Coffee Lake to have similar TIM as KabyLake?

[For Science!]

Positive spin-doctor throw: 

If you own a delid-die-mate then rejoice! It will likely be compatible with Coffee Lake CPUs if they decide to do a crappy job with the IHS installation again.

 

This is quite speculative (perhaps too speculative for tech news, but I'll let the mods decide), but I picked up this nugget of information from der8auer's live stream yesterday.

 

Quote

"I can actually confirm that my delid-die-mate works for coffee lake"

 

Since der8auer doesn't recommend delidding soldered CPUs, it can be extrapolated that Coffee Lake is likely to have a similar TIM to Kaby Lake (and probably therefore likely to suffer from similar thermals?)

 

What do you think?

 

 

 

[porina]

I wouldn't expect Intel to suddenly turn around now and solder everything. As long as it is an easy delid it is no worse than current. 

[zMeul]

except it's not the TIM that is the problem, it's the z height between the die and the IHS

but, no matter how many times this is told and repeated, here, on Anand and everywhere else ... you blame the TIM 

[For Science!]

1 minute ago, zMeul said:

-snip-

Not entirely convinced this is actually true, I have delidded and relidded my CPU with liquid metal (I.e. re-introduced said Z-height difference) and observed substantially improved thermals, and there are also plenty of others in the same boat. 

 

On the other hand you have Linus who delidded a 6700K and used conventional thermal paste and no relid (and therefore no Z-height) and only oberseved marginal temperature improvements. So I would argue that the paste material does make a notable impact on performance.

 

 

[Princess Luna]

To be honest I have been defending since Skylake that overclocking mainstream Intel CPUs is a waste of money, my experience shows the i7 6700 and i7 7700 performs sufficiently for any high end gaming rig and light workstation purposes being that the bump in the core frequency adds insignificant gains once compared to the absurdly high priced tag of "overclocking".

 

Z170/270 chipset, cooling and delidding are very high expenses that 99% of people buying those didn't really need to have the end user experience they wanted... but reasons like aesthetics, bragging rights and misleading youtube crap will ensure Intel keeps getting easy money...

 

Ryzen on the other hand kept alive the old true of overclocking provides FREE welcome gains, thanks AMD.

[Cookybiscuit]

I really, really hope they are using toothpaste again. The 7700K isn't a bad housefire by any means, but a 6-core equivalent still using toothpaste will be a really nice housefire for the price, AMD can't hope to match it's thermal outputs.

[tom_w141]

1 hour ago, For Science! said:

Coffee Lake is likely to have a similar TIM to Kaby Lake

Intel has used the exact same TIM for years and if they aren't soldering 18 core cpus then they sure as shit aren't soldering the mainstream. Coffeelake like kabylake is apparently too small to solder anyway.

[tom_w141]

41 minutes ago, For Science! said:

I would like to think Intel have their reasons for using paste instead of solder

Extreme longevity. Using paste means the chips are good for a crazy amount of years, like 15+. Personally i'd rather a soldered chip that dies in 5-10 years but hey ho I guess the mainstream users keep their cpus longer. Also solder can wreck a chip in production so you cut cost and increase yields by using paste. Lastly as said above the skylake onwards die is too small to be soldered apparently.

[For Science!]

3 minutes ago, tom_w141 said:

Extreme longevity.

I understand this point very well, but I also am a bit salty about Intel's strong stance on "If its under a 100 degrees celcius it is within spec" approach, since overclocking on these KabyLake seems to almost guarantee you are in the high 80's and 90's if you don't delid and overclock. 

 

I would question the long term longevity over a chip running with paste at close to 90 degrees vs a delidded chip running in the 60's-70's. Clearly once bought in a system, an intel chip rocking liquid metal is also very stable and provides adequate headroom that was arguably "factory throttled" by the manufacturer.

 

With 2 more physical cores on the same PCB and similar clocks, I dread to think what the thermals will be like on a non-delidded Coffee Lake CPU.

 

Or perhaps the consumer perspective needs to change and having a CPU at stock load temperatures of 80 degrees should be now perceived as "20 degrees of overclocking headroom" instead of "oh no its too hot". Although I doubt that is going to change while delidding provides a quick way to drop temperatures close to 20 degrees. While voiding the 3 year warranty, you "probably" extend the lifetime of the chip this way anyway.

