Can someone explain TDP to me? Went from Dual to Quad and temps dropped

[reallyscrued]

Scenario:

I had a Phenom II X2 550 BE (Callisto, or Deneb with two defective cores) clocked at 3.7 Ghz being fed 1.45 volts. Says 80 Watt TDP.

Small FFT test - Maxed out 53 Degree Celsius on stock hs/f

Upgraded to a Phenom II X4 B93 (basically an 830 like I have in my sig) clocked at 3.2 Ghz at 1.4 volts. Says 95 Watt TDP.

Small FFT test - Maxed out 44 degrees - same hs/f


I'm a little confused. The Quad had a higher TDP at stock speed (I have it at a mild 400 mhz OC right now) and it has 2 additional cores, why did the temperatures drop?

 

[Puppies04]

The quad core has twice the surface area to dissipate heat (even though the chips look the same size the vast majority of the heat is produced by the cores themselves which are located in the middle of the chip). I know the cover over the cores is called an "integrated heat spreader" but they really don't spread heat laterally very well.

 

[2is]

You also have your dual core clocked higher and with more volts. bring it down to 3.2 and whatever voltage you need for stability at 3.2 which would be less than 1.4 i'm sure and you'll see different results.

 

[pantsaregood]

Those temperatures you have don't have much to do with the TDP of the CPU, in all honesty.

The 95W Phenom II X4 will dissipate 15W more heat than the 80W Phenom II X2 at stock voltages. The X2 cores could be running at temperatures of 900 C, and the X4 running at temperatures of 20 C - the X2 would still dissapate 15W less heat.

Temperature is a measure of heat concentration.

 

[KingFatty]

Can you re-run the test, except set both chips to the same voltage and frequencies?

I think your results just reflect the difference in voltage and frequency. I mean, you'd probably see the same difference on one chip if you just changed its settings. You'd probably see the same difference regardless of which chip you used.

 

[Ferzerp]

As pants said, operating temperature does not equate to TDP.

Not at all.

Nor does it have more than a loose correlation to heat produced. There are many other factors that decide your operating temperature.

 

[_Rick_]

Also, TDP is just a sticker number, that has no fixed correlation to actual heat output. (Though in some scenarios it has actually become quite close, i.e. self-throttling ULV CPUs and PowerTune limits on GPU's - but on most desktop CPU's there's at best a mild correlation of TDP and actual thermal output.)

 

[pantsaregood]

In my experience, CPUs generally do stay below their advertised TDP. An i7-3820, for example, cant possibly pull 130W at stock speeds.

Think of it like this - If you drop a 5000 C needle made of carbon into the ocean and do the same with a car that's been sitting in the sun a day, which one releases more energy?

The car, of course. No single point on the car is above 100C, but the car is quite large. The needle is just highly concentrated heat.

 

[Denithor]

And one more point - if you used the same HS/F you obviously re-seated it, probably using an aftermarket thermal paste of some kind. The new thermal paste probably has better transfer properties than the original stock paste.

If heat transfers out faster, temps will be lower.

 

[reallyscrued]

I see.

Okay, if I clocked them exactly the same, it looks as though their temperatures are pretty much equal.

From the responses, I can equate that because even though there's more cores active, there's also more surface area cooling those cores so in effect, it is as though the same amount of heat is being pulled away from the chip?

 

[reallyscrued]

In my experience, CPUs generally do stay below their advertised TDP. An i7-3820, for example, cant possibly pull 130W at stock speeds.

Think of it like this - If you drop a 5000 C needle made of carbon into the ocean and do the same with a car that's been sitting in the sun a day, which one releases more energy?

The car, of course. No single point on the car is above 100C, but the car is quite large. The needle is just highly concentrated heat.

This was my issue with understanding it.

I equate the 4 cores with the car and the 2 core with the needle.

Shouldn't the quad core make more heat than the dual?

And one more point - if you used the same HS/F you obviously re-seated it, probably using an aftermarket thermal paste of some kind. The new thermal paste probably has better transfer properties than the original stock paste.

Same goop along with the same hs/f. I should have clarified.

I don't think my technique of installing the heatsink was the issue, been doing this for a while now, unless this was a freak instance.



I'm temped to put my dual core back in and see why it acts like this.

At the exact same speeds, they display the exact same temps, although one has 100% more working, physical, cores.

 

[Ferzerp]

I see.

Okay, if I clocked them exactly the same, it looks as though their temperatures are pretty much equal.

From the responses, I can equate that because even though there's more cores active, there's also more surface area cooling those cores so in effect, it is as though the same amount of heat is being pulled away from the chip?

Technically, with all else being equal, (and that's a lot of "all else" here), twice the surface area can dissipate twice the heat.

 

[pantsaregood]

TDP and surface area aren't related at all.

Your 95W CPU could be the size of the moon. If it is operating at 95W, then it is producing 95W of heat. If you measured temperature at any given point, it would be incredibly cold.

Your 80W CPU could be the size of a pin head. The temperature on the pin head would be very high, but it doesn't contain much thermal energy - the energy it does contain is just highly concentrated.

What cools a tub of 95 C water better - one ice cube at -100 C, or 15 at -10 C?

Also, the X4 DOES produce more heat than the X2. The X2 is dissipating 80W of heat. The X4 is dissipating 95W. The dual-core produces less heat overall, but the heat it does produce is more highly concentrated.

Don't mistake the lack of correlation between TDP and surface area with that of temperature and surface area. At a given TDP, higher surface area will equate to lower temperatures - lower surface area will raise temperatures.

