CPU Temps

[awebos]

Hi Guys!

 

Following your advice, I went for upgrading my PSU & CPU instead of getting an overkill GPU. Now, I made an error when running the numbers and ended up with a $300+ CPU and a $30 cooler (7700k & 212 EVO). I made a custom fan curve that goes like this:
- 40% at 30°

- 60% at 50°

-100% at 70°

 

I haven't overclocked just yet and when running Aida64, the max temp I got was 71°. However, would you consider that the fan curve might be optimized? Does anyone with a similar setup has similar temps? 

 

Also, what would be a good upgrade when it comes to cooling the CPU? Would you rather go with air or AIOS?

 

Thanks in advance for your help, have a great day!

[Majestic]

Intel applied liquid aids between the chip and the heatspreader with Kaby Lake. They just run hot because of it.

That said, the 212 Evo is not a great cooler. Mugen 5 is probably a solid upgrade. Or a NH-U14S.

[PCGuy_5960]

17 minutes ago, Majestic said:

Intel applied liquid aids between the chip and the heatspreader with Kaby Lake.

 Nope  The thermal paste is fine... But Intel uses too much adhesive so the IHS doesn't make good contact with the die. If you delid the CPU and don't remove the adhesive, the CPU will continue to run hot.

[paddy-stone]

34 minutes ago, awebos said:

Hi Guys!

 

Following your advice, I went for upgrading my PSU & CPU instead of getting an overkill GPU. Now, I made an error when running the numbers and ended up with a $300+ CPU and a $30 cooler (7700k & 212 EVO). I made a custom fan curve that goes like this:
- 40% at 30°

- 60% at 50°

-100% at 70°

 

I haven't overclocked just yet and when running Aida64, the max temp I got was 71°. However, would you consider that the fan curve might be optimized? Does anyone with a similar setup has similar temps? 

 

Also, what would be a good upgrade when it comes to cooling the CPU? Would you rather go with air or AIOS?

 

Thanks in advance for your help, have a great day!

It's fine as it is, real world situations would be unlikely to run that load, so should be cooler than that during gaming for example. I wouldn't change anything TBH, 71C isn't a bad temp at all for stress test.. but if you're going to OC you might want to look at an AIO or a better air cooler.

[Majestic]

5 minutes ago, PCGuy_5960 said:

 Nope  The thermal paste is fine... But Intel uses too much adhesive so the IHS doesn't make good contact with the die. If you delid the CPU and don't remove the adhesive, the CPU will continue to run hot.

No the TIM isn't fine, it's not fine at all. It's just some random 5 w/mk gunk, when solder is atleast 80+ w/mk. Removing the TIM and adhesive, but placing the same 5 w/mk stuff back on the chip results in virtually the same temperatures. It's def. the thermal paste.

 

I know because I actually tested it myself.

[PCGuy_5960]

6 minutes ago, Majestic said:

No the TIM isn't fine, it's not fine at all. It's just some random 5 w/mk gunk, when solder is atleast 80+ w/mk. Removing the TIM and adhesive, but placing the same 5 w/mk stuff back on the chip results in virtually the same temperatures. It's def. the thermal paste.

 

I know because I actually tested it myself.

I am not sure, according to @done12many2 the excessive adhesive is more of an issue than the thermal paste itself. 

And soldering isn't an option:

http://overclocking.guide/the-truth-about-cpu-soldering/

[jools]

fine temps.....for evo 212 cooler.....90+ is hot

[jools]

fyi

http://www.tomshardware.com/reviews/intel-kaby-lake-core-i7-7700k-i7-7700-i5-7600k-i5-7600,4870-8.html

[Majestic]

9 minutes ago, PCGuy_5960 said:

I am not sure, according to @done12many2 the excessive adhesive is more of an issue than the thermal paste itself. 

And soldering isn't an option:

http://overclocking.guide/the-truth-about-cpu-soldering/

Well yes it's a tricky process to do by yourself, but that's not what i'm suggesting here.

[PCGuy_5960]

2 minutes ago, Majestic said:

Well yes it's a tricky process to do by yourself, but that's not what i'm suggesting here.

