Extrapolating heat from TDP, base clock, and binning?

[Machinus]

Intel provides estimates of "TDP" which is supposed to correspond to the "base clock" of a chip, even though that frequency is not actually used by the system (???). Intel also sells chips in suites, with different frequencies and core counts, so you could make some guesses about how the process and architecture will behave from them. In the same family, I assume that they still bin their own chips and sell the better ones at higher clocks.

So, for example, the 8700K is specified 500MHz higher than the 8700, and it draws 30W more. So you would guess that the an average 8700 would draw at least 30W extra power if you overclocked it by 500MHz. (Note that this is a 46% increase in heat for a 16% increase in frequency.)

But Intel's specifications do not lead to a consistent picture of power consumption. If you just look at the official specifications of 14nm chips, power per core ranges from 8W to 28W. Trying to estimate overclocking headroom looks impossible because the factory specifications don't suggest a pattern for power consumption. Now that we're getting 8 (or 20...) cores per socket, knowing your real-life thermal flux is necessary if you want your chip to live. The 8700K is specced at 16W per core at "base clocks," but really runs 600MHz higher out of the box on all four cores. If we compare to the 8700 power difference, that suggests that the real life stock heat output reaches 150W, which is obviously a problem if you built a system based on the 95W rating. A lot of good coolers can handle that no problem, but remember that's just the factory setting, which is really 55W higher than claimed. Without even considering overclocking, the same Turbo vs. "base clock" problem means that, if Intel sells an 8-core mainstream chip with an unlocked multiplier in Coffee Lake Refresh, realistic extrapolation suggests you will get out of the box power spikes of 200W. The problem is compounded for i9s: if the 10+ core parts add 5-10W of power in out of the box use you are adding 50-200W of additional heat which is not getting through that polymer very well. This might be part of why Intel no longer claims per-core Turbo settings.

Actually overclocking any of these chips will then lead to unsustainable heat levels in these chips, no matter what fluid is flowing past the lid. It seems very important to accurately test the actual heat given off by the 14++ chips. Is there anyone doing tests of this type to find out how hot they really are?

[Charles Wirth]

Correct way is to use a DC camp on amp meter to the CPU power connectors near the CPU socket.

Other method is XTU, it shows the package TDP real time.

I can set my 8700K BIOS to stock, I have a super crappy Intel celeron heat sink, the real small one that is quick and easy to install.

I ran XTU and my chip is 60c ~ 70c, the package TDP is exceeding 120W. When idle it sits around 9W package TDP with BIOS pure stock.

If I set all cores to 47x the chip moves to 70c ~ 80c, the package TDP exceeds 150W (screen shot) and max temp of 78c

If I turn off turbo and the chip maxed to 95W package TDP, max temp 63c. On the XTU graph I can see the chip regulating time on core and core count active to stay under the package TDP with turbo disabled.

This chip is not a delid, package TDP values will be similar on AIO except for the 47x setting, I think the chip will allow higher package TDP with the extra cooling, there is not much worse than the sink I am using now other than no fan or no sink at all.

This confusion may be part of why Intel wants to be transparent in stating correct TDP values, Intel can accurately state this number with confidence and no confusion to how this value is determined.

http://www.esaitech.com/intel-d75716...CABEgL1zfD_BwE