These cpus suffer from crappy thermal paste Intel has used in them, they must be delidded.

AberTim wrote on 2025-12-25, 11:16:

It is my "assumption" that the 'k' CPUs have all been damaged by Overclocking.

Yes, that's possible. I was overclocking the 3770K ES and trying to get it stable at 4.6-4.7 GHz, but it seems I overdid the core voltage.
After that, it won't overclock above 4.3 GHz.

Desktop Ivy-Bridge has a regular paste under IHS instead of solder. These CPUs are almost 14 years old.

Thanks guys,

I did get the impression that the temperature of the spreader lid wasn't consistent with the temperature of the core, but also thought the die to spreader connection was of the "hard" glue bond. Now I know different, I've ordered a de-lidder from AliExpress. ("delidders" who'd have guessed?)

Also something I didn't mention about the CPU's, was that all three had patches of discolouration on the underside, which although did buff off with an oily rag, they very much struck me as heat generated and were around groups of 4 pads. My guess is that the motherboard current sense circuitry may well have saved the CPUs from complete destruction. due to thermal runaway. I must have around a 100 CPUs, but have never seen this discolouration before. That said I've not "knowingly" had such an easily over-clocked type either!

The last actual over-clocking I did was a 80386 DX-33 around 1991. I killed it, and it cost me about 2 weeks wages, and I sort of lost interest in it after that.

Pretty sure 386 can't be killed by any common kind of overclocking. There's no voltage adjustments and it's feeding directly from +5V line from PSU.
Modern CPUs? Yeah, those are very easy to kill, they already run on the edge of possible from the factory (cough cough 14900K).

Ivy Bridge is somewhere in the middle, but statistically it's very uncommon to encounter a degraded from overclocking CPU, not a lot of people bother with that, especially on midrange stuff.

AberTim wrote on 2025-12-25, 11:16:

My question is, does anyone "know for sure" if the 'k' versions (when new) did have the same low
power consumption as the plain i5-3570?

I am not prepared to confirm this from personal experience, as I am not involved in ‘high-performance sport’.
However, it is traditionally believed that CPUs with "high leakage currents" (i.e. with higher consumption at idle) achieve more highest frequencies in extreme overclocking.

The Serpent Rider wrote on 2025-12-25, 12:38:

Desktop Ivy-Bridge has a regular paste under IHS instead of solder. These CPUs are almost 14 years old.

It's not the age alone that degrades the thermal compound.

It's the length of time spent at high temperatures* that do it.

* high temperatures = anywhere north of 65-70C for most older common compounds, typically.

If that i5-3570k was used with one of those short Intel stock coolers with push pins (and it probably was, since -k parts generally shipped for the boxed version of the CPU, whereas non-k [OEM] parts were cooled by whatever cooler the OEMs [Dell, HP, etc.] specified... which were generally better than the stock Intel one), chances are it ran that hot under load quite frequently.

AberTim wrote on 2025-12-25, 11:16:

My question is, does anyone "know for sure" if the 'k' versions (when new) did have the same low
power consumption as the plain i5-3570?

Not 100% sure, since the -k versions are usually better-clocked parts too, IIRC.

How are you measuring the average power consumption?

I generally don't trust sensor data from older hardware too much - not because of age, but because it wasn't always calibrated well or too accurate. Voltage readings in BIOS on old motherboard are a prime example. By the 2010's, of course, some of the sensors became much better too, so I wouldn't completely discard your results either. But IMO, the best way to know for sure is to grab a Kill-A-Watt or similar power meter along with both the -k and non-k version of the CPU, and then swap between them in the same system to compare the results - ideally under desktop conditions and on an OS that doesn't have a "mind of its own" like Windows 10 or 11 (i.e. where random updates or whatever other background gags can screw with the results.) So that leaves Windows 7 and older or some such similar flavor of Linux... but obviously that might be a little more work for your here than you might be willing to do, which I would understand if you didn't. Just saying what would be the "ideal" way to measure the power consumption more accurately.

...Pretty sure 386 can't be killed by any common kind of overclocking.

Well I certainly can't be sure, but I think I was running it with a socketed dip14 40Mhz Oscillator and a small fan sitting on the cpu, propped up with 4 plastic motherboard stand offs, when it all stopped working. When the CPU was finally replaced the motherboard & ram (4 x 4Meg sips) were still functioning OK. (Occums razor ?)

...It's the length of time spent at high temperatures* that do it.

My own experience bares out that theory. In the early 2000s running win 2k 16 hours a day with AMD CPUs, I would lose a cpu every 18 months or so, despite a regular cleaning cycle and using all copper coolers with added heat pipes and larger slower fans and always using lower performance motherboard settings when available. It wasn't until I finally moved to XP, (APM) that the CPUs stopped dying.

CPUs with "high leakage currents" (i.e. with higher consumption at idle) achieve more highest frequencies in extreme overclocking.

That does 'sound' very counter intuitive... My own experience driving power Mosfets & Bjts at Khz & Mhz speeds would differ. Saying that, I'm also not involved in ‘high-performance sports’ either 😀

How are you measuring the average power consumption?

Just a basic motherboard setup on the workbench (GA-H61M-S2PV) 1GB ram, no boot devices (except a sometimes USB stick with dos & FDAPM ).
All measurements are relative using an averaging energy monitor.
Test subjects are 4x i5-3570 & 3x i5-3570k (1 is dead) (all purchased at different times)

The non-k versions hardly deviate from each other perhaps 0.5watt in total across all 4 units, but the K versions are quite a different matter.

My attention was immediately drawn when the cooler fan started to increase in speed within a few seconds after power-up and steadily increased. This made me think that the heat wasn't being transferred from the die to the heat spreader.

The upshot so far, is that I'm returning the dead and the hottest running cpu, but keeping the coolest running one and will de-lid it and see what happens. I'm sure I'll report here what happens, particularly as it was this thread that introduced me to the de-lidding process.

AberTim wrote on 2025-12-26, 11:43:

That does 'sound' very counter intuitive... My own experience driving power Mosfets & Bjts at Khz & Mhz speeds would differ. Saying that, I'm also not involved in ‘high-performance sports’ either 😀

Theoretically, the deeper the class B amplifier, the stronger the transistors off and the slower they exit this state.
This is why AB (with quiescent currents) is used.
In fact, these high-performance overclockers have their own terminology and mythology, which makes them difficult to understand.

It is quite possible that “K” processors have less aggressive power saving modes.
Or the supply voltage is higher to make overclocking easier for beginners.

from datasheet Desktop 3rd Generation Intel ®
Core™ Processor Family, Desktop
Notes:
1. Each processor is programmed with a maximum valid voltage identification value (VID), which is set at
manufacturing and cannot be altered. Individual maximum VID values are calibrated during manufacturing
such that two processors at the same frequency may have different settings within the VID range. This
differs from the VID employed by the processor during a power management event (Adaptive Thermal
Monitor, Enhanced Intel SpeedStep Technology, or Low Power States).

I measured consumption with a multimeter, in 10 amp mode (actually 20 on almost all multimeters) using adapter cables to 12V CPU connector.
But since I'm overclocking the hardware ‘to the top of the model range’ without aggressively playing with the power supply voltages, I haven't seen anything interesting.
Everything is the same as in the datasheet for the corresponding frequency model.