[tom_w141]

2 minutes ago, For Science! said:

I would question the long term longevity over a chip running with paste at close to 90 degrees vs a delidded chip running in the 60's-70's. Clearly once bought in a system, an intel chip rocking liquid metal is also very stable and provides adequate headroom that was arguably "factory throttled" by the manufacturer.

Exactly you aren't meant to overclock them. They are designed for extreme longevity at stock. If you overclock it and kill it then thats your problem as far as they are concerned.

 

EDIT: One thing for certain is that manufacturers of delidding tools will make bank  for X299 and the mainstream.

[Morgan MLGman]

1 hour ago, Princess Cadence said:

To be honest I have been defending since Skylake that overclocking mainstream Intel CPUs is a waste of money, my experience shows the i7 6700 and i7 7700 performs sufficiently for any high end gaming rig and light workstation purposes being that the bump in the core frequency adds insignificant gains once compared to the absurdly high priced tag of "overclocking".

 

Z170/270 chipset, cooling and delidding are very high expenses that 99% of people buying those didn't really need to have the end user experience they wanted... but reasons like aesthetics, bragging rights and misleading youtube crap will ensure Intel keeps getting easy money...

 

Ryzen on the other hand kept alive the old true of overclocking provides FREE welcome gains, thanks AMD.

I suppose that depends on your resolution and your GPU, I reverted my 6700K & RAM to their stock values when I got the 980Ti to be 100% sure I'll OC it to be stable and haven't got the time to OC those again, now when I'm playing BF1 (with my 980Ti at stock, which admittedly is a very high "stock" speed of fluctuating between 1405-1418MHz) on my potato monitor and resolution (24" 1920x1200) I'm seeing even up to 95-96% of CPU usage at times with regular processes in the background such as Firefox, Steam, Origin, MSI Afterburner and anti-virus+malware software so I'll definitely overclock the CPU and RAM back again as it will surely yield benefits in performance

 

I'd probably just buy a Ryzen 7 1700 & OC it if I were building my PC now, sadly I didn't want to wait either for Kaby which was known to bring basically nothing over Skylake besides higher OC potential or for Ryzen which I actually didn't believe was going to be competitive.

 

1 minute ago, Demonking said:

enthusiasts overclockers yeah right at 4Ghz woo hoo.

I'd take eight 4GHz cores over four 5GHz ones though, especially for quite a bit less money

[PCGuy_5960]

He also said that he doesn't have Coffeelake iirc

[tom_w141]

3 minutes ago, Demonking said:

enthusiasts overclockers yeah right at 4Ghz woo hoo.

*watches the point whoosh overhead*

[PCGuy_5960]

1 hour ago, For Science! said:

Since der8auer doesn't recommend delidding soldered CPUs, it can be extrapolated that Coffee Lake is likely to have a similar TIM to Kaby Lake (and probably therefore likely to suffer from similar thermals?)

4 minutes ago, PCGuy_5960 said:

He also said that he doesn't have Coffeelake iirc

1 hour ago, For Science! said:

What do you think?

Not that big of a deal IMHO. Delidding an 1151 CPU is dead simple

[For Science!]

30 minutes ago, PCGuy_5960 said:

He also said that he doesn't have Coffeelake iirc

I guess he doesn't have it but somebody else with his delid tool could have done it

 

25 minutes ago, PCGuy_5960 said:

Not that big of a deal IMHO. Delidding an 1151 CPU is dead simple

 

I guess with all this discussion of overclocking voiding warranty then you might as well go balls to the wall voiding the crap out of the warranty if you're going to do it

[MageTank]

2 hours ago, For Science! said:

Not entirely convinced this is actually true, I have delidded and relidded my CPU with liquid metal (I.e. re-introduced said Z-height difference) and observed substantially improved thermals, and there are also plenty of others in the same boat. 

 

On the other hand you have Linus who delidded a 6700K and used conventional thermal paste and no relid (and therefore no Z-height) and only oberseved marginal temperature improvements. So I would argue that the paste material does make a notable impact on performance.

 

 

@zMeul is right. First of all, the old linus video. That is exactly the result you see when you use a thermal paste that is terrible for thermal shock (especially under intense thermal cycles of bare die heat loads, both MasterGel and Noctua's stock paste are known for failing under this load) and secondly, he didn't spread the paste, and simply expected the IHS to do that for him. Your socket retention mechanism on the board absorbs the vast majority of the mounting pressure, so the IHS itself will not be able to evenly spread the paste across the die. Either way, linus did still see a per-core variance drop in his tests (meaning the cores max temperature differential was less after delidding) but he failed to mention it.

