"How d'ya like it?": Skylake package temperatures @ 4.6 Ghz

[BonzaiDuck]

I'd debated how or whether or when I would post this thread. After all -- Skylake overclockers seem to fall into two categories: people afraid to get their feet wet because they are understandably new to "the business;" and folks with experience with DC and Haswell, Ivy or Sandy Bridge -- a few holdouts from Nehalem who've take the upgrade plunge.

As Aigomorla or any veteran of the case and cooling forum will tell you, I've "looked" at everything from Rube Goldberg evaporative cooling possibilities to TEC water-chillers, AiOs and custom-water kits. But I've never pulled the string. This actually seems more weird for me, because I should've done it by now.

And for a year's time, I was planning to do it this time with the Skylake. I had my eye on the EKWB and Swiftech "customizable" AiOs, and even the Corsair H115i. I was even leaning toward a true EK "kit" that would fit my case.

Water-cooling isn't just more efficient: it's part of an art-form. What's involved? a waterblock in good working order (that doesn't leak); hose fittings (which don't leak); a radiator; a pump; a reservoir. By my thinking, this adds complexity to a cooling problem solution that doesn't exist with less effective, less efficient cooling strategies.

So what did I do? I figure I spent an extra $80 for a binned Skylake, but I also made sure that Silicon Lottery de-lidded and re-lidded the processor with CLU according to their promises for $50 extra.

I suspect from my correspondence with SL that they are particularly good at this. Their tech-support was able to explain to me exactly how the re-lid turned out, whether they had further lapped the flanged base of the IHS, whether they had done a little lapping on the underside of the IHS, and whether they were confident that the "resealing" with the same silicon rubber as factory-fresh Intel chips would be reliable with either air or water.

I had prepared my case so it would fit either a Predator 240 or an H240 X2. But at the last minute, I did some cross-references between several reviews that included one or the other of these AiO "customizables," and discovered this -- another heatpipe cooler:

http://www.newegg.com/Product/Produ...ThermalRight_Macho-_-9SIA7WF4NC0962-_-Product

Let's call it the TR LGM. If I don't like their name for it, it's because of a possibility it has marketing appeal to folks of a certain Neanderthal mindset.

On the up side: it's more compact than an NH-D15 or other double-tower options. For about $8, you can buy an accordion duct to place between the cooler and your case exhaust. It's promoted as a "passive" cooler, and the benchmarks show that "passive" is only 5C warmer than "semi-active" with a limp exhaust fan. You can hang a fan on it, and the front edge of the cooler's fins vertically line up with the edge of the socket latch-plate: they don't hang over your RAMs, when hanging a fan on the cooler further complicates that problem.

On the down side: it weighs in at 900 grams without the weight of a 160 gram fan. But that is 80 grams lighter than a fanless NH-D15. We've been hanging this sort of weight on our motherboards for several years now. However, Skylake apparently has a thinner PCB. While there were some advisories and "pictures" suggesting that these heavier coolers might warp the CPU-PCB, there are no substantive reports of it. It is a "shipping, transportation, or recklessness" situation. TR issued a shim or spacer that fits between the CPU and the latchplate; the spacer would further assure that users couldn't damage the processor in its socket by moving their computer around with some modicum of carelessness.

But on to the races . . . . How does a CLU relidded i7-6700K behave with this cooler?

For comparison, here's a Skylake overclocking review for a factory-fresh retail-box 6700K, using a dual-fan external EXOS water-cooling kit:

http://www.hardocp.com/article/2015...76700k_ipc_overclocking_review/7#.V-_7SsIVCUk

If you're impatient for the reading, skip down the page to the last screen-shot and paragraph that follows: "The max temperature values of 76/74/79/82C are something get excited about."

Here are my own screenies for Intel Burn Test, Prime95 and LinX respectively:

[Still have to install my licensed screen capture after removing from another system, so sorry about Winsnap's free-trial watermarks.]

Following overclocking guides for the voltage-timid, it took me a little time to adjust so that either LinX or Prime95 would run at length without error. Now I can pass 30 iterations of maximum problem-size and memory with Linx, when I was beginning to believe the advice "don't use Prime or LinX."

And there you are. I have preliminary settings for 4.7 Ghz, and the temperatures only edge up a few more C degrees and nowhere close the review results with the water-cooler. I suspect, if I can't reach 4.8, it will derive from both my motherboard limitations and my reticence about volting above 1.4V.