If you dropped a needle that was 5000C on your floor when your room was ~15C, do you think your room temperature would suddenly shoot up? Of course not, it would be relatively unaffected. If you dropped a mattress-sized block at 5000C into your room, however, your room would quickly heat up. The two objects are the same temperature, but one is emitting far more thermal energy than the other.

 

[2is]

The quad IS producing more heat. 4 cores at 40C is twice as much heat as 2 cores at 40C even though they are not running any hotter. (assuming same size cores)

 

[reallyscrued]

I know the cover over the cores is called an "integrated heat spreader" but they really don't spread heat laterally very well.

According to this thread:

http://www.xtremesystems.org/forums/showthread.php?182095-Naked-Phenom/page2

The heatspreader is soldered on.

And removing it doesn't result in lower temps.


I'd say the heatspreader does its job.

 

[reallyscrued]

TDP and surface area aren't related at all.

I don't think anyone was suggesting that.

I think it was suggesting that surface area and cooling efficiency are related.

The quad IS producing more heat. 4 cores at 40C is twice as much heat as 2 cores at 40C even though they are not running any hotter. (assuming same size cores)

That makes a lot of sense.

And that's where I'm confused.

Why, when there's double the heat in your scenario, is the same heatsink/fan capable of cooling the chip without rise in temperatures?

 

[Ferzerp]

According to this thread:

http://www.xtremesystems.org/forums/showthread.php?182095-Naked-Phenom/page2

The heatspreader is soldered on.

And removing it doesn't result in lower temps.


I'd say the heatspreader does its job.

If it is soldered, yes, leave it on, it is making it easier to cool (unless you can solder your own heatsink or water block directly to the die (hint: don't try this).

If it is paste, it is always a detriment to cooling. You can always get better temps by mounting directly on to the die, but it's easy to eff it up.

 

[lakedude]

Heat is not temperature.

The example I remember is the bathtub full of warm water @ 115 °F vs a red hot needle @1200 °F. Clearly the needle is at a higher temperature than the water, but which would melt a 3 inch cube of ice faster?

The quad is almost certainly using more power and putting off more heat. The dual core will have more trouble expelling heat so it will tend to get hotter quicker.

This is the same reason Sandy Bridge @ 32 nm is easier to OC on air than is the 22 nm Ivy Bridge. Ivy is smaller and uses less power but it has more trouble getting rid of heat.

 

[pantsaregood]

There isn't double the heat. 95W is ~20% more energy than 80W.

Think of it this way: a 95W power source with temperatures at 25 C is going to release more waste heat into your room than an 80W power source with temperatures at 250 C. That means the 95W unit will warm your room up faster than the 80W unit.

 

[reallyscrued]

If it is soldered, yes, leave it on, it is making it easier to cool (unless you can solder your own heatsink or water block directly to the die (hint: don't try this).

If it is paste, it is always a detriment to cooling. You can always get better temps by mounting directly on to the die, but it's easy to eff it up.

Thanks for the hot tip (...pun intended) but I wasn't planning on taking the heatspreader off.

 

[Ferzerp]

I don't think anyone was suggesting that.

I think it was suggesting that surface area and cooling efficiency are related.



That makes a lot of sense.

And that's where I'm confused.

Why, when there's double the heat in your scenario, is the same heatsink/fan capable of cooling the chip without rise in temperatures?

This can easily be the case if the limitation of your current cooling setup is the transfer of heat from the CPU to the cooler.

 

[reallyscrued]

This can easily be the case if the limitation of your current cooling setup is the transfer of heat from the CPU to the cooler.

If the limitation of cooling is the contact area of the heatsink to the heatspreader, (Say it can only dissipate 80 watts of heat per second) then wouldn't the residual 15 watts of heat stew within the CPU and raise temps ultimately causing temperatures to show on-die as being higher?

 

[2is]

There isn't double the heat. 95W is ~20% more energy than 80W.

Think of it this way: a 95W power source with temperatures at 25 C is going to release more waste heat into your room than an 80W power source with temperatures at 250 C. That means the 95W unit will warm your room up faster than the 80W unit.

Sorry I didn't mean to suggest it was actually producing twice as much heat. I used those numbers for simplicity just to illustrate that more heat does not necessarily mean higher temps.

 

[Denithor]

Ok, try this idea on for size:

At stock, the dual core (X2 550, 3.1GHz) is rated to produce up to 80W of waste energy (heat) that has to be dissipated.

At stock, the quad core (X4 B93, 2.8GHz) is rated to produce up to 95W of heat that has to be dissipated.

Now, here's the first kicker - AMD basically assigns this TDP based on the highest member of the family tree and the lesser models typically run somewhat lower. Just because they state a chip runs 80W or 95W doesn't mean it runs that constantly. The TDP is more of a guide to the OEM system builders out there (think Dell, HP, etc) to know the maximum heat they will have to dissipate from a given CPU at stock settings under load.

And the second kicker - you weren't running either of those chips at stock settings. As soon as you start overclocking, especially if you're feeding the chip more volts, power consumption/heat generation goes up quickly.

http://www.tomshardware.com/reviews/phenom-x4-965,2389-10.html

So I'd assume that you just happened to overclock each to a point where the power consumption was equal, the same TIM/HSF meant that equivalent heat was being dissipated at the same rate and therefore they ran at the same temperatures.

 

[KingFatty]

What cools a tub of 95 C water better - one ice cube at -100 C, or 15 at -10 C?

Trick question - ice cubes can cool water, without getting any warmer than 0 C themselves.