 

I've delidded many CPU's too. More than I am able to count, and have used various kinds of paste, ranging from CLU to Gelid, and I can tell you, the difference is less pronounced when you use a solid silicone based paste and spread it properly, vs liquid metal alone. Granted, LM is still far superior, offering a 10-15C difference over Gelid, but it's still not the 20-25C difference seen when removing the glue in the process. Not only that, but I've also purchased a gigantic jar of Dow Corning paste (the exact same blend used on Intel CPU's from Haswell and above) and compared to Gelid, it was only 2-3C difference. Meaning it still dropped temperatures substantially compared to Intel using the exact same paste along with their gluing process.

 

If you need further evidence that the process itself was at fault, look no further than X299. Yes, not soldering that platform's SKU's certainly resulted in higher temperatures, and a bottleneck at the die->IHS heat transfer, the per-core variance on those chips are extremely low. Almost identical to what we see when delidding/using solder. This means that the paste is perfectly even this time around, and I blame this on the raised substrate in which the die is raised on an additional substrate layer, rather than be on the same layer as the glue itself. 

 

TL:DR? Liquid metal is still the best way to go for thermals when delidding, but stock paste wasn't the reason people needed to delid out of necessity. High per-core variance caused by the gluing application process caused higher than tolerable per-core temp variance, and people had to delid to fix that. The paste itself (Dow Corning TC-1996) while not as great as liquid metal, was still perfectly fine for even temps had the gluing process not impact the application of the paste.

Edited July 31, 2017 by wkdpaul
Cleaned up

[WkdPaul]

Thread cleaned.

 

Please keep it on-topic and keep it civil.

[Ravager911]

2 hours ago, zMeul said:

except it's not the TIM that is the problem, it's the z height between the die and the IHS

but, no matter how many times this is told and repeated, here, on Anand and everywhere else ... you blame the TIM 

Reducing the distance between the die and the heatspreader would require much tighter manufacturing tolerances, which would cost significantly more.

That is the reason the manufactures don´t do that. Using liquid metal is also no option in terms of reliability. The only alternative is soldering which costs significantly more.

 

Only AMD is doing soldering but you gain no overclocking headroom on the Zen-architechture whatsoever. They could save that money too.

[For Science!]

14 minutes ago, MageTank said:

 

but if this were to be true (more questioning, than doubting) isn't it even a poorer excuse from Intel in for example not shaving down the shoulder on the underside of the IHS that increases the gap between the die and IHS?

 

I get the impression that for the 7700k while per core variance was an issue too, even the "good cores" were hot too (was the case for mine anyway). I honestly don't think my temps would be much better if I re-delidded and didn't use a new thin layer silicon glue.

 

I mean if it could be solved by just applying the paste better, or pushing the IHS down harder during glueing, or use less glue, whatever....then intel should have done it for earlier products, and people would not have to have this debate.

[JoostinOnline]

2 hours ago, For Science! said:

Not entirely convinced this is actually true, I have delidded and relidded my CPU with liquid metal (I.e. re-introduced said Z-height difference) and observed substantially improved thermals, and there are also plenty of others in the same boat. 

 

On the other hand you have Linus who delidded a 6700K and used conventional thermal paste and no relid (and therefore no Z-height) and only oberseved marginal temperature improvements. So I would argue that the paste material does make a notable impact on performance.

 

 

The reason it didn't work is because he didn't bother to remove the adhesive.  I went into more detail here.

[MageTank]

Just now, For Science! said:

but if this were to be true (more questioning, than doubting) isn't it even a poorer excuse from Intel in for example not shaving down the shoulder on the underside of the IHS that increases the gap between the die and IHS?

 

I get the impression that for the 7700k while per core variance was an issue too, even the "good cores" were hot too (was the case for mine anyway). I honestly don't think my temps would be much better if I re-delidded and didn't use a new thin layer silicon glue.

 

I mean if it could be solved by just applying the paste better, or pushing the IHS down harder during glueing, or use less glue, whatever....then intel should have done it for earlier products, and people would not have to have this debate.

