The Skylake-X Mess Explored: Thermal Paste And Runaway Power

[ElMojoMikeo]

Luckily I have stuff to do that doesn't require loading the processor. As things are the free running fans are coping ok. I will not be running Prime95 today. I like Corsair, I have loads of their stuff. Most of which is really good. But the Link software trying to be a closed loop monitoring input and having to deal with a whole array of other hardware is asking for what has just happened. It seems deathly silent on the forum at the moment. I hope that they haven't let the culprits go home before fixing it. Hmmm.

 

[Karadjgne]

CL has always been a mess, just another gimmick for the sales dept to use. Much prefer CAM, especially now that the ssd/hdd readings are fixed.

I mentioned before about having a bigger die might solve some issues, but was summerily shot down, but I'm still of that opinion. It's understandable that there's only so much real estate available on a mobo, but socket area takes precedence over anything. When you consider that even an asetek pump on a cpu still has a considerable area uncovered by Tim, and on that Tim there's only a 1/3rd? that's actually head to head with the die, you are using very little surface area to transmit all that heat. Spread the wealth? Bigger die, bigger lid, more area for heat transmission. You could even keep the same nano process, just put some distance between the cores, cache etc, which would cut down on radiated heat, etc etc.

 

[the nerd 389]

CL has always been a mess, just another gimmick for the sales dept to use. Much prefer CAM, especially now that the ssd/hdd readings are fixed.

I mentioned before about having a bigger die might solve some issues, but was summerily shot down, but I'm still of that opinion. It's understandable that there's only so much real estate available on a mobo, but socket area takes precedence over anything. When you consider that even an asetek pump on a cpu still has a considerable area uncovered by Tim, and on that Tim there's only a 1/3rd? that's actually head to head with the die, you are using very little surface area to transmit all that heat. Spread the wealth? Bigger die, bigger lid, more area for heat transmission. You could even keep the same nano process, just put some distance between the cores, cache etc, which would cut down on radiated heat, etc etc.

Spreading the CPU out would put an upper limit on the attainable clock frequency. Besides, generally a specific sub-section of the die generates most of the heat. By moving those portions away from each other, you'll bottleneck the rate that you can feed data to that portion.

You're also limited by the fact that the least conductive link in the transfer chain is the paste. That means that the die itself acts as a heat spreader to an extent. This means that spreading the circuitry out won't make as big a difference as you would think.

You'd get lower temperatures, but it'd be due to decreased activity rather than increased thermal conductivity. That's not really ideal in a CPU design.

To actually solve the problem, there's only one method that would be effective: reduce the thermal resistivity (either real or effective) of the die-to-IHS interface material. You can reduce the temperature in other ways, but those alternatives have very significant trade-offs.

 

[ElMojoMikeo]

Karadjgne said:

I remember you did bring this up before and I have thought about it. I have used my processor as comparison 3970X that is a 32nm fab against the new Skylake X as a comparison because their power hungry ratings are so similar. To say over five years have passed and several changes have occurred to X processors, including the power circuitry. It is not that remarkable how closely these two processors show a resistance to being cooled down. Despite all these fab changes and movement of VRM components from outside to inside. If I am not mistaken back outside again. They still exhibit very similar problems.

I also know that in 2012 Intel were heavily criticised for core castration concerning 3960X and the Xeon 2687W. Exactly as it is today the 8 x core Xeon was locked down for thermal reasons.

Well it went as all witch hunts eventually do. Intel takes flake for making good science based decisions.

Eventually it became clear that these two were in fact one. A 3960X was in fact a castrated 2687W. So the story goes is that Intel were charging lots of cents more than they should be for X processors. In fact Intel didn't let anybody down. A 3960X (6 core) overclocked to 4.5GHz will at ease out perform a 2687W (8 core). What does this tell you. Well there is an exponential decline in the performance increase for each core increase. That's not new, that's old, we knew that . Yes we did and did nothing! My point is right there.

Is there something sexy in delidding I am missing. We are thinking of operating though its bottom.

 

[ElMojoMikeo]

This the third day without cooling. Some closed loop I bought into. Now there is no response from Corsair at all. The support site is dead. I have even used my skills as a firmware engineer to seriously pee them off. Nothing, something strange is going on there. What, I don't have a clue. But strange.

 

[ElMojoMikeo]

I am going to programme with my heart from now on. I will do my best where it is needed. What esle can I do die

 

[Arjuna79]

Yay, we are going back to the throttling era, with flashy, bulky 2Kg of copper heatsinks with a myriad of heatpipes in awe-inspiring loops and coolers cranked to their noisy limits, white-hot processors causing 1st degree burns, and people considering a watercooling solution, when they should have no business in it!

Except all of that won't work, because Intel's gunked the chip in thermal paste BELOW the lid and you can't possibly extract 300W worth of heat from the processor fast enough by any means whatsoever, before it bangs its gazillion-headed core at a 105C head-height beam and go back to 95C, and then keeps doing that every 5 seconds like a kid revving a Honda in neutral!

