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

[GoldenLag]

3 minutes ago, adammmmmm said:

This is incorrect. 100% of electricity that gets put into a computer gets turned into heat.

not all the heat gets output at the CPU. some energy is used for signal transfer over PCIe. 

 

some small electrical fields should be present, but thats a very small factor. 

[adammmmmm]

56 minutes ago, GoldenLag said:

not all the heat gets output at the CPU. some energy is used for signal transfer over PCIe. 

 

some small electrical fields should be present, but thats a very small factor. 

It's pedantic but electricity used for "signals" still eventually turns into heat. I was referring to a computer system as a whole. A computer is a 100% efficient space heater that happens to let me play games on it before heating my room.

[GoldenLag]

Just now, adammmmmm said:

It's pedantic but electricity used for "signals" still eventually turns into heat. I was referring to a computer system as a whole. A computer is a 100% efficient space heater that happens to let me play games on it before heating my room.

you know we are reffering to the CPU right?

 

like since the beginning we have speculated ways to remove variables. 

 

 

also 100% space heater isnt that special. ever since the beginning of phase change cooling 100% is kinda wasteful in terms of energy. at least compared to phase change cooling/heating. 

[adammmmmm]

2 hours ago, GoldenLag said:

you know we are reffering to the CPU right?

 

like since the beginning we have speculated ways to remove variables. 

I'm following now. The way it was stated earlier about how power "disappears" was just misleading.

 

I think measuring power input to the CPU directly would just be too difficult.

 

Anthony and Linus, I believe, were on the right path by using water and it's measured temperature change but approached it in a very unsciencey way.

 

Measuring delta T across the water block with thermocouple immediately before and after the CPU block is probably the easiest way as long as they can measure and log mass flow rate. Then it's just a matter of plugging their logged data into Q = M x C x Delta T

[GoldenLag]

23 minutes ago, adammmmmm said:

I think measuring power input to the CPU directly would just be too difficult.

why? you just measure EPS input amperage.

 

and since its a comperative measurement. all you need is the same VRM. 

[adammmmmm]

1 hour ago, GoldenLag said:

why? you just measure EPS input amperage.

Because as you said, we want CPU only and measuring the EPS drags the VRMs with it. Even with the same VRM, there is no way to easily subtract out the losses through those modules.

[GoldenLag]

16 minutes ago, adammmmmm said:

Because as you said, we want CPU only and measuring the EPS drags the VRMs with it. Even with the same VRM, there is no way to easily subtract out the losses through those modules.

Except there is a way to subtract those losses. 

 

There is documentation on the VRM modules which makes you able to calaculate the losses.

 

Even so its a comperative benchmark. As such there isnt a big reason to include it.

[adammmmmm]

54 minutes ago, GoldenLag said:

There is documentation on the VRM modules which makes you able to calaculate the losses.

That's pretty cool. I didn't know that kind of thing was so well documented. Thanks.

[alextulu]

Are you sure you want Ring to be a sponsor in this video?

 

https://www.washingtonpost.com/technology/2019/08/28/doorbell-camera-firm-ring-has-partnered-with-police-forces-extending-surveillance-reach/

Quote

Doorbell-camera firm Ring has partnered with 400 police forces, extending surveillance concerns

https://gizmodo.com/ring-gave-police-stats-about-users-who-said-no-to-law-e-1837713840

Quote

Ring Gave Police Stats About Users Who Said ‘No’ to Law Enforcement Requests

 

[PlayerX]

I have to assent to what adammmmmm said:

The heat flow is proportional to the Delta_T and therefore it's convinient to measure somehow that (for example on a waterblock). I wrote down a "short" example of the logic and math behind all this, so you maybe understand how this should work:

 

Hope this is not too complicated and clears some questions on how accurate this method is.

On the question, why measuring the current that goes to the cpu doesn't give you the heat output: this will give you the approximate heat output of the cpu AND everything thats in that circuit after the CPU. Approximate because you have to also account loss in energy due to creation of electromagnetic fields and therma radiation, both negligible effects (as far as I'm aware).

[Thorium90]

Just a thought, instead of measuring the time it takes the water to heat from a starting temperature to an ending temperature, could you instead measure the heat dissipated by a radiator. You could set up a regular water cooling loop (pump, fans, and radiator) and measure the temperature of the water in and out of the radiator. Assuming that the CPU and cooling loop are at equilibrium and the loop is large enough not to bottleneck the heat transfer, wouldn’t the heat being kicked off by the CPU be the sum of the heat in the loop plus the heat kicked off by the radiator. I’m still debating with myself what equations would need to be used. I kind of think it would involve using integrals to find the area under the curve of the temperature of both the loop and regulator. Any thoughts?

[adammmmmm]

On 9/28/2019 at 10:36 PM, Thorium90 said:

Just a thought, instead of measuring the time it takes the water to heat from a starting temperature to an ending temperature, could you instead measure the heat dissipated by a radiator. You could set up a regular water cooling loop (pump, fans, and radiator) and measure the temperature of the water in and out of the radiator. Assuming that the CPU and cooling loop are at equilibrium and the loop is large enough not to bottleneck the heat transfer, wouldn’t the heat being kicked off by the CPU be the sum of the heat in the loop plus the heat kicked off by the radiator. I’m still debating with myself what equations would need to be used. I kind of think it would involve using integrals to find the area under the curve of the temperature of both the loop and regulator. Any thoughts?

This is exactly the same thought as I had above which was backed up with the better example and explanation by PlayerX above . All you need is the temperature in and out of the radiator and then the flow rate through the radiator and then those values can be plugged into the equation presented above by PlayerX. The only issue with this approach is that the temperature probes will be far away from the CPU block so some heat would be dissipated by the piping in the water loop before reaching the radiator which is why my first thought was to approach this by placing the temperature probes close to the entrance and exit of the CPU block itself.

[NICHIRENSHU]

I just posted on youtube, but I will post here too, hoping to be heard

 

Quote

To make the measurement more precise, use big amount of water. Using bigger water mass will reduce the influence of other factors.

Just get 5 litters water bottle insulate it and connect it to the pump. Measure how long it will need to rise the temperature with 10ºC. The one that takes longer is more efficient - thats it.

 

Bonus - to calculate the watts divide 209080 by the seconds it need to rise the temperature, lets say 30 minutes: 209080/1800 = 116 W. To be more precise insulate the pipes, the pump and the pump motor too.

 

Measuring bigger amount of something gives better accuracy. For example when I want to weight small things that weight 0.5 g that will hardly be detected by the scale at all. I put 10 of them on the scale I see that the result is 5 g and I just divide it. 

It is similar with the bigger amount of water, the thermal mass of the other components become negligible, the thermal losses to the surrounding  become less in % comparing to the smaller amount of water. So this gives better results in therms of thermal accuracy.

Also making the test with water temperature from 20 ºC to 30 ºC will allow the CPU to work at constant clock, This will remove the load fluctuations.

 

So everything will be more accurate and you can even measure the CPU power in watts "divide 20908 by the seconds".

(20908 is for 5 litters, for 1 litter is 41816) 

@GabenJr