How hot do CPUs REALLY get? - It's SCIENCE time!

[Emily Young]

We have two 8-core CPUs: One at 105W, the other at 95W – But those numbers are completely arbitrary. Let’s test them and see for ourselves which will heat up faster!

 

 

Buy a Core i9-9900K:
On Amazon: https://geni.us/A7fMnMq
On Newegg: https://lmg.gg/8KVmd

 

Buy a Ryzen 7 3800X:
On Amazon: https://geni.us/c8jRoGK
On Newegg: https://lmg.gg/8KVmy

[Lurick]

Stop your science!

In the name of the dumb!

[5x5]

Intel's TDP rating is bullshit. In other news, water is wet

 

EDIT: Worth noting - the 9900K wasn't even running at it's max clockspeed (I saw 4.2 which is low for a 9900K and the later performance delta confirms this) - so the difference can grow even more when both CPUs are actually going full tilt. In reality, the 9900K's TDP should be 125W or 140W

[Windows7ge]

Just what I needed to hear that we can't trust TDP as a means of measuring what volume of cooler a given chip requires or comparison between two chips.

[poochyena]

literally what even was this video.

 

If they wanted to measure heat and tdp, why not just measure the wattage being taken from the wall?

[5x5]

1 minute ago, poochyena said:

literally what even was this video.

 

If they wanted to measure heat and tdp, why not just measure the wattage being taken from the wall?

cause then they also catch RAM, board, cooler, drive and fans as well. Those aren't exactly part of the CPU TDP. It's too many variables and doesn't translate to heat almost at all

 

EDIT: PSU efficiency would also be part of the variables since it changes slightly depending on load - you end up with a massive number of messy metrics and data and would have to do a ton of math to get even close to a reading.

[Fraser Jones]

If ya'll ever need a heat transfer engineer shoot me an e-mail. I had a lot of fun watching this video. Our lab was saying that heat transfer engineers always tends to be the garbage men of engineering. Coming in and cleaning up other people's overheated mess. Fun video!

 

Fraser

[poochyena]

1 minute ago, 5x5 said:

cause then they also catch RAM, board, cooler, drive and fans as well.

That baseline would be the same for each CPU though, so they could at least test to see which cpu draws more power.

[5x5]

Just now, poochyena said:

That baseline would be the same for each CPU though, so they could at least test to see which cpu draws more power.

Not necessarily - you can't account for spikes or CPU efficiency that way.

[GoldenLag]

3 minutes ago, 5x5 said:

cause then they also catch RAM, board, cooler, drive and fans as well. Those aren't exactly part of the CPU TDP. It's too many variables and doesn't translate to heat almost at all

you could also just measure amperage at the PSU cables. then grab a top of the line Z390 hero board and a low-mid range x570 board and start measuring. 

 

since they share the same VRM, you can essentially ignore VRM losses as its comperative and not absolute. 

[5x5]

2 minutes ago, GoldenLag said:

you could also just measure amperage at the PSU cables. then grab a top of the line Z390 hero board and a low-mid range x570 board and start measuring. 

 

since they share the same VRM, you can essentially ignore VRM losses as its comperative and not absolute. 

That would actually work.

[Nystemy]

And this is why a lot of organizations looking at power efficiency look at actual performance of an application and total system power draw. Giving a figure of performance per watt.
Though, if one cares that much, then one likely is running a datacenter, or super computer, knows one's own application like the palm of one's hand, and has the money to set up a few test systems for some back of the envelope calculations. Not to mention that there is also other more important factors to overall power efficiency of a larger system.

Though, for the average PC buyer, it would be nice to have some standard for CPU power consumption.

A couple of simple ways would either be to:

• Measure power supply power. (AMPs and current voltage before VRM. Though, VRM losses is a thing, but this can be calculated relatively easily. (since it can be calculated, and most efficiency numbers are already in the datasheet for the VRM components as is.))
• Measuring heat flow between the CPU and cooler. (heat flow meters are though adding thermal resistance, so it will make the CPU throttle earlier.)
• Have a standardized cooling setup with a know delta C per watt. (Alex should be able to build one relatively simply. I would go for a water cooling loop, but with known airflow through the radiator, and fixed ambient temperature/humidity. Then apply the cooling setup to different CPUs/motherboards. Do note, the CPU block, water flow rate, radiators and airspeed needs to remain constant. (but measuring and regulating airspeed, flow rate and ambient is relatively simple. The last is just going to need a good air conditioning system. After all, 0.5 C variations in ambient temperature isn't going to be all that big of a difference in the end.))

