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

[dudmont]

For your motherboard reviews with this CPU, might I suggest that you look up the rated lifetime of the PWM caps?

If those are 5k caps, they'll only last 5000 hours at their rated temperature. Lifetime usually doubles for each 10C under that rating.

These CPUs often end up in entry level workstations. If the intended usage pushes the caps to 75 C, and they are the more common 5k/105C, then conventional wisdom indicates that 10% of the caps should fail by the 40,000 hour mark. In those cases where the CPU is fully loaded most of the time, this will occur 4.5 to 5 years of age. This means that there isn't much room for mistakes in the motherboard layout, case design, or airflow requirements. It's something that potential consumers should take into account if they want to get the most out of this platform.

I'd have to check in more detail to give any estimates for the MOSFET lifetime, but that's another factor to account for if longevity is a priority.

I think you're right on with this issue here. The usefulness of these cpus suggest that they'll be more than powerful enough to handle most tasks assigned to them for many years(maybe 8-10), but will the hardware last that long, and that is an interesting point.
People on the forums(usually gamers) all the time are asking about whether to upgrade their I5 2,3, and 4 series CPUS to newer models and the answer is almost all their queries is no. Their cpu is not a problem yet. Your post is in a similar vein of thought. People may not need more cpu resources 8 years from now, but will their hardware still be running? I'm not sure the mobo makers and intel want their hardware to still be running by that point. Forcing people to buy new is much more profitable than building really quality parts now that last long enough to actually reach obsolescence.
This has been an interesting article to read and equally interesting forum thread to read, thanks to all involved. Very thought provoking.

What I meant was that these boards toe the line in terms of longevity, if my assumptions are correct, and if they were only using a single phase.

Since the VRM phases are not redundant (they divide the load up, and all phases are needed to handle the load), it only takes one failure to brick a board. The odds of a at least one phase out of a set of 12 phases failing in the first year (assuming the L90 is 4.5 years, as above) is a whopping 25%. Note that I'm using a pretty basic estimation procedure to get that number, so it's probably only accurate to +/- 10%. However, even a 15% failure rate in the first year is ridiculous for a motherboard. I don't think you could find a z270 board that bad if you crossed a Biostar with an ECS.

I'm not talking about these boards making it through an extended life. I'm talking about whether or not they'll make it through the first year. I'm not even using unrealistic figures for any of this. Aside from the thermal issues, these CPUs would be among the best values on the market for rendering until ThreadRipper comes out. The thermal issues unfortunately prevent you from using these to their full capacity.

This raises more interesting questions. Do you think mobo manufacturers are taking warranty claims into account in their selection of components? Who in their right minds would plunk down 1300$ for a mobo and cpu, just to have a 25% chance of having to waste time and money troubleshooting and then shipping off a failed mobo, if in one years time there's another 25% chance of having to do it again?
I'm still digesting some of what you're getting at, but correct my thinking if I'm wrong, but you're saying that the mobo's voltage solution is so marginal that any tinkering could brick the mobo quick, fast, and in a hurry?

 

[the nerd 389]

For your motherboard reviews with this CPU, might I suggest that you look up the rated lifetime of the PWM caps?

If those are 5k caps, they'll only last 5000 hours at their rated temperature. Lifetime usually doubles for each 10C under that rating.

These CPUs often end up in entry level workstations. If the intended usage pushes the caps to 75 C, and they are the more common 5k/105C, then conventional wisdom indicates that 10% of the caps should fail by the 40,000 hour mark. In those cases where the CPU is fully loaded most of the time, this will occur 4.5 to 5 years of age. This means that there isn't much room for mistakes in the motherboard layout, case design, or airflow requirements. It's something that potential consumers should take into account if they want to get the most out of this platform.

I'd have to check in more detail to give any estimates for the MOSFET lifetime, but that's another factor to account for if longevity is a priority.

