Is there a possible Advisory? Skylake's more fragile CPU PCB substrate

[BonzaiDuck]

The mounting specifications are the same. It is probable that the newer substrate is weaker than the old one, but that just means that the old one exceeded the required strength by a larger amount than the new one. A cooler which was fine on the old processors but damages the new ones is a cooler which was grossly out of spec in the first place.

Somehow, the damage potential involves the tabs on the latchplate which bear down on opposing tabs of the processor cap/IHS. The force from the cooler mounting assembly bears down on the processor cap. And I'm still trying to figure out how putting a tapered shim between the CPU PCB and the latchplate, mating evenly with the lower lip of the processor cap, eliminates some or all of the risk.

ThermalRight says to use theirs; ASUS says to follow their manual and leave the "installation tool" in place. I'm not sure that the tech-reps answering my questions know enough of the particulars to explain it accurately. Especially, I'm more inclined to follow the ASUS recommendations.

Most of the better-performing heatpipe coolers in years since the copper TRUE exceed the spec, some more, some less. I'm inclined to think this is more a matter of a shipping and movement situation than simply installing the cooler and letting it hang on the board in a vertical tower case.

Anyway, this is all going to prove out one way or the other -- when I finally do a final check of electrical connections, plug it in, and fire it up for the first time.

 

[superstition]

Most of the better-performing heatpipe coolers in years since the copper TRUE exceed the spec, some more, some less.

And, apparently (according to what has been claimed here), Intel's CPUs exceeded the spec as well.

But, for some unknown reason, Intel decided that thinning the substrate is a good idea — despite the existence of so many popular and effective heavy air coolers on the market — coolers enthusiasts have invested considerable money in.

 

[Phynaz]

And, apparently (according to what has been claimed here), Intel's CPUs exceeded the spec as well.

But, for some unknown reason, Intel decided that thinning the substrate is a good idea — despite the existence of so many popular and effective heavy air coolers on the market — coolers enthusiasts have invested considerable money in.

When did you get your degree in Materials Science?

 

[Excessi0n]

So, the claim is that Intel changed the substrate thickness to make its product in-spec? It's been producing CPUs for years that are grossly out-of-spec?

No, I don't know why Intel chose to make the substrate thinner. The "claim" is that it doesn't matter because the new package is still perfectly capable of taking the maximum loads defined in Intel's mounting specifications.

Structures (whether it be a bridge, a ship, a building, or in this case a cpu package) are typically stronger in reality than what they're rated for. This is called a factor of safety, and it's a good thing. Thinning the substrate has presumably lessened the factor of safety by some degree, but any cooler which has stayed within Intel's mounting specs will cause no problems. Any cooler which places more load on the package than the maximum defined by the specs was dangerous to use in the first place.

 

[BonzaiDuck]

Well, they all exceeded the specs. I just revisited the Noctua website for the NH-D15. Without its fan (D15 S) or fans, the D15 weighs 980 grams or 80 grams more than the LG Macho. The Macho may (I'm not certain) have a center-of-gravity further from the processor cap. Either way, the spec was 500 gms of weight and 25mm distance from the processor cap. Apparently there are two sources of stress: the torque on the motherboard mounts, and the pressure of the mount at the processor cap. The greater the torque, the greater the necessary pressure to keep the cooler flush with the IHS.

Fellas -- Ladies -- the only way to get beyond this for me is to finish plugging it together and see if it fires up. But then the issue of whether it's good for the long haul.

 

[LTC8K6]

The specs changed. The substrate was thinned.

So, the only way it could be equally strong as before is if it's a different material, is reinforced with an additional material(s), or is somehow engineered with a stronger form factor.

None of those things have any data behind them, though, that I've seen.

The substrate is thinner, but meets the same specs and passes the same tests as the previous substrate. We can make parts out of different materials that are different thicknesses, but are the same strength. We can even have the thinner part be the stronger part if we want. You already know this, of course, yet you still say that Intel "weakened" the substrate...as if it's that simple.