You can delid and relid yourself, but I figured it was worth the $50 against the time, trouble and anxiety -- having it done by someone who has done it many, many times. I once replaced the tranny on my Honda by myself: it took me a day's time. I later had a mechanic do the same thing with another Honda. It took him less than 2 hours.

I could get another 5C improvement with the EKWB Predator 240, but why bother?

It means I have to drain, flush and bleed the system periodically.

It means that there's always some small risk of leaks (another thread here addresses a member's frustrations and an RMA-in-progress.)

It means that the pump on the unit might fail in 5 or 6 years, only as a matter of expected risk accepted by folks who water-cool.

It means that I don't get much of a "bling" boost for my "computer-as-art," because at least with the Predator, there's no visible reservoir and the hoses are black. With the Swiftech and its tinted water, it's not visible through the aluminum drive-cage wall anyway.

It means that I'm less flexible in how I can fit things into my midtower front-panel.

I can explain further what I intend to do with my video dGPU. For this build, I chose a "mini-OC" GTX 1070. So far, it warms up to a maximum 57 C under Unigen benchmarks and stressing. But I've got further plans for it that don't involve water. And I'm already wondering if I really, really need two of those 1070 jobs in SLI.

My hat is off to all the accomplished water-cooling aficionados. And -- I say AND! -- I'll go that route for sure if I ever build a six, eight or ten-core "E-machine," but I'd come to a conclusion that I can't use enough of all that processing power -- thus, Skylake's K chip. Of course, there's no need to delid an E processor, either.

DISCLAIMER: I don't work for ThermalRight or SiliconLottery. But many thanks to their tech-support via e-mail for their advice, insight and assistance.

FOOTNOTE: The parts inventory

Sabertooth Z170 S motherboard
TridentZ DDR4-3200 14-14-14 2x8GB 16GB kit
ADATA 480GB SP-550 SSD[temporary, until the budget plan says to trigger the checkout button for a Sammy 950 Pro M.2 card]
Fans -- whatever I already had for minimizing purchases: 2x Akasa Viper 140 "square", 1x Viper 140 "round," 2x Shark 140mm, 1x Noctua iPPC 3000 (120mm), 1x Noctua 40x40x10 mm max-5000rpm. the plan is to remove both the Sharks and replace with 1x Corsair mag-lev 140. Test to see if I can get rid of the CPU 140 "round" pusher fan. Then I may add a CM barrel fan and a duct I've mostly already built.

No doubt about it, though. The Saber Z170 S runs cool, and you can practically monitor the temperature of every spot on the motherboard without using any of the bundled two-wire thermal sensors. But only 8+4 phase-power-design. Not 16-phases like Maximus or Deluxe. Lose 100 Mhz. It's as good as any of the gaming board models.

Compromises will be made . . . .

 

[ClockHound]

Congrats on your new system Bonzai!

Temps look pretty good. But, remember to drain the air and replace it with boutique French Alps mountain air every three months. Sure, you could use room air, but that's for the average cheapskate user. Enthusiasts deserve the good stuff. ;-)

 

[BonzaiDuck]

Congrats on your new system Bonzai!

Temps look pretty good. But, remember to drain the air and replace it with boutique French Alps mountain air every three months. Sure, you could use room air, but that's for the average cheapskate user. Enthusiasts deserve the good stuff. ;-)

Somewhere just now, someone asserted "stability" at 4.8 Ghz with an H100 cooler and 70C. The reported voltage is about 20mV lower than where you'd expect it to be to do some honest stressing. All the other signs I see suggest that the temperatures with that cooler would be higher. But folks use all sorts of stressing programs, and the folklore says to avoid the Prime95 and Linpack tests, and it always boils down to heat if you investigate further. Then again, maybe the assertion on the forum post (different site) failed to mention that he, too, had a re-lidded CLU CPU. Or maybe he was just using RealBench, OCCT:CPU and XTU.

Or maybe he just has a better motherboard. I suspect, though, that I'll find out some more good stuff as I continue to play with it. My 4.6 voltage for rock-solid stability is somewhere between 1.35 and 1.38. LLC would make the VID shown in the screenies very close. I guess the only way I'll find out in a general sense of it is to tweak my 4.7 settings a bit more and then see how it shakes out.

But if I wanted to keep my processor cool at a 4.6 overclock, I've more than succeeded, I think.