I think you're right on with this issue here. The usefulness of these cpus suggest that they'll be more than powerful enough to handle most tasks assigned to them for many years(maybe 8-10), but will the hardware last that long, and that is an interesting point.
People on the forums(usually gamers) all the time are asking about whether to upgrade their I5 2,3, and 4 series CPUS to newer models and the answer is almost all their queries is no. Their cpu is not a problem yet. Your post is in a similar vein of thought. People may not need more cpu resources 8 years from now, but will their hardware still be running? I'm not sure the mobo makers and intel want their hardware to still be running by that point. Forcing people to buy new is much more profitable than building really quality parts now that last long enough to actually reach obsolescence.
This has been an interesting article to read and equally interesting forum thread to read, thanks to all involved. Very thought provoking.

What I meant was that these boards toe the line in terms of longevity, if my assumptions are correct, and if they were only using a single phase.

Since the VRM phases are not redundant (they divide the load up, and all phases are needed to handle the load), it only takes one failure to brick a board. The odds of a at least one phase out of a set of 12 phases failing in the first year (assuming the L90 is 4.5 years, as above) is a whopping 25%. Note that I'm using a pretty basic estimation procedure to get that number, so it's probably only accurate to +/- 10%. However, even a 15% failure rate in the first year is ridiculous for a motherboard. I don't think you could find a z270 board that bad if you crossed a Biostar with an ECS.

I'm not talking about these boards making it through an extended life. I'm talking about whether or not they'll make it through the first year. I'm not even using unrealistic figures for any of this. Aside from the thermal issues, these CPUs would be among the best values on the market for rendering until ThreadRipper comes out. The thermal issues unfortunately prevent you from using these to their full capacity.

This raises more interesting questions. Do you think mobo manufacturers are taking warranty claims into account in their selection of components? Who in their right minds would plunk down 1300$ for a mobo and cpu, just to have a 25% chance of having to waste time and money troubleshooting and then shipping off a failed mobo, if in one years time there's another 25% chance of having to do it again?
I'm still digesting some of what you're getting at, but correct my thinking if I'm wrong, but you're saying that the mobo's voltage solution is so marginal that any tinkering could brick the mobo quick, fast, and in a hurry?

My observations are less relevant for overclockers, as overclockers don't usually fully load their computers non-stop for more than a week or two of stress testing. I'm considering rendering boxes and other non-stop compute heavy situations. However, if you look at the temperatures reached in the review, and assume a non-stop load on the CPU, these are the numbers you end up with. They aren't pretty.

These all assume the use of standard 105C/5k caps.

At stock settings (160 watt scenario, using 63C as the cap temp), and in a fully loaded build that's crunching numbers non-stop, there's roughly a 10% chance you'll have a cap failure before the first year is over. There's a 20% chance it will fail before the end of the second year.

If you run stock settings with the Enhanced Turbo bug (230 watt scenario, using 76C as the cap temp), again non-stop, there's roughly a 25% chance the board will die before the first year is up. There's a 45% chance it will fail before the end of the second year.

If you overclock to 250 watts (using 80C as the cap temp), and load the computer non-stop, there's about a 32% chance the board will fail in the first year. There's about a 55% chance it will fail within the first two years.

If you overclock to 300 watts (using 85C as the cap temp), and load the computer non-stop, there's about a 43% chance the board will fail in the first year. There's about a 67% chance it will fail in the first two years.

Also, I'm not trying to bash the motherboard's VRM solution. It's actually quite standard, and there's nothing wrong with it on any other CPU. This CPU is so poorly designed that it renders the current industry standard solutions inadequate. The problem is not that the motherboard is badly designed, per se. It's that the CPU is so poorly designed that every other component has absolutely no room for mistakes if you want an acceptable system at the end of the day. Intel is basically offloading the engineering challenges they should have solved with Skylake-X onto motherboard designers, cooler designers, case manufacturers, and basically anyone else who has to work with Skylake-X in any capacity.

