 3600

69.3

42

0.420

65

62

Thanks to AMD for helping us fill out this chart with tCase, tAmbient, and HSF θca numbers, as well as answering some other questions that arose. The P0 numbers are not from AMD through official channels and are sourced from internal AMD design guidance documents obtained by GamersNexus through other sources, which define P0 as “the highest-power, highest-performance, non-boosted P-state.”

sTRX4/sWRX8 Thermal Requirements Leaked to GamersNexus
GroupTDPTambientTcase MAXTctl MAXThermal Resistance (C/W)
A28032601000.1
B28042811000.14

Note also that tAmbient changes for other processors, like Threadripper parts, where sTRX and sWRX Threadripper 3 parts have tAmbient values in the TDP formula of 32C and 42C between processor groups A and B, thus completely changing the TDP outcome.

Let’s get back to the big table for Ryzen 3000 desktop parts. We’ll highlight the TDP column first: It doesn’t take a genius to see that there’s something strange here. The 3800X, 3900X, and 3950X are all 105W TDP parts, despite all having different clocks and different core counts, and greater thermal requirements on the higher-end parts; the higher core count, higher frequency 3700X has the same 65W TDP as the 3600--not to mention that all of these numbers round cleanly to multiples of 5. We’re going to break down AMD’s explanation of TDP from the reviewer guide for the 3000 series Ryzen CPUs sentence-by-sentence to explain why that’s the case.

AMD’s guide reads the following: “It is a common mistake to conflate thermal watts (TDP) and electrical watts (‘power draw’). Accurate knowledge of what TDP is, and how to calculate it, is therefore vitally important when drawing conclusions about the electrothermal characteristics of a silicon device.”

A Watt is a Watt

A watt is a watt. Wikipedia says a watt is “a derived unit of 1 joule per second, and is used to quantify the rate of energy transfer.” In a Reddit post, AMD’s Robert Hallock says “TDP is about thermal watts, not electrical watts. These are not the same.”

In a literal sense, that’s not true: “thermal” and “electrical” watts are the same unit, in the sense that a cup of flour is the same as a cup of water or a pound of feathers is the same as a pound of bricks. Moreover, the rate that a chip draws energy (as electricity) and releases energy (as heat) should be the same over a sufficient amount of time, since no physical “work” is being done." There is no translation about any axis and no emission in any other form, such as light, and so heat out will be nearly perfectly 1:1 with power in.

Here are the two most important facts to remember during any internet argument about TDP:

  1. Once the system reaches steady state, the rate that heat is released in watts is the same as the rate that electricity is drawn in watts.*
  2. TDP is a man-made rule of thumb, not a precise measurement of thermal output (keep reading).

AMD next says this: “Thermal Design Power (TDP) is strictly the measurement of an ASIC’s thermal output, which defines the cooling solution necessary to achieve rated performance.”

From the confidential engineering documents GamersNexus obtained, AMD’s internal definition is this: “[TDP is] the recommended design target for power to use when designing a processor thermal solution. TDP is defined at nominal voltage and maximum specified case or die temperature. TDP represents the sum of power consumed for all processor voltage rails.”

TDP as defined by cooler manufacturers is different than TDP as defined by AMD or Intel alike, and we’ll cover the cooler manufacturers’ point of view in a moment. They tend to disagree with both Intel and AMD, and vary in their disagreement from casual to vehement, depending on whom you ask. AMD’s TDP does “define the cooling solution necessary,” but maybe not in a way that’s helpful to consumers browsing a cooler manufacturer’s store, for example.

AMD’s reference to measuring an ASIC’s thermal output also requires some explanation. An ASIC is an application-specific integrated circuit, referring to Ryzen CPUs in this context. CPUs are general-purpose and therefore by definition not ASICs, but that’s a technicality, and this explanation is an old one that has been patched and edited at least twice since Ryzen launched in 2017. This sentence in particular has become more loaded than it was back then, because Precision Boost 2 means that performance of Ryzen 3000 chips scales strongly with temperature, and so “rated performance” is now a trickier prospect. As we showed, one of the best ways to overclock Ryzen is to drive its temperature to sub-zero, and so this muddies the waters of what cooling solutions are really expected to achieve at a given TDP. Also, limited-duration boosting means that no single number can effectively summarize thermal output for AMD or Intel. TDP can’t be “strictly the measure of an ASIC’s thermal output,” because the thermal output changes based on workload and test length, among other things.