I know. Read the article:

Quote

Whenever I read sentences like “What a ripoff – Intel doesn’t even solder a 300 USD CPU” or “Why does intel save 2 USD on soldering” I’m thinking

Stop hating on Intel. Intel has some of the best engineers in the world when it comes to metallurgy. They know exactly what they are doing and the reason for conventional thermal paste in recent desktop CPUs is not as simple as it seems.

Micro cracks in solder preforms can damage the CPU permanently after a certain amount of thermal cycles and time. Conventional thermal paste doesn’t perform as good as the solder preform but it should have a longer durability – especially for small size DIE CPUs.

 

[Majestic]

6 minutes ago, PCGuy_5960 said:

I know. Read the article:

 

I did read it, and I thought the conclusion was very weak. He goes through great lengths to explain everything and basically ends it with "intel dindu nuffin".

Because of the material being rare, and them supposedly not being able to control for thermal expansion all of a sudden eventhough they have 'the best metallurgy engineers'.

[awebos]

Thank you all very much! 

[done12many2]

2 hours ago, Majestic said:

Intel applied liquid aids between the chip and the heatspreader with Kaby Lake. They just run hot because of it.

That said, the 212 Evo is not a great cooler. Mugen 5 is probably a solid upgrade. Or a NH-U14S.

 

Out of curiosity, what "liquid aids" did Intel use on Kaby Lake that differed from Skylake?

 

 

1 hour ago, Majestic said:

No the TIM isn't fine, it's not fine at all. It's just some random 5 w/mk gunk, when solder is atleast 80+ w/mk. Removing the TIM and adhesive, but placing the same 5 w/mk stuff back on the chip results in virtually the same temperatures. It's def. the thermal paste.

 

I know because I actually tested it myself.

 

Sounds like you used something you determined to be compatible and not necessarily the "same".  Did you actually use the same OEM TIM that Intel used on the 7700k?

 

The OEM TIM is not the issue.  Sure it's not the best, but it's far from the worst.  The single largest contributing factor is the application process itself.  The use of too much adhesive actually doesn't allow the IHS to compress down on the die much or at all.  The OEM TIM is simply filling that small gap, but without any pressure, it's doing a poor job of allowing the transfer of heat.

 

You can delid a 7700k, scrape away the OEM silicone adhesive on the substrate only, place the IHS back on with the OEM TIM still in place and temperatures will drop as a result.

 

I'm fairly confident that Intel was intentionally aggressive with the adhesive to serve as a means in protecting the die from excessive clamping force as a result of improperly installed coolers. 

 

1 hour ago, Majestic said:

I did read it, and I thought the conclusion was very weak. He goes through great lengths to explain everything and basically ends it with "intel dindu nuffin".

Because of the material being rare, and them supposedly not being able to control for thermal expansion all of a sudden eventhough they have 'the best metallurgy engineers'.

 

It's not that they can't do it, but more like at what cost?  Soldering a Kaby Lake chip is nothing like soldering a 2500k.  Sure they can figure it out, but I'm sure they determined early on that the cost to do it safely with Kaby Lake exceeded what it needed to be for a mainstream chip.  That's if they even considered it at all.  We're not talking about an Extreme chip.

 

 

 

@MageTank

@PCGuy_5960

[MageTank]

4 hours ago, Majestic said:

Intel applied liquid aids between the chip and the heatspreader with Kaby Lake. They just run hot because of it.

That said, the 212 Evo is not a great cooler. Mugen 5 is probably a solid upgrade. Or a NH-U14S.

Nah. They used the exact same Dow Corning TC-1996 paste they've used for years now. Nobody complained about Haswell's paste (maybe pre-DC, but that was mostly the process they used to apply it), nobody complained about Broadwell's paste, and nobody complained about Skylake. Why is it all of a sudden a problem on Kaby? The fact is, the paste has never been an issue. The changes to the IHS and substrate thickness (without changing the glue application process) has caused quite a pickle with thermals. 

 

Surely a man that has claimed to test this, like yourself, would have seen the paste and pockets of air on the bare die itself. You would have also seen that using the exact same TC-1996 paste (which you can get by the gallon on Ebay for cheap) would still improve temps by upwards of 5C over stock, just by manually spreading the paste yourself and not using an excessive amount of expanding silicone glue. 