 

[Karadjgne]

Ok, I got a question, see if someone smarter than me can give a logical answer. Intel cpu's generally have a much smaller die (only frame of reference being what's listed in the Thermal paste comparisons) than amd cpus. Now take a cooler with 4x direct contact heatpipes. Pretty much the center 2x heatpipes do the lions share of the work, the 2x outer heatpipes seeing next to nothing since there's nothing under them but air on an intel die. Amd having a larger die, will make more contact with more heatpipes as the die is large enough to cover all 4x, so should have better cooling ability with the same cooler (if TDP is somewhat equivalent) . Got that so far? So, wouldn't it stand to reason that if the Threadrippers larger die will also benefit from better cooling ability vrs the skylake-X? Just seems to me it's no different than a 200lb person wearing sneakers, vrs a 200lb person wearing Stilletos, all that heat is so concentrated in 1 small spot, even the largest coolers can't adequately absorb it as the heatpipes are sitting on the hollow part of the lid. Wouldn't this make large aios/fcl almost mandatory, since the micro-fin area of the pump and waterblock fully cover the die without worrying about thermal diffusion on a solid chunk of metal and some tubing?

 

[the nerd 389]

The premise of your question is flawed. You assume that the heat doesn't spread away from the actual die much. This is not the case. The thermal paste between the cooler and IHS has a higher thermal resistance than the heat spreader. This means that the heat basically builds up on the spreader until enough if there to push the thermal energy through the paste. This buildup of heat allows the heat to spread away from the die quite easily.

To tie it into your analogy, imagine those shoes encased in an 18" cube of dense memory foam. Technically, there's still a focal point of the pressure. Realistically, the force is spread evenly enough to effectively disregard the focal point, unless you're trying to optimize the ever loving snot out of the floor.

 

[bit_user]

after seeing the review i am thinking of using my car cooling system to cool down the intel new raging bull.

This could be the situation where setting up your rig to run out of a spare chest freezer makes sense again. :/

You'll be fighting condensation and the fact that refrigerators are made to run at a lower duty cycle. Plus, this could happen:

Yep, started by a refrigerator.

 

[10tacle]

and dont call me a dictator again , be nice or you will be reported. and dont expect that I will read anything you will post after this, dont waste your time. .

Moderators read these posts so you don't have to worry about that. And I said you were behaving as one, not accusing you of being one. You have a problem with someone pushing back on your original snide comments saying nobody needs to overclock (CPU tier irrelevant). Your five downvotes for your original post speaks for itself on what others think of it as well.

 

[bit_user]

Still, it appears that you're just using alternative names to broadcast intellectual superiority, and it's only working on a bit more than 10% of your readers. As for that ego, stick a pin in it.

I can think of more diplomatic ways of reminding folks that delta Kelvin == delta Celsius.

Seriously, this post just makes you look bad. Is it possible this was simply a translation issue?

 

[the nerd 389]

We have enough frame rates today in gaming and we dont need overclocking at all. Why are you focusing on overclocking so much ?

as for professional work , people never overclock workstations with ECC memory.

when is this overclocking hype going to end ?

Why overclocking? just to show off ? and say I got 20 more fps in a game that is already above 100 fps ?

move on people ... i you dont need to overclock there is no need to pay extra money just to hit useless fps that is not going to change anything.

The issue doesn't just manifest itself as an overclocking limitation. Numerous other factors are also affected. For instance, at stock settings, the design of Skylake-X requires you to pay for overclocking-grade cooling to get stock performance. It also requires even those coolers to run at high fan speeds, increasing noise. Beyond that, the finer points of the issue all reduce the longevity of systems based on these CPUs.

Basically, you're going to have to pay for everything an overclocker would if you want the CPUs to run at stock settings. For that money, you cannot overclock. You cannot reduce noise. You cannot maintain the reliability you would get out of a standard cooler.

By assuming that this is only an issue for overclockers you have entirely missed the scope of the problem and the magnitude of the impact of the Skylake-X design flaws.

 

[FPS-fan]

...and all the while, AMD shares continue to climb.

 

[bit_user]

take a cooler with 4x direct contact heatpipes. Pretty much the center 2x heatpipes do the lions share of the work, the 2x outer heatpipes seeing next to nothing since there's nothing under them but air on an intel die.