AMD expands and says the following of its formula: “The TDP formula is straightforward: TDP (Watts) = (tCase°C - tAmbient°C)/(HSF θca)”

tCase°C is externally defined as follows: Maximum temperature for the die/heatspreader junction to achieve rated performance. AMD’s internal definition is this: Maximum case temperature. The maximum temperature when measured at the package location specified by the appropriate thermal design guide. Tcase max is used for thermal solution design and in thermal simulations.

tCase means “case,” as in integrated heat spreader or IHS, not as in computer case. In a very strict sense, this refers to the temperature at the point where silicon meets IHS. Note that this isn’t “how hot does the CPU get” but “how hot can the CPU get before Precision Boost 2 starts to throttle back.” Lower tCase would beget lower TDP in the formula.

The next number in the formula is tAmbient, which is the subtrahend deducted from the minuend tCase before the result is divided by thermal resistance.

AMD defines tAmbient°C as the “maximum temperature at the HSF fan inlet to achieve rated performance.”

Its internal definition is this: “Local ambient temperature at processor heatsink inlet. This is not the external system ambient temperature. Tambient Max is the maximum local ambient temperature supported by the θda [die to ambient?] values stated in the thermal requirements section. If the ambient temperature exceeds Tambient Max, then the thermal solution must be designed to a lower θda to maintain Tdie max at TDP. Tambient Min specifies the minimum local ambient temperature [5°C for Ryzen 3000, but obviously this is just a guideline]. Systems operating in a lower ambient temperature are not supported.”

HSF refers to the heatsink and fan, so the CPU cooler. This is the temperature of the air around the heatsink, whether it’s on an open bench or in a PC case. Lower tAmbient means higher TDP, but tAmbient is defined by AMD in its TDP formula and is not defined by your own tAmbient.

AMD defines HSF θca (°C/W) as: The minimum °C per Watt rating of the heatsink to achieve rated performance.

Its internal definition, for comparison, says “the minimum required heatsink resistance necessary to maintain the case temperature within specification for the thermal design power (TDP) and assumptions for the external ambient temperature and system temperature rise (Tsys).”

“Theta CA”

The HSF is what stands between the CPU and the surrounding air, and θca is the thermal resistance between the CPU and the air, so HSF θca is the thermal resistance of the heatsink. Lower is actually better here, not higher, so AMD’s phrasing has some interpretive gray areas. AMD’s reviewer document should instead read “maximum” instead of “minimum,” so it should be the, quote, “maximum *C per Watt rating of the heatsink,” as lower is better and so maximum would be the last value permissible for rated performance before becoming insufficient for rated performance. When we reached out, AMD clarified that “you can interpret the original copy to mean ‘the [minimum standard]’ where lower values produce superior results.”

HSF θca is a direct measure of required heatsink quality, and from this guide, it appears to be how AMD rates its own coolers. Lower HSF θca means a higher TDP.

But again, none of these numbers are constant. AMD changes the numbers based on which CPU it is, and so each is a variable that has some interpretive dance and voodoo rituals involved for defining the nice, round number that comes out the other side. There are also variables that require other variables in order to define, which means that it’s all just chosen.

Let’s revisit that formula:

All things defined, TDP in watts equals the (difference between recommended max CPU temperature and recommended max ambient temperature) divided by (maximum recommended thermal resistance of heatsink). TDP is a result of plugging recommended room temperature, CPU temperature, and cooler quality into a formula and getting a hard number out, even though the variables that go in can be freely defined and data can be massaged to equate nearly any TDP value, as you’ll see below.

AMD says this in its guide: Using the established TDP formula, we can compute an example in the form of the 105W AMD Ryzen™ 9 3900X: (61.8-42)/0.189 = 104.76 TDP, [with] tCase°C [as] 61.8°C optimal temperature for processor lid.”

The “maximum temperature” wording from the previous section has now been replaced by “optimal temperature.” As we’ve proven in content with liquid nitrogen, stock settings, and frequency scale -- which comically doesn’t technically violate the warranty, even though PBO does -- Ryzen 3000 frequencies keep scaling up with temperature decreases even below zero thanks to Precision Boost 2. Saying 61.8°C is an “optimal” temperature seems arbitrary, but this is the temperature AMD says the chip should be at or below to hit the frequency numbers on the box (though they still wouldn’t until recently). Examples AMD has used in past iterations of this guide were 71.3°C for the 1300X, 60°C for the 1600X, and 56°C for the “Threadripper 1600X” (this was supposed to read 1950X, but was a typo). These numbers are complicated by AMD’s policy of misreporting CPU temperatures on older Ryzen chips, too, where tCTL creates an offset against tDie.