Actually, yes. Bits Power actually made a custom IHS that does exactly that (and also has a cutout to form-fit the IHS to the die) that shows a 2-3C improvement over the stock IHS, that most are attributing to exactly that process. 

 

Another thing that made the 7700k hotter than it's 6700k predecessor, was the redesigned fin pitch, which is how they were able to get the extra overclocking headroom over Skylake. However, it did come at the cost of higher thermals. 

 

Nobody is (or should be) defending Intel for cutting corners on their thermal paste application. We have been tired of it for years. Need I remind everyone of the original Haswell CPU's? Devil's Canyon was basically a requirement for Haswell to survive, as the original Haswell CPU's were a hot mess. 

 

If you do want to test, you can use some of Shin Etsu's cheaper pastes (which is roughly the same quality as Dow Corning's TC-1996, and Dow Corning is hard to find at the moment), though you would kinda need a stock, undelidded CPU to see the difference. If anything, I can probably delid one of my spare CPU's and post the results (along with evidence of me using the pastes). Got a few different CPU's to test (4790k, 4690k, 6600T). Either way, the paste that Intel uses is designed for longevity. No other paste that I have tested, has even come close to the pastes offered by DC. Shin Etsu's offerings are about as close as we can get from a consumer purchasing standpoint. Gelid is pretty durable, but even that requires a repaste every 1-2 years.

 

If longevity is your intent, you can't beat an extremely durable silicone based paste. Solder, for as well as it performs, still has a very finite lifespan due to the nature of solder under intense thermal cycles. I just wish Intel took more precautions when applying the paste, and adhering the IHS to the substrate.

[The Benjamins]

13 minutes ago, MageTank said:

Actually, yes. Bits Power actually made a custom IHS that does exactly that (and also has a cutout to form-fit the IHS to the die) that shows a 2-3C improvement over the stock IHS, that most are attributing to exactly that process. 

 

Another thing that made the 7700k hotter than it's 6700k predecessor, was the redesigned fin pitch, which is how they were able to get the extra overclocking headroom over Skylake. However, it did come at the cost of higher thermals. 

 

Nobody is (or should be) defending Intel for cutting corners on their thermal paste application. We have been tired of it for years. Need I remind everyone of the original Haswell CPU's? Devil's Canyon was basically a requirement for Haswell to survive, as the original Haswell CPU's were a hot mess. 

 

If you do want to test, you can use some of Shin Etsu's cheaper pastes (which is roughly the same quality as Dow Corning's TC-1996, and Dow Corning is hard to find at the moment), though you would kinda need a stock, undelidded CPU to see the difference. If anything, I can probably delid one of my spare CPU's and post the results (along with evidence of me using the pastes). Got a few different CPU's to test (4790k, 4690k, 6600T). Either way, the paste that Intel uses is designed for longevity. No other paste that I have tested, has even come close to the pastes offered by DC. Shin Etsu's offerings are about as close as we can get from a consumer purchasing standpoint. Gelid is pretty durable, but even that requires a repaste every 1-2 years.

 

If longevity is your intent, you can't beat an extremely durable silicone based paste. Solder, for as well as it performs, still has a very finite lifespan due to the nature of solder under intense thermal cycles. I just wish Intel took more precautions when applying the paste, and adhering the IHS to the substrate.

this reminds me how much I HATE my 4770k, impossible to cool, and get high OC due to the TIM.

 

I am done with non soldered CPU's.

[For Science!]

Well, perhaps I can put a positive spin to the story:

 

If you own a delid-die-mate then rejoice! It will likely be compatible with Coffee Lake CPUs if they decide to do a crappy job with the IHS installation again.

 

 

[MageTank]

1 minute ago, For Science! said:

Well, perhaps I can put a positive spin to the story:

 

If you own a delid-die-mate then rejoice! It will likely be compatible with Coffee Lake CPUs if they decide to do a crappy job with the IHS installation again.

 

 

Hopefully the same is said for the Rockit Cool delid tool. Am not enthusiastic about having to buy a new platform to use Coffeelake, but at the very least, I would be able to take solace in the fact that I could delid my friends CPU's if they decided to go that route. 

[themctipers]

when will intel run out of their expired mayo paste?

 

at least solder the K series and the HEDT, people who buy those want good temps and cverclock, and won't keep their computers for a decade unlike OEM machines.