 

Then again, what do I know? I've only done this several times before, with various pastes, and have also discussed my findings with other enthusiasts that have done the same over the past few years. Let's just blame Intel's choice of the most durable paste, and not their terrible adhesion process. 

[Majestic]

3 hours ago, done12many2 said:

Sounds like you used something you determined to be compatible and not necessarily the "same".  Did you actually use the same OEM TIM that Intel used on the 7700k?

 

The OEM TIM is not the issue.  Sure it's not the best, but it's far from the worst.  The single largest contributing factor is the application process itself.  The use of too much adhesive actually doesn't allow the IHS to compress down on the die much or at all.  The OEM TIM is simply filling that small gap, but without any pressure, it's doing a poor job of allowing the transfer of heat.

 

I'm fairly confident that Intel was intentionally aggressive with the adhesive to serve as a means in protecting the die from excessive clamping force as a result of improperly installed coolers. 

 

It's not that they can't do it, but more like at what cost?  Soldering a Kaby Lake chip is nothing like soldering a 2500k.  Sure they can figure it out, but I'm sure they determined early on that the cost to do it safely with Kaby Lake exceeded what it needed to be for a mainstream chip.  That's if they even considered it at all.  We're not talking about an Extreme chip.

 

 

 

@MageTank

@PCGuy_5960

It was a 5 w/mk paste. Dow Corning TC-1996 is 4 w/mk. Any improvements I notice are from the slightly higher thermal conductivity.

 

If the application process was bad, i'd see more substantial gains. So I looked around the web for something to break the subjective argument between us and found this:

 

 

So, the glue adds about 0.06mm of spacing in interface material, using 4 w/mk of the 1996 paste, 177mm2 from the 4670K we can look at how much it degrades the thermal conductivity. I don't know the baseline, but well take 0,5mm. We're only looking at relative numbers here.

 

(0,5*1000)/(4*177) = 0,706
(0,506*1000)/(4*177) = 0,715

 

So in theory it should decrease performance by about 2,3%. Unless I'm doing something wrong. As for protecting from clamping forces, the socket is already adding a huge amount of clamp. Not sure the cooler will exceed this.

 

That is a whole lot of assumptions there. They can do it with the X99 platform chips, it's just a way of cutting costs. And because people are so gullible and they had 0 competition we kept buying them. I'm not sure why the fuck you're defending this practice, it hurts you as a consumer aswell.

 

1 hour ago, MageTank said:

Nah. They used the exact same Dow Corning TC-1996 paste they've used for years now. Nobody complained about Haswell's paste (maybe pre-DC, but that was mostly the process they used to apply it), nobody complained about Broadwell's paste, and nobody complained about Skylake. Why is it all of a sudden a problem on Kaby? The fact is, the paste has never been an issue. The changes to the IHS and substrate thickness (without changing the glue application process) has caused quite a pickle with thermals. 

 

Surely a man that has claimed to test this, like yourself, would have seen the paste and pockets of air on the bare die itself. You would have also seen that using the exact same TC-1996 paste (which you can get by the gallon on Ebay for cheap) would still improve temps by upwards of 5C over stock, just by manually spreading the paste yourself and not using an excessive amount of expanding silicone glue. 

 

Then again, what do I know? I've only done this several times before, with various pastes, and have also discussed my findings with other enthusiasts that have done the same over the past few years. Let's just blame Intel's choice of the most durable paste, and not their terrible adhesion process. 

Well I complained about Haswell, hence my chip was delidded a couple months after buying it... I don't care about their rationals, I care about them cutting costs whilst not cutting the price of each unit. And it's not like they were pressed for margins.

 

You can maybe get a slight decrease on stock, but once you put higher voltages on it (1.3V+) any non liquid metal TIM is going to fumble. So tbh it doesn't really matter which TIM they use, it's just utter crap and they shouldn't have done it.

 

Not sure why you're being so divisive. I have done the same testing and my conclusions differ from yours. Why you're using this misplaced sarcasm is beyond me.

[MageTank]

1 minute ago, Majestic said:

Well I complained about Haswell, hence my chip was delidded a couple months after buying it... I don't care about their rationals, I care about them cutting costs whilst not cutting the price of each unit. And it's not like they were pressed for margins.

 

You can maybe get a slight decrease on stock, but once you put higher voltages on it (1.3V+) any non liquid metal TIM is going to fumble. So tbh it doesn't really matter which TIM they use, it's just utter crap and they shouldn't have done it.

 

Not sure why you're being so divisive. I have done the same testing and my conclusions differ from yours. Why you're using this misplaced sarcasm is beyond me.

It's not being divisive, it's making sure the blame is properly attributed to the real problem at hand. We have people here, in this very thread, that have tested conventional pastes after delidding, seeing improvements. I've even gone as far as to buy the exact same paste used by Intel (and EVGA, since they use the same paste on their GPU's, and even sent me a tube during that thermal pad fiasco) and the results were very clear, a 5C improvement. When delidding several CPU's over the past few generations, one thing became very clear. The black silicone glue they used, caused gaps to form between the die and IHS. I don't know if this was due to some sort of expansion (which seems to be the most likely culprit), or if it was applied thicker in one area of the substrate, but it's the only thing I can think of that would cause this kind of variance in thermals when using the exact same paste. It's not like TC-1996 has a burn-in period, and it's highly resistant to copious amounts of thermal cycles.

 

In order for us to have different results, our methodology must be different. My application method is always the same (manual spreading), the amount I use is always the same (I use those inner-ear scratchers with the cupped ends to get exactly the same amount of paste each time, these things are amazing) and I always reattach my IHS using my delid tool's Spyder Relid attachment, to make sure the IHS doesn't move around when re-mounting it. I did buy some Permatex Black Silicone glue, per the recommendation of @done12many2 so I'll try to see exactly how much glue results in the sporadic thermals we see on Intel's stock CPU's. So far, it seems to be universally accepted among enthusiasts that the glue is the #1 issue, which is why almost all of us insist on not re-using the glue. As far as conventional silicone based paste goes, we would never see this drastic of a temperature difference with paste alone. You don't see much of a difference comparing the ancient AS5 to say, MX4 or Gelid (maybe a few C tops), and I certainly wouldn't expect to see an extreme difference between these same pastes on the bare die unless the paste itself cannot withstand severe thermal shock. 

[Majestic]

But the drop could very well be because of the age of the paste, amount of thermal cycles it already endured or damage during storage (sub 0 temperatures on planes or warehouses). Unless you can get a chip with a fresh paste out of the factory, you can't rule this out.

 

But one thing to me is certain, it's applied very crappy regardless of where the fault lies (TIM or Glue). And it shouldn't be the case on these products when before it was possible at the same pricepoint.

[done12many2]

6 hours ago, Majestic said:

Intel applied liquid aids between the chip and the heatspreader with Kaby Lake. They just run hot because of it.

That said, the 212 Evo is not a great cooler. Mugen 5 is probably a solid upgrade. Or a NH-U14S.

 

Still waiting on the class on what mysterious "liquid aids" Intel used on Kaby Lake that makes them some makes Kaby Lake run hotter then Intel chips without the new "liquid aid"  

 

Quote

If the application process was bad, i'd see more substantial gains. So I looked around the web for something to break the subjective argument between us and found this:

 

 

So, the glue adds about 0.06mm of spacing in interface material, using 4 w/mk of the 1996 paste, 177mm2 from the 4670K we can look at how much it degrades the thermal conductivity. I don't know the baseline, but well take 0,5mm. We're only looking at relative numbers here.

 

(0,5*1000)/(4*177) = 0,706
(0,506*1000)/(4*177) = 0,715

 

So in theory it should decrease performance by about 2,3%. Unless I'm doing something wrong. Any idea how much clamping force the socket puts on the CPU? lol. I don't think the cooler will add much.

 

You just did all of that fancy math based on measurements taken from a 3770k.  We're talking about a 7700k and the fact that the change in the application process, resulted in even MORE silicone adhesive being used when compared to previous generations. 

 

Seriously, have you delidded a 7700k, or are you basing this stuff off of experiences with other Intel CPUs?  Not a confrontational question at all, but one that could clarify some misunderstanding.  