 

[LTC8K6]

The way the socket is made, it's going to take a lot of torque to do what has been seen in the rare pics showing damage.

Most likely, that damage was not from the static load of the cooler.

 

[superstition]

Intel said they had not changed the specs.

The specs changed. The substrate was thinned.

So, the only way it could be equally strong as before is if it's a different material, is reinforced with an additional material(s), or is somehow engineered with a stronger form factor.

The substrate is thinner, but meets the same specs

So, the CPUs produced prior to Skylake were grossly out of spec?

and passes the same tests as the previous substrate.

Except for compatibility with coolers that were based on working with all those out-of-spec – the thickness spec – Intel parts with their too-thick substrates.

We can make parts out of different materials that are different thicknesses, but are the same strength. We can even have the thinner part be the stronger part if we want.

So, the only way it could be equally strong as before is if it's a different material, is reinforced with an additional material(s), or is somehow engineered with a stronger form factor.

I don't see what you've added.

You already know this, of course, yet you still say that Intel "weakened" the substrate...as if it's that simple.

Because we're seeing a level of problems with coolers that didn't exist before the substrate was thinned.

A thinner substrate is also questionable given the rise of the delidding practice.

Given the effectiveness and popularity of heavy air coolers, Intel should have made sure its CPUs are made to be compatible with them.

 

[BonzaiDuck]

The way the socket is made, it's going to take a lot of torque to do what has been seen in the rare pics showing damage.

Most likely, that damage was not from the static load of the cooler.

Also -- in recognition of Superstition's response . . . .

So far, I got the system together. Haven't loaded the OS, but I'm now testing memory. I resolved a problem with a CPU fan that wouldn't spin up for what was apparently a loose connection on the 8-pin "CPU-to-PSU" connector for ATX or EATX power. 12V, I think. Isn't that right?

Then, you find there are things that must be done for the PCI Express feature and the use of SATA ports individually. Particularly, an SATA optical disk has to be configured as "hot-swap" or "hot-plug," which in turn won't function without a power-efficiency feature, which must be enabled to also have your onboard controller recognize the external SATA optical drive.

At this point, all I see are the default settings and changes I made just to get the hardware to work, so the processor is running at a default 4.0 Ghz in BIOS monitor. The memory seems to configure easily as DDR4-3200. The temperature from BIOS monitor doesn't vary outside of 25C to 29C.

Whatever damage could be caused by the TR LGM (and adding in another 170gm Viper fan), it has not materialized so far. I believe this is a matter that requires a "shipping" situation, or just plain carelessness.

IF anyone is familiar with any of the Sabertooth Z170 motherboards, send me a PM. I may have some questions, if you have advance experience with this board.

I WOULD THINK that you could use either of two spacers or shims: The TR unit, or the one provided with ASUS boards. I simply installed CPU the way ASUS told me to, and put the TR spacer aside. Lord knows what this means for later CPUs or the implications.

 

[LTC8K6]

I have seen very few pics documenting any problems, and the pics I have seen were clearly not related to the thinner substrate. The whole board and socket had been warped.

And I would never expect Intel to test aftermarket coolers, or make changes to accommodate aftermarket coolers, or even care about ridiculously heavy and tall coolers.

It is the cooler mfgs responsibility to stay within spec.

It is very silly to ship a computer with a heavy cooler attached.

The problem is rare, as far as I am concerned, and limited to heavy coolers, dynamic loads, and tightening of cooler screws.

There is no problem with Intel's substrate, imo. I have no concerns at all.

If you are using a heavy and tall cooler to get that max overclock, then you need to use common sense when installing it, moving the PC, or shipping the PC.

 

[crashtech]

The real problem comes with delid/relid, since after the perimeter adhesive is removed, the die itself is taking the entire load of the retention mechanism and any load transferred from the cooler to the heat spreader. The sealant is load-bearing, so using a shim between the latch mechanism and the PCB is very prudent after delidding Skylake.

