AMD Ryzen TDP Deep-Dive & What Cooler Manufacturers Think of TDP
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
Rory
Lol, I got a bit lost, so the processor cooler now has to pay a mortgage? Many years ago AMD budget for coolers was 8 USD. They were thinking of doing away with them all together. I did a video on the wraith prism. Now I spoke in English throughout. CPU temperature is a variable as is ambient air. The thermal mass, takes the variable of the CPU only. The cooler has two values, the tcase to HSF and the thermal conductivity. Then the fan/fins ability to cool the mass as dispensation from the mass, This is highly related to air temps and dispensing. Or and case tower internal temperature. As you state the ambient of 5oC is a delusional number to use, sit for four hours in that temp and you would surely be dead. Now 15oC would be more realistic. That would make there coolers 66% off mark. Now in my tests, I made the 2700x 120w The cooler 20% defuncked to cool the CPU at 15-20oC And the thermal mass about 10% inadequate. Also the AMD software, peaks system parameters out side of spec, and design. In short is not fit for purpose as it runs hotter and over powered that it needs to be. When cutting all auto and going manually I droped 20oC and droped voltages to stable levels on which AMD even test and bench the CPUs for sales and quality. In short, like my findings during summer or when heating your home or office with a stock cooler, you will get too hot and throttle back. This will last some time due to the fact the cooler is in effective in that variable (the real world. Load +voltage /(cooling dispensation/air temp) equals fit for purpose. Max performance equals voltage/temp/ cooling effect X transference abientT/rate (T/M X t) Aka. If you got mass to take the heat and a way to get the heat away at a rate which allows cooling faster or equal to the rise of temperature. The fall in temp relates to airflow and temperature. Mass is required to absorb variable temps at the CPU case and the coolers ability to spread heat is the quality of the cooling device(rating)
reply
Lol, I got a bit lost, so the processor cooler now has to pay a mortgage? Many years ago AMD budget for coolers was 8 USD. They were thinking of doing away with them all together. I did a video on the wraith prism. Now I spoke in English throughout. CPU temperature is a variable as is ambient air. The thermal mass, takes the variable of the CPU only. The cooler has two values, the tcase to HSF and the thermal conductivity. Then the fan/fins ability to cool the mass as dispensation from the mass, This is highly related to air temps and dispensing. Or and case tower internal temperature. As you state the ambient of 5oC is a delusional number to use, sit for four hours in that temp and you would surely be dead. Now 15oC would be more realistic. That would make there coolers 66% off mark. Now in my tests, I made the 2700x 120w The cooler 20% defuncked to cool the CPU at 15-20oC And the thermal mass about 10% inadequate. Also the AMD software, peaks system parameters out side of spec, and design. In short is not fit for purpose as it runs hotter and over powered that it needs to be. When cutting all auto and going manually I droped 20oC and droped voltages to stable levels on which AMD even test and bench the CPUs for sales and quality. In short, like my findings during summer or when heating your home or office with a stock cooler, you will get too hot and throttle back. This will last some time due to the fact the cooler is in effective in that variable (the real world. Load +voltage /(cooling dispensation/air temp) equals fit for purpose. Max performance equals voltage/temp/ cooling effect X transference abientT/rate (T/M X t) Aka. If you got mass to take the heat and a way to get the heat away at a rate which allows cooling faster or equal to the rise of temperature. The fall in temp relates to airflow and temperature. Mass is required to absorb variable temps at the CPU case and the coolers ability to spread heat is the quality of the cooling device(rating)
reply
KX36
So from an electrical engineering point of view: The formula is TDP = delta temperature / thermal resistance. This is very similar to equations electrical engineers use every day to specify hearsinks. You would design for a whole chain of thermal interfaces from the heat source (the die) to the ultimate heat sink (the air. Thermal resistances of each layer add up (die to IHS to paste to heatsink to air. The outcome of the formula in electrical engineering is a sentence like the combined thermal resistance of the paste and heatsink must be less than X degrees C per Watt. The problem comes as power is an integral over time rather than an instantaneous thing. ( i. e. Joules per second) and depending on the timespan (1 second? 1 hour? 1 year) the power will be different because of things like turbo. So AMD and intel have to make a design decision as to what power rating to use in this specification, and this single estimation is called TDP. In EE, the delta of temperature from die to ambient would typically be conservative and use the maximum safe die temperature and approximate the worst case ambient air temperature. (As performance of a CPU changes dramatically with temperature, they could substitude maximum safe temperature with optimum temperature for peak performance. I can see why they'd specifiy a different ambient air temperature for server parts for instance. Theta of die to IHS is controlled and know by AMD/intel. Theta of IHS to air is not and therefore they must make this recommendation to a heatsink manufacturer. I think the whole thing would be much simpler if AMD and intel therefore simply specified a maximum theta_ca for 3rd party heatsinks and kept the estimated thermal power used in the equation to themselves. Or better still, specify a range of theta_ca (based on a range of ambient temperature and a fixed optimum die temperature) which the heatsink+fan should completely cover by way of its variable fan speed.
reply
So from an electrical engineering point of view: The formula is TDP = delta temperature / thermal resistance. This is very similar to equations electrical engineers use every day to specify hearsinks. You would design for a whole chain of thermal interfaces from the heat source (the die) to the ultimate heat sink (the air. Thermal resistances of each layer add up (die to IHS to paste to heatsink to air. The outcome of the formula in electrical engineering is a sentence like the combined thermal resistance of the paste and heatsink must be less than X degrees C per Watt. The problem comes as power is an integral over time rather than an instantaneous thing. ( i. e. Joules per second) and depending on the timespan (1 second? 1 hour? 1 year) the power will be different because of things like turbo. So AMD and intel have to make a design decision as to what power rating to use in this specification, and this single estimation is called TDP. In EE, the delta of temperature from die to ambient would typically be conservative and use the maximum safe die temperature and approximate the worst case ambient air temperature. (As performance of a CPU changes dramatically with temperature, they could substitude maximum safe temperature with optimum temperature for peak performance. I can see why they'd specifiy a different ambient air temperature for server parts for instance. Theta of die to IHS is controlled and know by AMD/intel. Theta of IHS to air is not and therefore they must make this recommendation to a heatsink manufacturer. I think the whole thing would be much simpler if AMD and intel therefore simply specified a maximum theta_ca for 3rd party heatsinks and kept the estimated thermal power used in the equation to themselves. Or better still, specify a range of theta_ca (based on a range of ambient temperature and a fixed optimum die temperature) which the heatsink+fan should completely cover by way of its variable fan speed.
reply
Siana
There's no emission of energy as light - oh yes there is, it's called black body radiation and is deep infrared for any surface you might vaguely consider touching. Unfortunately the CPU is confined from all sides, it has no radiating surface, though the heatsink does. Then again the contribution of that even at the heatsink level is negligible. Not all of thermal output of the processor needs to be handled by the CPU cooler, but on higher TDPs it really doesn't matter that much. But yes claiming that electrical Watt and thermal Watt is not the same is extremely misleading. For that to be the case, the CPU would have to be doing useful work, rather than just sitting idle and occasionally running Counter Strike. (just kidding) Mhm is the modern AM4-socket CPU an ASIC? To an extent it is, it has only one application as in one type of product - it's designed to be used in a PC, and any product that contains this IC becomes a PC. What is not an ASIC is for example an FPGA, a RAM IC or a 74-series logic IC, those ICs are designed independent of any particular product they might go into. This is a modern trend. Back in the day, the 6502 or Z80 was definitely not an ASIC because the product it might go into might not in fact be any kind of computer, but a control circuit or appliance of some kind.
reply
There's no emission of energy as light - oh yes there is, it's called black body radiation and is deep infrared for any surface you might vaguely consider touching. Unfortunately the CPU is confined from all sides, it has no radiating surface, though the heatsink does. Then again the contribution of that even at the heatsink level is negligible. Not all of thermal output of the processor needs to be handled by the CPU cooler, but on higher TDPs it really doesn't matter that much. But yes claiming that electrical Watt and thermal Watt is not the same is extremely misleading. For that to be the case, the CPU would have to be doing useful work, rather than just sitting idle and occasionally running Counter Strike. (just kidding) Mhm is the modern AM4-socket CPU an ASIC? To an extent it is, it has only one application as in one type of product - it's designed to be used in a PC, and any product that contains this IC becomes a PC. What is not an ASIC is for example an FPGA, a RAM IC or a 74-series logic IC, those ICs are designed independent of any particular product they might go into. This is a modern trend. Back in the day, the 6502 or Z80 was definitely not an ASIC because the product it might go into might not in fact be any kind of computer, but a control circuit or appliance of some kind.
reply
Artur
I've watched about 5 minutes of the video, until AMDs formula. And what do I get by just reading the formula is that TDP is a bunch of crap that manufacturer puts on the box! Watt is a measurement of power, since a CPU doesn't transform energy in any other way than heat (theres no lighting or movement production for instance) power consumed = power dissipated! Thats simple. If the power is not being dissipated is because its beign redirected trough any IO pins but that can be in margin of error for the purposes and shoudn't be accounted since its not the CPU itself that is consuming its another device. In resume, when we are talking about watt which is a unit of power and on a CPU specifically since it just transform power into heat we can conclude that power dissipated = power comsumed and since TDP stands for thermal design power and is measured in watts any value thats different from thats the CPU consumes is manufacturer BS! And yes! Your computer is a fancy and expensive heater with RGB! That's all it does, produces heat!