[samcool55]

43 minutes ago, GoldenLag said:

you could also just measure amperage at the PSU cables. then grab a top of the line Z390 hero board and a low-mid range x570 board and start measuring. 

 

since they share the same VRM, you can essentially ignore VRM losses as its comperative and not absolute. 

Not only will that allow them to measure power consumption in a similar workload. It also allows you to measure power usage over a timeperiod.

Which means that they can figure out which cpu is actually more efficient.

 

Let's say you have cpu A and cpu B. Cpu A's power consumption is 100 watts, CPU B's power consumption is 120 watts.

People would say "hey go for A, that's more efficient". But if cpu B runs the benchmark in let's say 5 minutes and cpu A, while having a lower powerdraw, takes 10 minutes. Suddenly those 20 extra watts are actually not that bad.

 

In the video the AMD one was slightly ahead power consumption wise and while they didn't plot it out, the numbers they mentioned tells us that the AMD cpu did also do a LOT more actual work compared to the Intel equivalent. But then again we don't really have the exact times it took so it's meaningless because if the AMD took twice as long, it's suddenly looking quite bad. (probably not, but we don't have exact data to work with which is the issue here)

[GoldenLag]

2 minutes ago, samcool55 said:

Not only will that allow them to measure power consumption in a similar workload. It also allows you to measure power usage over a timeperiod.

Which means that they can figure out which cpu is actually more efficient.

 

Let's say you have cpu A and cpu B. Cpu A's power consumption is 100 watts, CPU B's power consumption is 120 watts.

People would say "hey go for A, that's more efficient". But if cpu B runs the benchmark in let's say 5 minutes and cpu A, while having a lower powerdraw, takes 10 minutes. Suddenly those 20 extra watts are actually not that bad.

 

In the video the AMD one was slightly ahead power consumption wise and while they didn't plot it out, the numbers they mentioned tells us that the AMD cpu did also do a LOT more actual work compared to the Intel equivalent. But then again we don't really have the exact times it took so it's meaningless because if the AMD took twice as long, it's suddenly looking quite bad. (probably not, but we don't have exact data to work with which is the issue here)

well they would need to do one thing to remove variables and that is controll temps of the VRM. as a hotter VRM generates more heat than a cold one. but since its the same VRM the difference should be so small that it shouldnt matter

[Devryd]

How much water was in the loop after all?
I am trying to calculate the actual wattage 

[Emily Young]

6 minutes ago, samcool55 said:

But then again we don't really have the exact times it took so it's meaningless because if the AMD took twice as long, it's suddenly looking quite bad. (probably not, but we don't have exact data to work with which is the issue here)

We cut the test at around 19 minutes or so into each run (when the first Intel test started throttling hard and the water appropriately enough hit 69°C). By that time, AMD was done and its water temperature was lower (it never got a chance to significantly turn the dye black like the Intel runs if you notice).

[agent_x007]

I think everyone should ALWAYS lock Vcore and other voltages to one/pre determined value, BEFORE doing anything that requires measuring Amps/Temps between different stuff.

[samcool55]

5 minutes ago, GoldenLag said:

well they would need to do one thing to remove variables and that is controll temps of the VRM. as a hotter VRM generates more heat than a cold one. but since its the same VRM the difference should be so small that it shouldnt matter

Well as long as ambient, airflow and the VRM components are the same, it should be fine?

Technically the cooler should also be the same but that's really tricky.

 

One CPU will draw more power than the other one and will load the VRM more, run less efficient and draw more power. But I would say that's fine because VRM power draw/heat is linked to the CPU.

 

You can compensate a higher power-drawing CPU with a better VRM, but then it's no longer fair.