 

[BonzaiDuck]

The real problem comes with delid/relid, since after the perimeter adhesive is removed, the die itself is taking the entire load of the retention mechanism and any load transferred from the cooler to the heat spreader. The sealant is load-bearing, so using a shim between the latch mechanism and the PCB is very prudent after delidding Skylake.

According to Silicon Lottery, they -- he -- reseal(s) the IHS with the same adhesive. If SL notes that the cap (IHS) is closer to the die for more even or thorough transfer of heat through the CLU used to replace the original TIM, they are also confident about how precisely they prepared the IHS.

I'm only a tad uncertain about which is the better correction for any vulnerability: the TR shim, which causes the latch-plate spring to bend too much; or the ASUS "Installation Tool," which the user is instructed to leave in place before closing the latch. ASUS says "follow our directions;" TR tech-support says "use our spacer."

If I used their spacer, I'd want to lap it down to a point where the latch-spring has the right tension, as opposed to too much.

 

[crashtech]

The configuration of the shim would be critical, for sure. Properly designed, it would apply a more consistent pressure to the outer lands, possibly alleviate excess pressure from the die and center lands, and tend to stabilize heat spreaders that are not re-attached to the PCB.

 

[BonzaiDuck]

The configuration of the shim would be critical, for sure. Properly designed, it would apply a more consistent pressure to the outer lands, possibly alleviate excess pressure from the die and center lands, and tend to stabilize heat spreaders that are not re-attached to the PCB.

I didn't like it when the latch-spring bowed like a tree in a hurricane. So I replaced it with the ASUS "installation tool," set e-mail to TR and ASUS for advice. TR says to use their spacer; ASUS says to follow the directions in their manual, to leave the installation tool in place when you close the latch-plate and lock the spring.

You cannot install both shims in that socket at the same time, or Lord knows what will happen to that latch-spring or maybe even the motherboard.

Does anyone have any information about boards from MSI, Gigabyte, AsRock, Biostar, EVGA etc.? For that you know, for instance, does the board bundle include an "installation tool" -- something like a thin, pliable plastic frame that clips onto the processor when you're ready to drop it in the socket? If ASUS is being cautious regarding users' heatsink applications, TR says to use their tool which makes the latch-spring bow.

I may try and lap off some plastic from one of the three TR spacers they sent me.

But the system is already set up and running with the LG Macho and the ASUS Installation Tool . . .

 

[Arachnotronic]

I've built a ton of Skylake systems with a range of coolers and I haven't managed to hurt the PCBs of a single one. And I am pretty much a clumsy brute.

 

[crashtech]

I've built a ton of Skylake systems with a range of coolers and I haven't managed to hurt the PCBs of a single one. And I am pretty much a clumsy brute.

But I wager you haven't built a delidded one with a large air cooler for somebody who doesn't realize how carefully it must subsequently be handled. This is where the potential trouble lies, but it's not any responsibility of Intel's.

 

[Arachnotronic]

But I wager you haven't built a delidded one with a large air cooler for somebody who doesn't realize how carefully it must subsequently be handled. This is where the potential trouble lies, but it's not any responsibility of Intel's.

True, I'd never mess with delidding.

 

[sxr7171]

AIO will change your life.

 

[sxr7171]

But I wager you haven't built a delidded one with a large air cooler for somebody who doesn't realize how carefully it must subsequently be handled. This is where the potential trouble lies, but it's not any responsibility of Intel's.

Exactly, if you intend to move the PC you're really risking it. Even putting it in car and driving around could damage it.

 

[BonzaiDuck]

True, I'd never mess with delidding.

Nor would I, until I could pay Silicon Lottery to do it with CLU and provide a reasonable warranty for returns after busting the Intel warranty. either the person behind SL or their tech-rep seemed confident for what I provided as a description of the cooler. He might otherwise have urged water-cooling, but he didn't. He only warned of a difficult bare-die mounting for even a waterblock requiring permanent changes to the height of the socket -- a forbidding prospect if it requires something like a dremel.

