So how is 95W less than 84W exactly?

Toggle sidebar Toggle sidebar
L

lyssword

Diamond Member
Dec 15, 2005
5,630
25
91
TDP is not always real world usage, Skylake is actually cooler/uses less power at same mhz than devil's canyon in most of reviews I've read.
 
Dufus

Dufus

Senior member
Sep 20, 2010
675
119
101
TDP is for a thermal solution at default specs and base frequency. The thermal resistance from junction to case is fixed and dependent on manufacturing which isn't a perfect process and as such based on a worse case scenario.

If a device with a TDP of 88W has a junction to case thermal resistance of 0.49°C/W then at 88W the temperature differential from cores to case (IHS) at 88W will be 88 x 0.49 = 43°C. A device with a TDP of 91W that has a junction to case thermal resistance of 0.395°C/W will be 91 x 0.395 = 36°C which is lower. These temperatures are additive to local ambient and 3rd party cooling temperatures of which the 3rd party cooling also makes a difference.

Note that Skylake's 91W TDP CPU is based on a 130W cooling solution while the 65W TDP Skylake CPU with a 65W cooling solution is actually rated at 47W TDP.

What SKU were you interested in? The Intel ark currently has destop 64.. 65.. 66.. and 67.. SKU's listed.
 
know of fence

know of fence

Senior member
May 28, 2009
555
2
71
Dufus said:
TDP is for a thermal solution at default specs and base frequency. The thermal resistance from junction to case is fixed and dependent on manufacturing which isn't a perfect process and as such based on a worse case scenario.

If a device with a TDP of 88W has a junction to case thermal resistance of 0.49°C/W then at 88W the temperature differential from cores to case (IHS) at 88W will be 88 x 0.49 = 43°C. A device with a TDP of 91W that has a junction to case thermal resistance of 0.395°C/W will be 91 x 0.395 = 36°C which is lower. These temperatures are additive to local ambient and 3rd party cooling temperatures of which the 3rd party cooling also makes a difference.

Note that Skylake's 91W TDP CPU is based on a 130W cooling solution while the 65W TDP Skylake CPU with a 65W cooling solution is actually rated at 47W TDP.

What SKU were you interested in? The Intel ark currently has destop 64.. 65.. 66.. and 67.. SKU's listed.
Click to expand...

In general terms, a better heatsink should result in a lower temperature (Tcase=T Heatspreader/IHS ) even if more heat has to be dissipated. That doesn't say anything about core temperatures, those may still are getting higher every generation.

Found the value for 4790K's PCG 2013D Heatsink (or Thermal Testing Vehicle as Intel calls it) 0.381 °C/W (page 77, Data Sheet)

I couldn't confirm if 0.395 Degree per Watt is accurate for Skylake, though it is a higher value, which is worse it terms of conductivity.
Tcase max for 91W Skylake with the PCG 2015D Heatsink
should be 63.72 °C. (page 96, Data Sheet)
 
Last edited:
Dufus

Dufus

Senior member
Sep 20, 2010
675
119
101
know of fence said:
I couldn't confirm if 0.395 Degree per Watt is accurate for Skylake, though it is a higher value, which is worse it terms of conductivity.
Tcase max for 91W Skylake with the PCG 2015D Heatsink
should be 63.72 °C. (page 96, Data Sheet)
Click to expand...

We're talking about 2 different regions here. You are talking about case to ambient (ΨCA) TIM and cooler, while I'm talking about junction to case (ΨJC) CPU.

For the 4790K @ 84W and 0.381°C/W, local ambient @ 40°C then
84 x 0.381 + 40 = 72°C. Case temperature (Tcase).

From the same table for maximum Tcase with a local ambient @ 45°C we need better cooling, 0.33°C/W (ΨCA).
84 x 0.33 + 45 = 72.72°C. Tcase

Now this is typically calculated at DTS = -1 which means as the temperature target (Tjmax) is 100°C then core temperature is at 99°C. Any higher and thermal throttling should take place unless it has been disabled or an offset to the temperature target has been applied.

So if at 84W core temperature is 99°C and Tcase is 72.72°C then
99 - 72.72 = 26.28°C. Temperature junction to case (Tjc)
26.28 / 84 = 0.313°C/W. Junction to case thermal resistance (ΨJC)

Add the two together ΨCA + ΨJC = ΨJA
0.33 + 0.313 = 0.643°C/W. Junction to ambient thermal resistance (ΨJA)
So at 84W and local ambient of 45° we get a junction (core) temperature of 0.643 x 84 + 45 = 99°C

Now you can make ΨCA better by using a good TIM and a better cooler but we're stuck with ΨJC, barring removing the IHS.

IOW a CPU with a poor ΨJC but good cooler can run hotter using less power than a CPU with a good ΨJC.
 
Last edited:
dave_the_nerd

dave_the_nerd

Lifer
Feb 25, 2011
16,992
1,621
126
So:

1) TDP doesn't correlate to either real world or maximum power use.
2) TDP doesn't correlate to temperature in any meaningful way.

