one CPU core 15c lower than the rest?

[MaxTehLegend]

Hi, i've been messing around with my overclock and stumbled on something extremely strange. I hadn't noticed it before and i thought that it was maybe time for me to apply some fresh thermal paste on my CPU, So i did. And i made sure i put alot on there and mounted my cooler on pretty tight and im still seeing weird temps.

when stress testing with Prime95 v26.6 Small FFT's, AIDA64 CPU & FPU, intel burn test stadard 3 of my CPU's cores are significantly hotter than my 4th core and i had task manager and other monitoring software to see if ALL my cores were being utilised and they all were at 100%. i'm also 100% positive that this oc is stable.

I'm using an i7 4770K overclocked to 4.5GHz core ratio & ring ratio/cache ratio with my core VID @1.3v (1.320v) a under load and ring voltage @1.35v. my RAM is running at 2400MHz (XMP profile 1 and memory try it CL11 compatibility).

Also, if you're wondering why my temps are so high its because i was using AIDA64 & intelburntest which made my CPU get unrealistically hot. In Prime95 Small FFT's it only hovers around 83c on the package (around 85c hottest core and around 72c coldest core) which i consider perfectly fine as it's a steady workload and perfect stress test for testing thermals. I want to try and fix this without having to delid if possible.
Thanks for reading.

 

[jay32267]

I've seen this happen numerous times and most times....the general consensus was....a bad thermal sensor.

Is it always about 15 degrees off? Under load? Under no load?

If it were a bad sensor, I would expect it to be off around 15 degrees all the time.

 

[mjbn1977]

I rather think that this specific core won the silicone lottery, whereas the other 3 cores came out average. Core 1 and 2 were probably on the edge or close to the edge of the silicone waver and the 2 and 3 one row further away and core 4 was "produced" perfectly.

 

[MaxTehLegend]

I've seen this happen numerous times and most times....the general consensus was....a bad thermal sensor.

Is it always about 15 degrees off? Under load? Under no load?

If it were a bad sensor, I would expect it to be off around 15 degrees all the time.

You can see the screenshot. the 4th core is idling at 38c. same as the 3rd core. the sensor is fine.

 

[mjbn1977]

yes, because as hotter the chip the larger the differences. There is a lot of science and physics behind it. I have a business background, but while I was in college I used to work for a large front-end quality control department of a large semiconductor company and you have no idea how much factors influence the quality of a chip and how much the quality from one chip to another on the same silicone wafer can fluctuate.

 

[CompuTronix]

... I hadn't noticed it before and i thought that it was maybe time for me to apply some fresh thermal paste on my CPU, So i did. And i made sure i put alot on there and mounted my cooler on pretty tight and im still seeing weird temps...

... testing with Prime95 v2.66 Small FFT's, AIDA64 CPU & FPU, intel burn test stadard 3 of my CPU's cores are significantly hotter than my 4th core and i had task manager and other monitoring software to see if ALL my cores were being utilised and they all were at 100%...

... i7 4770K overclocked to 4.5GHz core ratio & ring ratio/cache ratio with my core VID @1.3v (1.320v) a under load and ring voltage @1.35v...

... Prime95 Small FFT's it only hovers around 83c on the package (around 85c hottest core and around 72c coldest core) which i consider perfectly fine as it's a steady workload and perfect stress test for testing thermals. I want to try and fix this without having to delid if possible...

MaxTehLegend,

• What was your ambient temperature?

• What cooler are you using?

Too much TIM can act as an insulator, thereby increasing Core temperatures. However, as the symptom didn't change when you reapplied, TIM is not likely the problem.

Your test methods are correct, but your Core & Ring voltages are on the high side. Maximum recommended Vcore for 22 nanometer processors (3rd & 4th Gen) is 1.300 volts. Reducing Vcore will, of course, result in lower Core temperatures.

Your Thread title says your thermal deviation is 15°C, your text description is ~12°C, and your screenshot shows 10°C. Keep in mind that deviations diverge as load increases, and converge at idle.

Intel’s specification for Digital Thermal Sensor (DTS) accuracy is +/- 5°C, and are typically quite reliable. This means at a steady 100% workload on all Cores, deviations between the highest and lowest Cores can be up to 10°C, so your 4770K is definitely out of spec.

Deviations on processors that have an uneven application of TIM might exceed 10°C by several degrees. Again, since the symptoms didn't change when you re-TIM'd, the problem is elsewhere.

Consider this:

The 4th Gen Haswell 4770K was first launched 2nd Quarter 2013, so yours could be 5+ yrs old. Intel's TIM between the Die and IHS dries out over time and hardens to the consistency of chalk, losing it's thermal bond with the Die. Older 3rd Gen Ivy Bridge (launched 2nd Quarter 2012) as well as newer 4th Gen Devil's Canyon processors (launched 2nd Quarter 2014) with the same problem as yours have been delidded, revealing that thermal performance will progressively degrade over time due to internal TIM failure.

See Section 9 - The TIM Problem - Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

I know you don't want to hear this, but thanks to our friends at Intel, the only way to correct the problem is to delid. If you don't want to do it yourself, then you can send your CPU to Silicon Lottery - https://siliconlottery.com/collections/all/products/delid

They warranty a first class professional job for a fair price. Your Core temperatures will be ~18°C lower at 100% workload, and you'll also be able to reduce Vcore by about 20 millivolts, which will further reduce Core temperature while maintaining your stable 4.5 overclock. If you don't want to delid, then your only other option is to just live with it. Sorry.