[GoldenLag]

1 minute ago, samcool55 said:

Well as long as ambient, airflow and the VRM components are the same, it should be fine?

different VRM heatsink. you could run the bare or just ignore it. its very small losses as the VRMs are very good to begin with. also another issue i realize. because Asus is dumb with how phasecounts work. the VRM may not run all the phases and as such a lot of powerstages may be inactive. tho it shouldnt be different unless the 9900k has a significantly higher powerdraw than the r7 3800x. 

 

3 minutes ago, samcool55 said:

Technically the cooler should also be the same but that's really tricky.

thats just CPU powerconsumption during normal operation. any high end cooler should be fine as that is normal load. 

[Streetguru]

1 hour ago, GabenJr said:

 

Why didn't you guys just go with a battery drain test? Get the battery to 90%, let it idle until it's at 80%, then start the render until the battery is down to 20% or something, and measure how long the PC was able to run at 100% on that battery. With like a GT 1030 or something ultra low power for the GPU, and using the 9900KF

Also don't forget that intel's solder isn't as good as AMD's solder at transferring the heat away.

Just need maybe 3 runs on each PC, with each battery swapped between runs for a total of 6 runs for each PC.

[samcool55]

4 minutes ago, GoldenLag said:

different VRM heatsink. you could run the bare or just ignore it. its very small losses as the VRMs are very good to begin with. also another issue i realize. because Asus is dumb with how phasecounts work. the VRM may not run all the phases and as such a lot of powerstages may be inactive. tho it shouldnt be different unless the 9900k has a significantly higher powerdraw than the r7 3800x. 

 

thats just CPU powerconsumption during normal operation. any high end cooler should be fine as that is normal load. 

I was thinking about a VRM cooler, like a heatsink, probably should have made that a bit more clear.

CPU cooler is indeed less important. Anything that keeps temps low enough to let the CPU run at full boost is good enough. I would go for a watercooler btw because than you can control the airflow better so you are sure the VRM cooling is the same.

 

About VRM power consumption, there is more to it than phasecount. Which components they use is also very important. It's really complex when you dive in it but not every phase is the same.

So if you want to have the same VRM, almost every single component that makes the VRM has to be the same (mainly high-side, low-side and phase count are important)

[Tetanus]

Fun video!

 

Just out off curiosity, why didn’t you just weigh the water instead of using the measuring cup? Seems like it would be much easier and more precise. 1 liter is one kilogram, so 250 ml would be 250 grams. 350 ml would be 350 grams etc. 

 

I have to admit that using the metric system seems far easier than imperial, but then again I live in Europe so that’s what I am used to. I have no idea how much a cup is, but you seem to have a good grasp on the conversion.  

[samcool55]

8 minutes ago, Streetguru said:

Why didn't you guys just go with a battery drain test? Get the battery to 90%, let it idle until it's at 80%, then start the render until the battery is down to 20% or something, and measure how long the PC was able to run at 100% on that battery.

Just need maybe 3 runs on each PC, with each battery swapped between runs for a total of 6 runs for each PC.

Well that would give false results because the motherboards aren't identical.

Different chipsets consume a different amount of power. Also if one board has an extra USB controller or an extra SATA controller or whatever, it all adds up and causes your results to be incorrect because there are too many variables.

[GoldenLag]

Just now, samcool55 said:

About VRM power consumption, there is more to it than phasecount. Which components they use is also very important. It's really complex when you dive in it but not every phase is the same.

asus x570 at low and mid end use the exact same VRM as the Z390 maximus hero. so thats not an issue. (i mean ctrl+c and ctrl+v) the issue is phases being turned on or if they are used at all due higher or lower powerconsumption. 

 

2 minutes ago, samcool55 said:

I was thinking about a VRM cooler, like a heatsink, probably should have made that a bit more clear.

issue then is layout. but in the end it really shouldnt matter has its running stock and neither will get very hot. neither board use actual heatsinks. they just use aluminium slabs. 

[Streetguru]

2 minutes ago, samcool55 said:

 

Just use a close enough model from the same vendor, the VRM components are the same between the platforms.