CPU

tCase°C

tAmbient°C

HSF θca (°C/W)

TDP (W)

Rated P0 Power (W)

Ryzen 9 3950X

?

?

?

~105

?

Ryzen 9 3900X

61.8

42

0.189

104.76

[127W listed, may have been lowered before release]

Ryzen 7 3800X

61.8

42

0.189

104.76

[listed number outdated]

Ryzen 7 3700X

69.3

42

0.420

65

87.8

Ryzen 7 PRO 3700

?

?

?

~65

87.8

Ryzen 5 3600X

69.3

42

0.287

95.12

87

Ryzen 5 3600

69.3

42

0.420

65

62

We should get that chart of values for the formula back on the screen. We should also clarify that the tCase temperatures AMD lists in the chart at the start of this content are not hard thresholds beyond which Precision Boost stops working, and likewise they are not hard thresholds under which Precision Boost instantly hits maximum performance (see our LN2 content). Instead, “the CPU will simply draw back on boost if the tCase is exceeded," as AMD says.

Let’s next look back at the tAmbient value: AMD says, for this example, that “42°C [is] optimal ambient temperature for the case at HSF inlet.”

Even if we trust that tCase is based on chip specs, tAmbient certainly is not. Increasing tAmbient makes TDP go down, decreasing tAmbient makes TDP go up, and AMD can freely choose tAmbient. Examining old reviewer guides again, optimal ambient is 32°C for the 1950X, 42°C for the 1300X, and 42°C for the 1600X. It’s 42°C for every Ryzen 3000 chip released so far, which is a hot-but-realistic temperature for air inside a poorly designed PC case.

The next number is for thermal resistance. In its example, AMD chose HSF θca (°C/W): 0.189 θca, then it noted the following: “0.189 θca is the objective AMD specification for cooler thermal performance to achieve rated CPU performance.”

TDP is (according to AMD) a number that indicates required cooler quality. To calculate this number, you need HSF θca, which is “the objective AMD specification for cooler performance.” This is a Catch-22. From AMD’s official Reddit explanation of TDP mentioned earlier, we learn the following: “TDP is a cooler spec to achieve what's printed on the box.” From earlier in the same explanation: “Specifically, θca represents thermal resistance between the CPU heatspreader and the ambient environment. The lower the θca, the better the cooler is.” Because the formula for TDP divides by θca, plugging in a better recommended cooler results in a higher TDP. Past versions of this explanation often include a line like “[stock cooler for the example CPU] corresponds with this specification.”

What TDP Actually Means

Our opinion is this: the AMD TDP formula takes three individually useful numbers and blends them all together into one un-useful number that can’t be reverse engineered back into its original parts. It’s like summarizing someone’s personality by adding their birth date and the number of letters in their last name together. AMD did provide tCase, tAmbient, and HSF θca to us upon request, but those aren’t the numbers that get printed on the box.

Here’s another helpful line from AMD’s commentary: “Thermal conductivity of the processor die, heatspreader, HSF, and junction solder allow the AMD Ryzen processor to amortize the tCase implications of peak power values over time, allowing the CPU to automatically increase performance while remaining inside the thermal boundaries defined by the TDP. In our example processor, Precision Boost 2 will level off at 61.8 tCase°C or 142W peak PPT (whichever comes first).”

Firstly, amortize means “to pay off (an obligation, such as a mortgage) gradually, usually by periodic payments of principal and interest or by payments to a sinking fund.”

This quote is saying that there’s a lot of metal on top of the CPU that acts as a thermal buffer, so that even with spikes in power consumption (i.e. turbo), thermals will be smoothed out into an average that obeys TDP constraints. Intel’s works similarly. An example from an AMD document: “When instantaneous power exceeds TDP, the processor returns to MaxPower within 10 ms and further converges to TDP within 30 ms. The long-term power consumption of the processor computed using an exponential moving average accumulator will not exceed TDP. The potential to consume MaxPower exists, but only in exceptional circumstances when Power Management is disabled.” This is the same argument Intel makes: just because you see short-term power consumption beyond TDP doesn’t mean TDP has been broken. Verifying the truth of that argument is up to reviewers.