Now perhaps this thread could fit on Cases and Cooling (or I already said that). So far, I've only run the processor at a turbo-speed of 4.2 -- the default, but with all cores synched.

The mobo makers seem to be overvolting these processors needlessly with their "auto" settings. Secondhand, this was confirmed by another poster at Tom's who spoke to
asus about it. The motive is simple and sloppy: They want to just make sure the board works with as few problems as possible out of the box.

So I dropped the voltage so that peak turbo voltage under XTU or OCCT doesn't exceed 1.35, and the base 4GHz voltage is about 1.29V.

This is still way to high according to what I've read, but the stress temperatures don't go over 47C (peak package) or an average around 45C.

So on to the point about AiO. I can probably squeeze another 5C in cooling from a Predator 240 -- maybe 6 or 7C with an H240 X2. But with temperatures as low as I expect, and temperatures I've seen so far, it hardly seems worth it. The cooler seems more compact than others with weight on the high-end of performance. It is built so the front face of the fins aligns vertically with the forward edge of the latch or heatsink base.

$80 for the cooler, $50 for the delid-relid. The price-bump for binning the processor doesn't figure if the cooler and the motherboard do their parts. I might have spent up to $200 for either the EKWB or Swiftech -- less for an H115i. But you'd still have to add the $50 to any of those choices. Is 5+C worth the extra $50 to $120 between the Macho and these other coolers? At the temperatures I should expect, I'll have as long as I wish to think about it.

I have some reading to do, must set up my notebook, and begin to find the lowest settings. I think a 1.2V base VCORE and an adaptive 0.1V increase from there to 1.3V buys an easy 4.4 Ghz. In fact, I probably need to run the AI Tweaker-overclocker-whatever from BIOS to test both modes "air" and "water." Just to see what is possible with sloppy volting.

 

[crashtech]

My Skylakes seemed to like lots of voltage when cranked up past 4.4GHz. It's not the same as Haswell or even Ivy. The way they like the volts reminds me more of Sandy, really.

 

[coercitiv]

The mobo makers seem to be overvolting these processors needlessly with their "auto" settings. Secondhand, this was confirmed by another poster at Tom's who spoke to asus about it. The motive is simple and sloppy: They want to just make sure the board works with as few problems as possible out of the box.

They are indeed overvolting, and not just Asus. The way I used to determine a more appropriate stock voltage for my Skylake was to switch voltage from "Auto" to the smallest offset available: the result was a significant drop in voltage, something like 0.05V, even though using a positive offset should actually increase it. That can only be explained if the "Auto" settings are not representative of stock values at all.

I think a 1.2V base VCORE and an adaptive 0.1V increase from there to 1.3V buys an easy 4.4 Ghz.

Voltage for 4Ghz should be in the realm of 1.15 - 1.17V so setting 1.2V is a safe value, for oc voltages you might want to take a look at the table in this overclock.net thread. (direct link to google docs here)

 

[BonzaiDuck]

They are indeed overvolting, and not just Asus. The way I used to determine a more appropriate stock voltage for my Skylake was to switch voltage from "Auto" to the smallest offset available: the result was a significant drop in voltage, something like 0.05V, even though using a positive offset should actually increase it. That can only be explained if the "Auto" settings are not representative of stock values at all.


Voltage for 4Ghz should be in the realm of 1.15 - 1.17V so setting 1.2V is a safe value, for oc voltages you might want to take a look at the table in this overclock.net thread. (direct link to google docs here)

I'm actually astonished at your post, for its timeliness -- even "clairvoyance" -- to what I'd been doing with this thing. I was actually thinking to start another thread to just . . . ask . . . the question:

"What is the minimal stable voltage for a top-17% Skylake at stock settings and for either no LLC or "Level 3" (out of 8)? Or what is the respondent's choices for fixed "voltage override" and LLC?""