We should sticky that.
 
know of fence

know of fence

Senior member
May 28, 2009
555
2
71
Oh, I get it. If Intel is measuring a strange Tcase/IHS that is between heatsink and CPU. We can have two different thermal conductivietes, of which one would be extremely revealing. Both also combine to the total thermal resistance.

\\\\ Heatsink or TTV ////
[Thermal Resistance ΨCA]
______IHS / Tcase_____
[Thermal resistance ΨCA]
____Junction / CPU_____

So I dug up the really boring one, the thermal resistance of the crappy Intel cooler or test vehicle. Secretly hoping that perhaps it would describe total thermal resistance. Without the Intel Thermal Test Vehicle (inluding TIM) we cannot even test total resistance or determine ΨCA.
That said Intel seems to be referencing a much better heatsink with low thermal resistance 0.22 °C/W for Skylake. http://ark.intel.com/compare/80807,88195
Which you originally stated and it is now confirmed.

Dufus said:

We're talking about 2 different regions here. You are talking about case to ambient (ΨCA) TIM and cooler, while I'm talking about junction to case (ΨJC) CPU.

For the 4790K @ 84W and 0.381°C/W, local ambient @ 40°C then
84 x 0.381 + 40 = 72°C. Case temperature (Tcase).

From the same table for maximum Tcase with a local ambient @ 45°C we need better cooling, 0.33°C/W (ΨCA).
84 x 0.33 + 45 = 72.72°C. Tcase

Now this is typically calculated at DTS = -1 which means as the temperature target (Tjmax) is 100°C then core temperature is at 99°C. Any higher and thermal throttling should take place unless it has been disabled or an offset to the temperature target has been applied.

So if at 84W core temperature is 99°C and Tcase is 72.72°C then
99 - 72.72 = 26.28°C. Temperature junction to case (Tjc)
26.28 / 84 = 0.313°C/W. Junction to case thermal resistance (ΨJC)

Add the two together ΨCA + ΨJC = ΨJA
0.33 + 0.313 = 0.643°C/W. Junction to ambient thermal resistance (ΨJA)
So at 84W and local ambient of 45° we get a junction (core) temperature of 0.711 x 84 + 45 = 99°C

Now you can make ΨCA better by using a good TIM and a better cooler but we're stuck with ΨJC, barring removing the IHS.

IOW a CPU with a poor ΨJC but good cooler can run hotter using less power than a CPU with a good ΨJC.
Click to expand...

Thanks for walking through these examples, at least we can say with updated values (for TCase and TDP from ark.intel) that for the 4790K the thermal resistance for Junction to Case should probably be lower than:
(&#936;JC) < 99 - 74.04 / 88 = 0.2836 C/W

dave_the_nerd said:
So:
1) TDP doesn't correlate to either real world or maximum power use.
2) TDP doesn't correlate to temperature in any meaningful way.
Click to expand...
1) Who cares about peak maximum power use? Extreme OCers and motherboard makers perhaps. TDP as maximum sustained power use is meaningful. Well, PSU's need to be bigger than TDP, so this is a "caveat" at best.
2) TDP relates to temperatures in a simple linear fashion.

Temperature [°C] = Therm.Resist.[°C/W] * TDP [W] + Ambient [°C]
 
Q

Qwertilot

Golden Member
Nov 28, 2013
1,604
257
126
To answer the original question a bit more thoroughly - they've clearly optimised the chip for low(ish) power operation.

There's an expectation of a very highly clocked K quad core though, and with no massive IPC speed ups, they 'had' to drive the i7k to the same clock speeds as devils canyon to make it comparable/faster than to the previous gen of K chips.

Something about the design/process doesn't seem brilliantly suited to that to be honest. Compare to Broadwell K's 65w TPD/more moderate clocks, the 65w locked Skylake chips.

In terms of 90w power use with Skylake, think you'd definitely do better with a slightly lower clocked 6 core model. Not what the mainstream line is about of course.
 
F

Fun Guy

Golden Member
Oct 25, 1999
1,210
5
81
Qwertilot said:
To answer the original question a bit more thoroughly - they've clearly optimised the chip for low(ish) power operation.

There's an expectation of a very highly clocked K quad core though, and with no massive IPC speed ups, they 'had' to drive the i7k to the same clock speeds as devils canyon to make it comparable/faster than to the previous gen of K chips.

Something about the design/process doesn't seem brilliantly suited to that to be honest. Compare to Broadwell K's 65w TPD/more moderate clocks, the 65w locked Skylake chips.

In terms of 90w power use with Skylake, think you'd definitely do better with a slightly lower clocked 6 core model. Not what the mainstream line is about of course.
Click to expand...
Not a huge advantage for 14nm over 22nm, power-wise anyway...
 
F

Fun Guy

Golden Member
Oct 25, 1999
1,210
5
81
ShintaiDK said:
Thats not true. Lowest Core based CPU is 11.5W on 22nm. 4.5W on 14nm.

And substantial performance improvements across the mobile lines.
Click to expand...
Will enjoy seeing how they in fact perform - not just in theory, but in practice.
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom