Intel Skylake / Kaby Lake

[RichUK]

Well there it is. Same crap paste as usual. Whatever their reasoning is, it's still unacceptable when we know they could've and should've used solder.

Oh dear, that's bad news indeed. Delidding a Skylake-X is just too risky in my opinion (given the de8auer video).

What a shame.

 

[lolfail9001]

Now here's the interesting part.

Woah, that is more savings than i would imagine. Way more.
Eh, whatever, Intel's reaction to people crying about lack of solder is probably befitting that one old gif.

 

[tamz_msc]

Woah, that is more savings than i would imagine. Way more.
Eh, whatever, Intel's reaction to people crying about lack of solder is probably befitting that one old gif.

Your attempted sarcasm hints to me that you thought I'd come up with numbers at least 10 times as much, while my original intent was to show that there is no reason to believe that there's no cost angle involved, whatever its magnitude.

 

[lolfail9001]

Your attempted sarcasm hints to me that you thought I'd come up with numbers at least 10 times as much,

Your sarcasm detector is malfunctioning, i was sincerely surprised that it actually adds up to that much.

 

[tamz_msc]

Your sarcasm detector is malfunctioning, i was sincerely surprised that it actually adds up to that much.

Well indium is produced mostly in China, and it is known that China is tightening its grip on its post-transition elements production for a number of years.

 

[beginner99]

What IS a fact:

Indium: 81.8 W/m K
Thermal Grizzly Cryonaut: 12.5 W/m K

You forgot:
Air-Gap between TIM and IHS: 0.024 W/m K

 

[beginner99]

invoking long term stability as pro-TIM argument is a sad joke.

Not if you mean long term stability of intels finances. It saves money in production and with the current lack of CPU gains they need to have built in obsolescence to force users to buy new chips. So add crappy TIM and 3 years down the line after warranty expires they need to buy new and can't keep it for 8+ years.

 

[tamz_msc]

You forgot:
Air-Gap between TIM and IHS: 0.024 W/m K

While I'm not sure, I don't think there is any air gap, based on IDC's testing here on the 3770K. It seemed to me that the thinness of the TIM layer as the IHS was dropped down on the die is what ultimately matters.

An air gap would be disastrous.

 

[tamz_msc]

So der8auer's comments on the delid video on GN has made me think, he said that the Skylake-X IHS required a much larger amount of force to remove - so much so that he had to redesign his delid tool. This might mean that the adhesive being used forms a thicker layer - more glue, tighter fit. This raises the IHS and results in suboptimal contact between the TIM and the underside of the IHS, just like IDC's findings on the 3770K.

 

[imported_ats]

Claim 1: TIM is better for long-term reliability(as per speculation)
Claim 2: Solder doesn't mean better heat transfer(yours)

These are separate claims.

Neither claim is from official channels, so there is no other way to test except to see how overclocking is affected.

What IS a fact:

Indium: 81.8 W/m K
Thermal Grizzly Cryonaut: 12.5 W/m K

The TIM Intel is using is actually is pretty much the best you can get in the industry for long-term reliability. And no solder can run into issues with heat transfer over time due to cracking and fracturing as it is non-pliable.

And Thermal conductivity numbers really mean jack in isolation. After all, copper is 400 W/mk and makes a shitty TIM. You have to take into account BLT, spread-ability, compression, etc.

 

[imported_ats]

What??? It's a known fact that thermal paste doesn't last as long as solder. When have you ever heard of problems about CPUs that are soldered? Take for example an Athlon 64 X2 5200+ brisbane that I used to own, it used thermal paste. A few years after purchasing it and many experiments at high temps and overclocks the chip became impossible to properly cool (it wouldn't drop below 50c). It turned out that the paste used on the die turned into crud and at that point I promptly replaced it and then regained proper temps.

Sure Intel probably isn't using paste that crappy but paste is paste. I sure as hell won't be buying any HEDT chip that isn't soldered.

You were using a shitty paste that likely had a high solvent content. Something like DC 5022 or 5026 will last virtually forever.

 

[beginner99]

While I'm not sure, I don't think there is any air gap, based on IDC's testing here on the 3770K. It seemed to me that the thinness of the TIM layer as the IHS was dropped down on the die is what ultimately matters.

An air gap would be disastrous.

I more meant it as a stupid joke due to the given high-tolerance of the manufacturing process. For me the problem is not the TIM itself but that Intel could not get it right. And due to comments from Asus they still did not get it right. 30°C drop after deliding shows how disastrous the TIM solution is. Typical too much TIM issue.

 

[imported_ats]

So der8auer's comments on the delid video on GN has made me think, he said that the Skylake-X IHS required a much larger amount of force to remove - so much so that he had to redesign his delid tool. This might mean that the adhesive being used forms a thicker layer - more glue, tighter fit. This raises the IHS and results in suboptimal contact between the TIM and the underside of the IHS, just like IDC's findings on the 3770K.

Or it is simply the reality of a larger IHS which means more contact area resulting in more adhesion.

 

[tamz_msc]

The TIM Intel is using is actually is pretty much the best you can get in the industry for long-term reliability.

See this is precisely what *cannot* be tested, because nobody carries out long-term reliability tests. Might as well say that TI voltage regulator is better in long-term stability than a Fairchild Semiconductor one.

And Thermal conductivity numbers really mean jack in isolation. After all, copper is 400 W/mk and makes a shitty TIM. You have to take into account BLT, spread-ability, compression, etc.