Yeah, so the better solution is to use a vapor chamber, as I believe Nvidia did with the GTX 1080 Founders Edition.

 

[the nerd 389]

take a cooler with 4x direct contact heatpipes. Pretty much the center 2x heatpipes do the lions share of the work, the 2x outer heatpipes seeing next to nothing since there's nothing under them but air on an intel die.

Yeah, so the better solution is to use a vapor chamber, as I believe Nvidia did with the GTX 1080 Founders Edition.

Vapor chambers are more costly, complicated to design, and not exactly appropriate for CPU coolers. Now, if Intel were to turn the CPU itself into a vapor chamber (i.e. the die would end up inside said chamber), that would be a different story altogether.

 

[dudmont]

after seeing the review i am thinking of using my car cooling system to cool down the intel new raging bull.

This could be the situation where setting up your rig to run out of a spare chest freezer makes sense again. :/

You'll be fighting condensation and the fact that refrigerators are made to run at a lower duty cycle. Plus, this could happen:

Yep, started by a refrigerator.

that certainly was quite the conflagration, sadly.
Run a custom look with only your radiators and fans inside a small freezer, I'm thinking a cheapy Danby would work rather well. Make sure you have some anti-freeze in your loop(is that possible?). Or am I out in left field on this.

 

[the nerd 389]

after seeing the review i am thinking of using my car cooling system to cool down the intel new raging bull.

This could be the situation where setting up your rig to run out of a spare chest freezer makes sense again. :/

You'll be fighting condensation and the fact that refrigerators are made to run at a lower duty cycle. Plus, this could happen:

Yep, started by a refrigerator.

that certainly was quite the conflagration, sadly.
Run a custom look with only your radiators and fans inside a small freezer, I'm thinking a cheapy Danby would work rather well. Make sure you have some anti-freeze in your loop(is that possible?). Or am I out in left field on this.

Or just run a vapor absorption refrigerator. They run just fine at 100% duty cycle, even with varying load. If you also cool-then-heat the intake air to dehumidify it, there's no issue running sub-ambient coolant temperatures. Unfortunately, you'd still have to deal with the insane delta-T inside the CPU. There's a point at which the thermal stress alone will break the CPU, and I have a feeling that it wouldn't take much to find out exactly what that point is with Skylake-X.

 

[alextheblue]

after seeing the review i am thinking of using my car cooling system to cool down the intel new raging bull.

Thermostat delete, reprogram the PCM to ramp the fans at a lower temperature... 😛

 

[FormatC]

About °C vs. Kelvin Wikipedia said:

In science and engineering, degrees Celsius and kelvins are often used simultaneously in the same article, where absolute temperatures are given in degrees Celsius, but temperature intervals are given in kelvins.

I'm a bit oldschool and using this since years to express the difference between both things.

 

[Crashman]

About °C vs. Kelvin Wikipedia said:

In science and engineering, degrees Celsius and kelvins are often used simultaneously in the same article, where absolute temperatures are given in degrees Celsius, but temperature intervals are given in kelvins.

I'm a bit oldschool and using this since years to express the difference between both things.

I think there might be a little confusion in the case articles if I labeled charts ΔTK 😀
I don't think simplifying the label to ΔK would make it any easier 😛
At this point any further ribbing might be a bit too much 😀 😀 😀

 

[bit_user]

take a cooler with 4x direct contact heatpipes. Pretty much the center 2x heatpipes do the lions share of the work, the 2x outer heatpipes seeing next to nothing since there's nothing under them but air on an intel die.

Yeah, so the better solution is to use a vapor chamber, as I believe Nvidia did with the GTX 1080 Founders Edition.

Vapor chambers are more costly, complicated to design, and not exactly appropriate for CPU coolers.