Finally, the AMD says this: “Conversely: If the smart algorithms governing Precision Boost 2 detect thermal or electrical conditions beneath these peak values (‘headroom’), it should be well understood by now that the CPU’s boost algorithm is free to convert such headroom into higher average frequencies.”

OK.

So, that’s it for AMD’s official explanation. The big takeaway here is that it’s entirely possible for AMD to pick a TDP that sounds good and work backwards from there, defining the other variables in their algorithm to equal what they want. From AMD’s Reddit statement, again, there’s this quote: “Notice also that this formula allows you to poke things around: a lower ϴca (‘better cooler’) allows for a higher optimal CPU temp.” In response to our own questions, Robert Hallock mentioned that, “as you would expect, the higher-end models assume a superior junction temperature stemming from a better thermal solution.”

Again, TDP is a number chosen by AMD -- or Intel -- and used to work backwards into a formula.

None of the numbers used to calculate TDP are direct measurements of the heat the CPU puts out. The closest one is tCase, which is AMD’s recommended maximum temperature for getting rated performance, NOT a measure of heat output. AMD is telling the truth when they say that TDP has nothing to do with power consumption. Anyone can tell that the 3950X is going to consume more power and run hotter than the 3800X, but the 3800X, 3900X, and 3950X are all 105W TDP parts. That’s completely possible: if AMD recommends the same tCase temperature, same tAmbient temperature, and same cooler quality for all three CPUs, then by definition TDP is the same. Our best guess for why AMD has even bothered to list TDP is that it justifies the use of the same stock cooler for all three chips, aside from usefulness in marketing and proxy internet commenter wars with Intel.

COOLER MANUFACTURER OPINIONS

We spoke to three major CPU cooler manufacturers about AMD’s TDP definition as well. Because of the nature of the responses we received, none of them were willing to go on-record and identify themselves. These are major players in the cooler market with combined experience totaling in the decades, all three of which are engineering-focused, not rebranding-focused.

When a new chip is announced, AMD sends a detailed document to the cooler designers called the Thermal Design Guide, which we’ve obtained, and which contains all the information about power consumption and temperature that are necessary to accurately simulate the chip, as well as the physical hardware to do so (we’re saying AMD because they’re the focus of this article, but all this applies to Intel and NVIDIA as well). This ONLY covers the AMD spec for the CPU, and accounting for breaking spec with overclocking or harsh environments is up to the cooler company. Companies test their own coolers according to their own criteria and may provide their own TDP rating, which may or may not correspond with AMD’s. Either way, presenting a single catch-all number for cooler performance is just as problematic for cooler companies as it is for AMD, maybe even more so--it’s impossible to tell at a glance whether a 200W TDP cooler means “this is suitable for CPUs that AMD has designated 200W” or “we tested this ourselves and it can dissipate 200W of actual power input reliably.”

None of the companies we contacted had a high opinion of TDP as a usable number, ranging from a polite recommendation to rely on test data from reviewers instead, to absolute discontent over what the companies see as a misleading measurement that causes confusion in buyers. One issue is transparency: AMD is very open about the formula used to calculate TDP, but why the number is used at all and why certain numbers are picked for certain processors is a little more vague, and values like tAmbient, and tCase for each CPU weren’t readily available before this content piece. Theta CA isn’t listed on cooler boxes, so that’s also minimally helpful.

CONCLUSION

TDP is an overly reductive number, and it’s only becoming more difficult to define with features like AMD’s Precision Boost 2 that make every individual processor perform differently in different situations. The safe assumption is that any TDP rating, whether it’s from AMD or Intel, is just an easy-to-swallow marketing device that looks good on a box. When buying a cooler, it’s a far better idea to check the cooler manufacturer’s website for recommendations: many have moved away from listing TDP entirely in favor of detailed CPU/cooler compatibility lists. NEVER try to compare Intel and AMD CPUs to each other based on the manufacturer-provided TDP, especially not in all-caps forum posts. It’s not even advisable to make concrete comparisons between different AMD CPUs using TDP. Read a review, pick a CPU, check the website of a reputable cooler manufacturer for recommendations.

If you’re looking for a TLDR, stop: Don’t do yourself a disservice. If you really want to know the answer to the question, read the whole piece. If you need it shortened, well, best not to try and summarize it, because it’s not a simple answer.

Editorial: Patrick Lathan
Additional Research: Steve Burke
Video: Josh Svoboda, Andrew Coleman