And I think you answered my question. Because -- I found instability at around 1.152V after 17 minutes of OCCT:CPU. So I set the VCORE in the Manual regime to 1.200V after lowering LLC to 3 and losing video (I had to cold boot the system, and I had to cycle it twice.)

1.152 is the drooped voltage minimum during OCCT. I think that LLC Level 3 now shows a stable value closer to 1.168V. So I think the vDroop is somewhere around 48mV.

At this voltage setting (passed 1hr 10min so far):

_____________MAX_____MIN______Prevailing_AVG

VCORE_____1.200V___1.152V____________1.166V

C-deg (pkg)____47C______40C______________42C

I'm not plugging the ThermalRight LG Macho [LGM], but this looks like a good start on the road to 4.7, at most 4.8Ghz. Somehow I think I should still be impressed with the delid/CLU-relid results. I've seen forum posts with folks reporting temperatures for stock-stressing, and the temperatures were higher by at least 10C to 20C, depending on the cooler. The 10C was maybe reported from someone using a Seidon water-cooler (or was it an X61?). I have an IBT 3470 (non-K) system (stock settings) which reaches ~65C with a 212EVO.

I chose to find this minimum [stock-custom-overclock] because I need a place from which to start, and I wanted to assess the cooler to better follow progressive temperatures when raising multi and voltages. I also need the best fall-back for stable operation.

But someone could also respond to another question: What do you think of those temperatures at those voltages for stock 4.0 Ghz to Turbo 4.2? The CPU die hasn't cracked, everything is working. And this is a freaking . . . . air-cooler . . . .

FORGOT TO MENTION: My "experiments" tell me that "Auto" for LLC at the default speed would seem to be either at Level 5 or Level 6 LLC. Some OC guides want you to shortcut directly to Level 5. I'd rather put it at level 3 for the "stock-but-tweaked" settings I'm looking for before I start raising the processor speed and vcore again.

 

[NeoPTLD]

Somehow, the damage potential involves the tabs on the latchplate which bear down on opposing tabs of the processor cap/IHS. The force from the cooler mounting assembly bears down on the processor cap. And I'm still trying to figure out how putting a tapered shim between the CPU PCB and the latchplate, mating evenly with the lower lip of the processor cap, eliminates some or all of the risk.

Center of gravity is everything. Your bicycle can obviously support your weight, but you'll bend the wheel if you stick one of the bike wheels in a groove and lean over. If you had a pair of training wheels, it will limit how far you can lean over before one of them comes into contact and prevent you from leaning any further.This should give you an idea how the shim works.

 

[BonzaiDuck]

Center of gravity is everything. Your bicycle can obviously support your weight, but you'll bend the wheel if you stick one of the bike wheels in a groove and lean over. If you had a pair of training wheels, it will limit how far you can lean over before one of them comes into contact and prevent you from leaning any further.This should give you an idea how the shim works.

Yes. Just in handling the cooler, I imagined the center of gravity exceeded the distance to IHS I had imagined with other coolers. But the board doesn't bow after cooler installation, and there's not a lot of torque to put on the screws holding the HS base against the IHS. I just have to decide whether to attempt a DIY lapping of the ThermalRight spacer so that I don't feels so uncomfortable over the bow in the latch-spring.

It does its job, though. All the core temperatures vary by no more than +/- 3C and are mostly in a tight grouping. My last OCCT:CPU test at 4.4Ghz with RAM at 3200Mhz gave a peak pkg temperature of 53C.

So this balances the risk. You have to be extra careful moving the system, as opposed to a fetish about not moving water-hoses -- if that would pose any risk for a fitting coming loose or a leak. There are advantages to either the air or water approach, but what happens if your air-cooler and processor-purchase yield temperatures better than various AiO coolers? This could almost seem to be a dilemma, unless I admit that I may spring for the water kit just out of curiosity. It may still be the EKWB, or than again possibly the H240 X2.