Well there is a reason why indium is used - one of the highest coefficient of thermal expansion for a metal plus its wetting properties, meaning it can expand and stick to the IHS better as temperatures are increased. This has its drawbacks in case of small dies, but it is superior to any polymer-based TIM as far as thermal performance does.

 

[tamz_msc]

Or it is simply the reality of a larger IHS which means more contact area resulting in more adhesion.

Is it any larger than previous socket 2011CPU dies? I don't think so.

 

[DrMrLordX]

And it should be pointed out that the TIM they do use is some of the best TIM on the market. The pump out with thermal cycling is all but non-existent and the degradation with time is also basically non-existent. Year 1, it is probably behind in thermal performance, but it quickly catches up and surpasses other TIMs as they get subjected to pump out and degradation. Intel isn't designing chips for people who think that replacing TIM on a yearly basis is a normal thing and they never have.

It's a good thing they used TIM instead of solder! Solder pumps out so fast and I have to replace it so often and . . . and . . . wait.

Solder doesn't mean better heat transfer than the paste/TIM Intel uses.

Sorry, but that's absolute nonsense. I can't think of a single CPU run in an air/water scenario where the solder cracked from thermal cycling. LN2 runs and the like, maybe. Otherwise, no.

And Thermal conductivity numbers really mean jack in isolation. After all, copper is 400 W/mk and makes a shitty TIM. You have to take into account BLT, spread-ability, compression, etc.

Solder works, it forms a seamless bond between the IHS and die, and it has better W/mk numbers than any polymer TIM by a long shot. The thermal conductivity numbers for indium solder certainly do "mean jack".

 

[LTC8K6]

Perhaps Intel's solder process is just not that good? Delidding the 6950X and using TIM improved the temps. So in that case, TIM was better than whatever solder Intel used.

 

[imported_ats]

See this is precisely what *cannot* be tested, because nobody carries out long-term reliability tests. Might as well say that TI voltage regulator is better in long-term stability than a Fairchild Semiconductor one.

They quite literally do carry out long term testing for TIM reliability. Even baseline TIM testing takes ~1/2 year of continuous thermal cycling @ ~10 minute intervals (8 minute heat up to 90+ degrees C followed by 2 min cooldown to RT) for 20k-25k cycles. And they measure the W/mk continuously.

Well there is a reason why indium is used - one of the highest coefficient of thermal expansion for a metal plus its wetting properties, meaning it can expand and stick to the IHS better as temperatures are increased. This has its drawbacks in case of small dies, but it is superior to any polymer-based TIM as far as thermal performance does.

It is superior when new, it can have significant degradation issues as it ages.

 

[RichUK]

Perhaps Intel's solder process is just not that good? Delidding the 6950X and using TIM improved the temps. So in that case, TIM was better than whatever solder Intel used.

What was the delta?

 

[RichUK]

They quite literally do carry out long term testing for TIM reliability. Even baseline TIM testing takes ~1/2 year of continuous thermal cycling @ ~10 minute intervals (8 minute heat up to 90+ degrees C followed by 2 min cooldown to RT) for 20k-25k cycles.



It is superior when new, it can have significant degradation issues as it ages.

If that's the case (where's the evidence?), then why has Intel been using it for quite some time on previous HEDT gens and I also presume Xeons, too?

 

[imported_ats]

Sorry, but that's absolute nonsense. I can't think of a single CPU run in an air/water scenario where the solder cracked from thermal cycling. LN2 runs and the like, maybe. Otherwise, no.

There is quite literally a report in the last page or so that documents solder cracking from thermal cycling.

Solder works, it forms a seamless bond between the IHS and die, and it has better W/mk numbers than any polymer TIM by a long shot. The thermal conductivity numbers for indium solder certainly do "mean jack".

Only if you haven't done literally any actual reading of the research in the area. The reality is that it has issues just like any TIM and the raw W/mk numbers are meaningless as the material changes over thermal cycles. Indium solder literally degrades into multiple composite materials over time as the Indium combines with the various metals and migration occurs.

 

[vissarix]

Guys take a break talking about TIM since 20 pages already without having benchmarks or anything...

If we did this on a Ryzen thread we would have been reported 100 times for threadcrapping trolling etc...

You can reach 5GHZ on an i7 7700k without delid with an cheap AIO which can go as low as $50 these days...but you seems to want to overclock a Skylake X to 4.8GHz with a cheap $20 air cooler

 

[imported_ats]

Is it any larger than previous socket 2011CPU dies? I don't think so.

The previous dies he was delidding using his tools were 1151 not 2011/2066, so yes they are larger than the previous IHS.

 

[tamz_msc]

There is quite literally a report in the last page or so that documents solder cracking from thermal cycling.




Only if you haven't done literally any actual reading of the research in the area. The reality is that it has issues just like any TIM and the raw W/mk numbers are meaningless as the material changes over thermal cycles. Indium solder literally degrades into multiple composite materials over time as the Indium combines with the various metals and migration occurs.

The only advantage I can see that polymer TIM has is that it's composition over time would not change. Nowhere in the report does it say that indium solder must be replaced with something else for better long term performance, just that there are various things to consider in applying indium solder for optimal operation.

We would have seen old processors with solder heating up quite regularly if that were the case.

 

[Edrick]

Does anyone know if Intel is using TIM on their LGA 3647 Xeon Gold/Platinum CPUs?

If they are going to TIM on these CPUs as well, I think that would disprove any stability concerns with using TIM.