reply
I've watched about 5 minutes of the video, until AMDs formula. And what do I get by just reading the formula is that TDP is a bunch of crap that manufacturer puts on the box! Watt is a measurement of power, since a CPU doesn't transform energy in any other way than heat (theres no lighting or movement production for instance) power consumed = power dissipated! Thats simple. If the power is not being dissipated is because its beign redirected trough any IO pins but that can be in margin of error for the purposes and shoudn't be accounted since its not the CPU itself that is consuming its another device. In resume, when we are talking about watt which is a unit of power and on a CPU specifically since it just transform power into heat we can conclude that power dissipated = power comsumed and since TDP stands for thermal design power and is measured in watts any value thats different from thats the CPU consumes is manufacturer BS! And yes! Your computer is a fancy and expensive heater with RGB! That's all it does, produces heat!
reply
Quetzalcoalt
So according to GN, a CPU is a magical device that consumes 100% of the electricity and is released as heat. ( the 1 to 1 part in the beginning) No electrical data (1s and 0s, they are still electricity) is passed to other devices, and the resistance of the CPU is 0 that's why 100w power = 100w of heat output. Watts are not the same. That's why we have watts power, watts heat, watts whatever. You can't say temperature is just temperature because we have C, F and K. When you buy a stove, and it's a 1800w one, that's the pulling watts not the heat released in watts. If a 1800w stove heats my food to cook the food, above 150C or who knows how much then my 1800w vacuum should melt if i use it for more than 3 minutes, if electrical watts are the same as heat watts. When i have a 2x50w power amplifier. The amp is pulling 150-200watts, and the amp does NOT heat up to those numbers. Some of the heat is send to the speaker coils. They also heat up. Or maybe they take the energy from the universe?
reply
So according to GN, a CPU is a magical device that consumes 100% of the electricity and is released as heat. ( the 1 to 1 part in the beginning) No electrical data (1s and 0s, they are still electricity) is passed to other devices, and the resistance of the CPU is 0 that's why 100w power = 100w of heat output. Watts are not the same. That's why we have watts power, watts heat, watts whatever. You can't say temperature is just temperature because we have C, F and K. When you buy a stove, and it's a 1800w one, that's the pulling watts not the heat released in watts. If a 1800w stove heats my food to cook the food, above 150C or who knows how much then my 1800w vacuum should melt if i use it for more than 3 minutes, if electrical watts are the same as heat watts. When i have a 2x50w power amplifier. The amp is pulling 150-200watts, and the amp does NOT heat up to those numbers. Some of the heat is send to the speaker coils. They also heat up. Or maybe they take the energy from the universe?
reply
Chilledfish
I'm definitely a bit late, and it has probably have been said, but thermal watts and power watts don't refer to the same thing! That's because power correlates to energy, and different types of energy aren't interchangeable. A power plant produces produces X amounts of watts; both thermal and electrical. Because converting one to the other is always at a loss, you're always going to end up with thermal watts that aren't going to get to any powerline. In fact, in nuclear reactors, even if the reactor is completely shut down and isn't creating a watt of electricity, it's still putting out a few megawatts of thermal power. The point that you are probably correct about is that a certain amount of power draw, should produce a comparable amount of heat. That's because the loss to heat is probably pretty much the same regardless of the component. That's not to say that a CPU that draws 300 watts, will require to dissipate 300 watts as heat. That would leave 0 watts for actual work.
reply
I'm definitely a bit late, and it has probably have been said, but thermal watts and power watts don't refer to the same thing! That's because power correlates to energy, and different types of energy aren't interchangeable. A power plant produces produces X amounts of watts; both thermal and electrical. Because converting one to the other is always at a loss, you're always going to end up with thermal watts that aren't going to get to any powerline. In fact, in nuclear reactors, even if the reactor is completely shut down and isn't creating a watt of electricity, it's still putting out a few megawatts of thermal power. The point that you are probably correct about is that a certain amount of power draw, should produce a comparable amount of heat. That's because the loss to heat is probably pretty much the same regardless of the component. That's not to say that a CPU that draws 300 watts, will require to dissipate 300 watts as heat. That would leave 0 watts for actual work.