 

Quote

That is a whole lot of assumptions there. They can do it with the X99 platform chips, it's just a way of cutting costs. And because people are so gullible and they had 0 competition we kept buying them. I'm not sure why the fuck you're defending this practice, it hurts you as a consumer aswell.

 

Kaby Lake and Haswell-E are two completely different architectures.  Differences in substrates, dies, and who knows what else are some obstacles I can see.  Those smart guys at Intel may have seen more.  Point being, I'm sure they could do it, but at what cost and was it worth it on a MAINSTREAM chip?  Probably not. 

 

By the way, I'm not defending the process at all and I'd love for you to show me where I did?

 

I delidded my 7700k.  Obviously, in doing so, I acknowledged that there was a problem.  All I was trying to do was point out that you mistakenly identified the wrong thing as the problem.  

[Majestic]

Just now, done12many2 said:

 

Still waiting on the class on what mysterious "liquid aids" Intel used on Kaby Lake that makes them some makes Kaby Lake run hotter then Intel chips without the new "liquid aid"  

 

 

You just did all of that fancy math based on measurements taken from a 3770k.  We're talking about a 7700k and the fact that the change in the application process, resulted in even MORE silicone adhesive being used when compared to previous generations. 

 

 

Kaby Lake and Haswell-E are two completely different architectures.  Differences in substrates, dies, and who knows what else are some obstacles I can see.  Those smart guys at Intel may have seen more.  Point being, I'm sure they could do it, but at what cost and was it worth it on a MAINSTREAM chip?  Probably not. 

 

By the way, I'm not defending the process at all and I'd love for you to show me where I did?

 

I delidded my 7700k.  Obviously, in doing so, I acknowledged that there was a problem.  All I was trying to do was point out that you mistakenly identified the wrong thing as the problem.  

Don't be pedantic, I didn't really mean much by it.

 

It was relative, not absolute. Doesn't matter how much it is in absolute numbers. Increase the thickness by twice that amount, and the results will still be nowhere near the numbers we're seeing with LCU. It could be a factor, but the TIM is of more importance to the temperatures.

 

Who cares. Haswell was pasted, Haswell-E was soldered. Why solder the enthusiast platform if it can't be done, or adds more risk. Surely they'd take less risk with $1000 chips.

 

You rationalized the reasons why they did it, other than cynically cut cost.

 

It's pretty obvious it's the TIM. As using LCU gives you a much better result than reapplying the TIM and scraping off the glue.

[done12many2]

 

6 minutes ago, Majestic said:

Don't be pedantic, I didn't really mean much by it.

 

It was relative, not absolute. Doesn't matter how much it is in absolute numbers. Increase the thickness by twice that amount, and the results will still be nowhere near the numbers we're seeing with LCU. It could be a factor, but the TIM is of more importance to the temperatures.

 

Who cares. Haswell was pasted, Haswell-E was soldered. Why solder the enthusiast platform if it can't be done, or adds more risk. Surely they'd take less risk with $1000 chips.

 

You rationalized the reasons why they did it, other than cynically cut cost.

 

It's pretty obvious it's the TIM. As using LCU gives you a much better result than reapplying the TIM and scraping off the glue.

 

Have you delidded a 7700k?

[Majestic]

1 hour ago, done12many2 said:

Have you delidded a 7700k?

No my 4670K, but I don't see how this is relevant. Using LCU has better results than just reapplying paste and removing the lift caused by the glue.

It COULD be a factor, but it's not the most important factor in the high thermals.

[PCGuy_5960]

16 hours ago, Majestic said:

Who cares. Haswell was pasted, Haswell-E was soldered. Why solder the enthusiast platform if it can't be done, or adds more risk. Surely they'd take less risk with $1000 chips.

They do take less risk. Haswell-E has a die size of 365mm^2 and Broadwell-E has a die size of 246mm^2 while Skylake has a die size of 122mm^2. The larger the die the easier it is to solder without any issues. 

http://overclocking.guide/the-truth-about-cpu-soldering/

The article is written by der8auer who probably knows a bit more than both of us 

@done12many2

@MageTank