[Fireworks] [Burnouts] [Eagles screaming]

I miss those days. Back to the "Which one is the best air-cooler for 150W TDP" reviews.

 

[bit_user]

...

Wow. Well said! Please post more.

 

[ElMojoMikeo]

Yay, we are going back to the throttling era, with flashy, bulky 2Kg of copper heatsinks with a myriad of heatpipes in awe-inspiring loops and coolers cranked to their noisy limits, white-hot processors causing 1st degree burns, and people considering a watercooling solution, when they should have no business in it!

Except all of that won't work, because Intel's gunked the chip in thermal paste BELOW the lid and you can't possibly extract 300W worth of heat from the processor fast enough by any means whatsoever, before it bangs its gazillion-headed core at a 105C head-height beam and go back to 95C, and then keeps doing that every 5 seconds like a kid revving a Honda in neutral!

[Fireworks] [Burnouts] [Eagles screaming]

I miss those days. Back to the "Which one is the best air-cooler for 150W TDP" reviews.

 

[ElMojoMikeo]

Hmmm! You seem to have completely left out the essential lighting tech we could mount on these beasts. Ooops. Why isn't that grey box in here.

 

[eltouristo]

I did more reading on reasons to not use solder for die to IHS. I wasn't completely convinced but sorta. Is more problematic to use solder for smaller dies. But Intel already used solder in past for dies this small. Also among observations such as micro-cracks and reduced areas of thermal conductivity in solder, I never saw it established that even compromised solder is not still better conductive layer than most pastes. Also it wasn't really clear what conditions produce solder problems. Saw quote of a thermal cycle being -50C to 125C. If solder was put throught that yeah so what. That is not typical or normal unless using exotic colling, and chips throttle or cut off around 100 or 105. So normal cycle is more like 20-100C. EVEN after studying this in depth it's not clear that Intel still hasn't just copped out. Purely speculating...IF they had some concern about deeper heat cycles from Overclockers using liquid Nitrogen or something, they could easily offer pre-delidded chips for those that really really didn't want solder. The 99.997% of us want solder so we can run fans lower, get a little more comfy OC from chip, nothing super crazy. And we are not going to be cycling chips through larger deep thermal cycles. If however there is an sleeping issue that deep study of past solder designs has shown to truly be an issue, where solder starts underperforming paste over time, that would be worthy of consideration. But that doesn't smell plausible at the moment. It seems like if that was true we would know it, but that is an assumption. Still, lack of solder has me for one less impressed at this time. The least Intel could have done is make a better and more public explanation that is actually compelling. That surely has not happened.

 

[eltouristo]

Actually was not quite right to say 'run fans faster' as that may help a little, but even with cold IHS you can't get heat out any faster than conductive capacity of TIM layer. Further cooling of IHS only helps marginally despite great effort, expense, noise of that. It help because higher temp differential flows heat a bit faster than smaller. But if you are trying to do that through a 'thermal bottleneck' you can see it is a misplaced effort with nonsensical gain vs effort.

 

[Karadjgne]

The reason Intel (imho) isn't using solder, opting for paste is cost. That's it in a nutshell. A trained monkey working for 3 bananas a day can slap paste on a lid before its welded to the die. Solder is a whole different beastie. It's gotta be heated to exacting temps, exacting amounts have to be applied, the lid needs to be applied in very short order or there's no weld between the two etc etc etc. Solder is temperamental, finicky, exacting and it's a one shot deal. Screw it up in any way, toss the cpu, it just failed QI. Screw up paste (like that's never happened before,) test cpu if it works, keep it, sell it, it's good (haha) or pop the lid, wipe it off and throw it back on line. That's not counting the fact that pastes can be slightly different in makeup, and still work, impurities being next to meaningless, whereas if solder is off, that changes the melting point, cooling speeds, even viscosity. All in all, it's a much more complex application than paste.

And when you have to pay a Hollywood actor to tell the world that 95% of the world's internet is run by Intel processors, the money had to come from somewhere, so go cheap paste or dig into stockholders profits.

 

[eltouristo]

It would be interesting to repeat these test in a proper case with the cpu cooler and additional fans moving air in and out of the case to see how the VRM's react when they have proper airflow.

I know I'm an outlier, but I still like down-draft air coolers (not that it'd be an option for this beast). With a 120 or 140 mm fan, you get a fair amount of VRM and DIMM cooling.

I like that too...if it's all-copper. I think all air coolers should be all-copper. I've never seen one as thick (as much fin area) and some of the double or triple layer side-draft ones. I think it would be hard to do layers cuz where would you run the heatpipes? But I like the concept. Ive got one cooling my q9650, that still runs great, I keep it OC'd to 3.6, with no overvolt I think. It cools great and keeps fan slow. I also put a cute little all-copper northbridge? cooler in with it's tiny fan. It's side-draft.