reply
Lars
Loved this video, I will have to watch this again to understand better. I struggled with a lot of the formulas and equations. There were a lot of missed opportunities for visual explanations that were filled with b-roll. There was a previous GN video that had really cool 3D effects to show how cooling works, I wish some of those could have been updated and used here. Also please give us some hypothetical examples. You say that these numbers can be changed to fit the marketing, so show us! What would it look like if the 3700X had shipped with a Wraith Stealth? Could AMD have marketed it as 45w by changing the numbers in the equation? It's hard to apply these concepts when the explanation stays only in the theoretical realm. Maybe revisit when 3950X comes out and the inevitable discussion on power consumption is happening. Thanks for your hard work GN. You guys rock.
reply
Loved this video, I will have to watch this again to understand better. I struggled with a lot of the formulas and equations. There were a lot of missed opportunities for visual explanations that were filled with b-roll. There was a previous GN video that had really cool 3D effects to show how cooling works, I wish some of those could have been updated and used here. Also please give us some hypothetical examples. You say that these numbers can be changed to fit the marketing, so show us! What would it look like if the 3700X had shipped with a Wraith Stealth? Could AMD have marketed it as 45w by changing the numbers in the equation? It's hard to apply these concepts when the explanation stays only in the theoretical realm. Maybe revisit when 3950X comes out and the inevitable discussion on power consumption is happening. Thanks for your hard work GN. You guys rock.
reply
N'ihl
My BeQuitet! Dark Rock Advanced is rated at 180W TDP, but when you add second fan on back for push-pull, it can handle 200W+ TDP, with its single silent wing fan, a FX chip clocked to draw roughly 180W under load, and this cooler crap itselfs. also, my Zalman CNPS7000, which is rated 95W, handle FX8320 at 125W TDP stock at ease, keeping it roughly at 65c on full load with boost engaged with push-pull negative pressure in case and roughly 22c ambient. pretty loose adaptation of 180W TDP, 125W TDP and 95W TDP. hard to really fit in, if you noise-optimise, you can drop that TDP on cooler significantly, if it require high FAN RPM to achieve it.
reply
My BeQuitet! Dark Rock Advanced is rated at 180W TDP, but when you add second fan on back for push-pull, it can handle 200W+ TDP, with its single silent wing fan, a FX chip clocked to draw roughly 180W under load, and this cooler crap itselfs. also, my Zalman CNPS7000, which is rated 95W, handle FX8320 at 125W TDP stock at ease, keeping it roughly at 65c on full load with boost engaged with push-pull negative pressure in case and roughly 22c ambient. pretty loose adaptation of 180W TDP, 125W TDP and 95W TDP. hard to really fit in, if you noise-optimise, you can drop that TDP on cooler significantly, if it require high FAN RPM to achieve it.
reply
Brett
My 2 cents Total electrical power can never equal thermal output. that would mean the only work the cpu is doing is creating heat, i, e a resistor. Energy is being used to switch gates in the transistors which is mostly what a cpu is made up of. The thermal energy that is generated is actually waste heat that is not used in doing work inside the cpu. If you subtract thermal waste heat from electrical input will actually give you a great idea of the efficiency of the cpu. On a side note higher switch speeds (core clock) generally decrease efficiency. Slower cpu speeds are always more efficient.
reply
My 2 cents Total electrical power can never equal thermal output. that would mean the only work the cpu is doing is creating heat, i, e a resistor. Energy is being used to switch gates in the transistors which is mostly what a cpu is made up of. The thermal energy that is generated is actually waste heat that is not used in doing work inside the cpu. If you subtract thermal waste heat from electrical input will actually give you a great idea of the efficiency of the cpu. On a side note higher switch speeds (core clock) generally decrease efficiency. Slower cpu speeds are always more efficient.
reply
Larry
The higher core count CPUs, lets say 3900x vs 3950x, _usually_ the higher core count has _lower_ max clocks. This may not be true, though, and if the 3950x had an equal or even a higher max clock speed, it will likely throttle faster (sooner. In the end, it's not really important enough to base anything off of. Read reviews, see what coolers perform well / not well with a particular CPU, understand if you overclock, these numbers are out the window, anyway. just like the heat it will produce. Thanks to Steve for diving deep into this with little reward at the end. I'm glad he does this stuff.
reply
The higher core count CPUs, lets say 3900x vs 3950x, _usually_ the higher core count has _lower_ max clocks. This may not be true, though, and if the 3950x had an equal or even a higher max clock speed, it will likely throttle faster (sooner. In the end, it's not really important enough to base anything off of. Read reviews, see what coolers perform well / not well with a particular CPU, understand if you overclock, these numbers are out the window, anyway. just like the heat it will produce. Thanks to Steve for diving deep into this with little reward at the end. I'm glad he does this stuff.
reply
Add a review, comment
Other channel videos