Well, they seem to be popular for cooling rackmount servers.

https://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=50001522&IsNodeId=1&Description=vapor%20chamber

There are a few enthusiast air coolers using vapor chambers, with the 3D models probably being most interesting (and pricey):

https://www.newegg.com/Product/Product.aspx?Item=N82E16835103208

Now, if Intel were to turn the CPU itself into a vapor chamber (i.e. the die would end up inside said chamber), that would be a different story altogether.

I had a similar thought. Maybe they're concerned about leaks, or maybe it's just too hard to find a good working fluid that's inexpensive, non-corrosive, and non-conductive.

 

[rantoc]

As suspected, intel rushed this product so the motherboard vendors didn't get the time to properly tune their motherboards. Its not helping either that the core architecture is aging making it almost netburst bad vs the competition and to make matters even worse the TIM approach vs solder made the product pretty much unfit due to extreme cooling needs to avoid throttling even during heavy non-overclocked loads.

Its clear as day that intel is shitting their pants since they got competition for once and they simply weren't ready for it resulting in rushing a product like this!

 

[cats_Paw]

I am wondering if the 18 core variant will even launch...

 

[bennie101]

Presscott otherwise know as PRESSHOT! Yep had one to.

 

[aldaia]

These 2 links show the size of the die and coolers needed:

https://www.lowyat.net/2017/133239/computex-2017-noctuas-amd-threadripper-cpu-cooler-massive/

http://www.pcgamer.com/amds-threadripper-is-huge-with-an-equally-large-socket-and-cooler/


Intel's I9-7900x has a tdp of 140w, with a 4.3ghz boost. Their die size is still roughly the same as that of the rest of their core lineup in comparison to Threadripper's monstrous size. The biggest mistakes Intel made was the thermal paste (as the author mentioned) and while not really a mistake; trying to cram too much into a tiny space for their socket.

What those links show is the size of the package. Next link shows the size of the dies compared to skylake-x
http://digiworthy.com/2017/06/22/intel-skylake-x-vs-threadripper-die-size/
16-core Theadripper die size (actually 2 smaller zeppelin dies) is marginally bigger than 10-core skylake-x dies, with the added benefit that its easier to get good yield from small dies.

 

[aldaia]

I am wondering if the 18 core variant will even launch...

Yes, rumor says it will be relased in the deep of the winter, but will only be available in Alaska, Canada, Scandinavia and Siberia. 😀😀😀

 

[bloodroses]

These 2 links show the size of the die and coolers needed:

https://www.lowyat.net/2017/133239/computex-2017-noctuas-amd-threadripper-cpu-cooler-massive/

http://www.pcgamer.com/amds-threadripper-is-huge-with-an-equally-large-socket-and-cooler/


Intel's I9-7900x has a tdp of 140w, with a 4.3ghz boost. Their die size is still roughly the same as that of the rest of their core lineup in comparison to Threadripper's monstrous size. The biggest mistakes Intel made was the thermal paste (as the author mentioned) and while not really a mistake; trying to cram too much into a tiny space for their socket.

What those links show is the size of the package. Next link shows the size of the dies compared to skylake-x
http://digiworthy.com/2017/06/22/intel-skylake-x-vs-threadripper-die-size/
16-core Theadripper die size (actually 2 smaller zeppelin dies) is marginally bigger than 10-core skylake-x dies, with the added benefit that its easier to get good yield from small dies.

That's for clarifying package vs die; I had used the wrong term. This is a very interesting read.

 

[motocros1]

anyone else get the feeling that intel has been sitting on this architecture for a while trying to milk every last drop out of its last tick-tock-tock strategy? seems like intel was waiting on a time to actually need this current architecture to battle with amd.

 

[mlee 2500]

I enjoy using my applications more than I enjoy trying to squeeze out an extra 4% of performance gains at the cost of stability, energy consumption, heat, and risk to my expensive hardware. All for gains I would hardly notice outside of benchmarks.

That being said, articles like this are useful even to folks like me, who overclock very little if at all, since running at the edge in a stock configuration, being susceptible to easy throttling, and compromising the value and return on a decent cooling solution (why bother in this case?), ARE all factors I take into consideration when building my machines.