CT

 

[MaxTehLegend]

... I hadn't noticed it before and i thought that it was maybe time for me to apply some fresh thermal paste on my CPU, So i did. And i made sure i put alot on there and mounted my cooler on pretty tight and im still seeing weird temps...

... testing with Prime95 v2.66 Small FFT's, AIDA64 CPU & FPU, intel burn test stadard 3 of my CPU's cores are significantly hotter than my 4th core and i had task manager and other monitoring software to see if ALL my cores were being utilised and they all were at 100%...

... i7 4770K overclocked to 4.5GHz core ratio & ring ratio/cache ratio with my core VID @1.3v (1.320v) a under load and ring voltage @1.35v...

... Prime95 Small FFT's it only hovers around 83c on the package (around 85c hottest core and around 72c coldest core) which i consider perfectly fine as it's a steady workload and perfect stress test for testing thermals. I want to try and fix this without having to delid if possible...

MaxTehLegend,

• What was your ambient temperature?

• What cooler are you using?

Too much TIM can act as an insulator, thereby increasing Core temperatures. However, as the symptom didn't change when you reapplied, TIM is not likely the problem.

Your test methods are correct, but your Core & Ring voltages are on the high side. Maximum recommended Vcore for 22 nanometer processors (3rd & 4th Gen) is 1.300 volts. Reducing Vcore will, of course, result in lower Core temperatures.

Your Thread title says your thermal deviation is 15°C, your text description is ~12°C, and your screenshot shows 10°C. Keep in mind that deviations diverge as load increases, and converge at idle.

Intel’s specification for Digital Thermal Sensor (DTS) accuracy is +/- 5°C, and are typically quite reliable. This means at a steady 100% workload on all Cores, deviations between the highest and lowest Cores can be up to 10°C, so your 4770K is definitely out of spec.

Deviations on processors that have an uneven application of TIM might exceed 10°C by several degrees. Again, since the symptoms didn't change when you re-TIM'd, the problem is elsewhere.

Consider this:

The 4th Gen Haswell 4770K was first launched 2nd Quarter 2013, so yours could be 5+ yrs old. Intel's TIM between the Die and IHS dries out over time and hardens to the consistency of chalk, losing it's thermal bond with the Die. Older 3rd Gen Ivy Bridge (launched 2nd Quarter 2012) as well as newer 4th Gen Devil's Canyon processors (launched 2nd Quarter 2014) with the same problem as yours have been delidded, revealing that thermal performance will progressively degrade over time due to internal TIM failure.

See Section 9 - The TIM Problem - Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

I know you don't want to hear this, but thanks to our friends at Intel, the only way to correct the problem is to delid. If you don't want to do it yourself, then you can send your CPU to Silicon Lottery - https://siliconlottery.com/collections/all/products/delid

They warranty a first class professional job for a fair price. Your Core temperatures will be ~18°C lower at 100% workload, and you'll also be able to reduce Vcore by about 20 millivolts, which will further reduce Core temperature while maintaining your stable 4.5 overclock. If you don't want to delid, then your only other option is to just live with it. Sorry.

CT

Thanks for the informative answer. I was honestly quite suprised that you answered my thread! I found your thermal guide extremely helpful and i ending up reading all of it, and I used your methods for my overclock. I did kinda knew in the back of my head that the thermal paste inside my CPU is old and dried up over time therefore resulting in higher temperatures, i just wanted to try and avoid delidding.

I tried doing it in the past with the razerblade method but my razerblade was too thick and my thin razerblade wasn't sharp enough, anyway getting a into a little too much detail. I'd might consider buying a delidding tool the one where you put the CPU in the vice and delid that way. I think that's pretty much the safest way to delid as the razerblade method has a chance of cutting off the resisters on the left side (triangle facing bottom left). Before i go and delid with the vice tool can you give me some heads up on any risks i might face when trying to delid so i don't break it?

Also, my ambient temps were at 25c.
My cooler is the NZXT Kraken X62 280mm AIO watercooler.
The thermal paste i used was Artic Cooling MX-2.

And is there any point overclocking my Ring Ratio and should i leave it stock? and will lowering it to stock and the ring's voltage lower my temps?

 

[CompuTronix]

There are several delidding tools now available on the Internet, any of which will provide a safe delid without the use of a vice.

The brutal "bench vice / wood block & hammer" method sometimes used on 22 nanometer processors prior to the advent of delidding tools, is absolutely destructive to the newer 14 nanometer processors, as the substrate is considerably thinner, and will not tolerate being pinched in a bench vice, or a lateral blow to the IHS.

Regarding the ring, as long as it's within ~300MHz of the Cores, the difference in performance is typically negligible. Ring voltage can be decreased accordingly, which will also affect Core temperatures.

CT

 

[MaxTehLegend]

There are several delidding tools now available on the Internet, any of which will provide a safe delid without the use of a vice.

The brutal "bench vice / wood block & hammer" method sometimes used on 22 nanometer processors prior to the advent of delidding tools, is absolutely destructive to the newer 14 nanometer processors, as the substrate is considerably thinner, and will not tolerate being pinched in a bench vice, or a lateral blow to the IHS.

Regarding the ring, as long as it's within ~300MHz of the Cores, the difference in performance is typically negligible. Ring voltage can be decreased accordingly, which will also affect Core temperatures.

CT

Oh, when I said the vice tool i meant something like this:

And i'll definitely take that into consideration. Thank you for the great info!