Should I use the Lavender or Wintergreen air freshener?

 

[aigomorla]

Should I use the Lavender or Wintergreen air freshener?

nonono... u need to use "squash" because its proven to make things go faster like cars for example.

All the j-import cars had that squash carmate air freshner inside it and they were really fast tuner cars.

(sarcasm)

lol...

 

[BonzaiDuck]

nonono... u need to use "squash" because its proven to make things go faster like cars for example.

All the j-import cars had that squash carmate air freshner inside it and they were really fast tuner cars.

(sarcasm)

lol...

well, for the time being, I can't feel short for the temperatures from the CLU-CPU mod.

One could wonder, with processor TDP and other improvements, if TEC coolers will suddenly reappear. I think they were feasible when the TDP was under 70 or 80W. But feasible might not mean "stunning performance."

 

[aigomorla]

well i think i have said this a lot lately with the new tech.

were at a point now where its not so much heat thats holding people down, but too much voltage required.

I have killed too many cpu's to count on voltage more so then heat.

When a chip gets too hot, it will throttle, letting you know its too hot and needs a better cooling system.

When a chip gets too much voltage, it will die up right without any warning, or foreshadowing.

 

[BonzaiDuck]

well i think i have said this a lot lately with the new tech.

were at a point now where its not so much heat thats holding people down, but too much voltage required.

I have killed too many cpu's to count on voltage more so then heat.

When a chip gets too hot, it will throttle, letting you know its too hot and needs a better cooling system.

When a chip gets too much voltage, it will die up right without any warning, or foreshadowing.

Your vast experience in these matters had become a joke in a member's sig, on the topic of buying a used chip from you.

This chip was binned for 4.8. After I placed the order, I wondered how it was done. Did SL simply select new retail chips at random, test them to that speed and a voltage (<= 1.424), and set them aside for sale under that verification? How many of those chips binned for only 4.6 or 4.7 would still run at 4.8? Hard to say.

But if I want to run it at that speed, voltage is going to be somewhere between 1.41 and 1.43, with VID reaching even a tad higher. And the only guideline we have now is how mobo makers volt the chips. Even then, you're running it out of spec even if peak unloaded voltage is less than that.

Of course, if the chip went south, you could always replace it if no other components went along for the ride.

Then there was the white-paper quoted to say that Intel had improved chip manufacture to make them more "resilient" to the effects of electro-migration and so forth. How much is anyone's guess unless you're inside Intel.

The CLU-relid is really a game changer though. I couldn't get goosey about Ivy or Haswell because of the TIM news, but no less couldn't bring myself to go socket-2011 instead. Without it, if you want 4.7 Ghz, a person is definitely going to incline to a water system -- AiO or custom.

5C degrees is going to matter when the package temperature is pushing 80 or higher. Doesn't mean so much if it would bring you from 70C under LinX stress down to 65C.

Remember Lepton? Haven't seen posts from him lately, but we pass e-mails. He had a leak with the configuration for that MO-RA-3 or 4. Lost a graphics card. But that monster E processor of his never gets beyond the high 40's C.

 

[VirtualLarry]

My (lacking) experience, is that 1.4+ V is somewhat dangerous for Intel CPUs, long-term. At least, since 45nm Core2. A (late) friend of mine, had an E5200, that I had built and OCed for him, for gaming, and he was doing DC for me in the background. We had it at 3.75Ghz and 1.425V in BIOS, which, under load, in CPU-Z, showed up around exactly 1.40V or so. Which, was "rumored" to be the "max safe voltage".

Well, six months later, we re-qualified the OC, and it failed Prime95 at current settings. Had to drop vcore and clocks down a bit, to become "Prime stable" again.

I can't help but imagine that the issue was CPU degradation, due to excessive vcore, over time. Temps weren't bad, but eventually, dust clogged up the tower heatsink we were using, and he was getting BIOS beep tones during gaming. Well, that was the overtemp alarm.

 

[BonzaiDuck]

My (lacking) experience, is that 1.4+ V is somewhat dangerous for Intel CPUs, long-term. At least, since 45nm Core2. A (late) friend of mine, had an E5200, that I had built and OCed for him, for gaming, and he was doing DC for me in the background. We had it at 3.75Ghz and 1.425V in BIOS, which, under load, in CPU-Z, showed up around exactly 1.40V or so. Which, was "rumored" to be the "max safe voltage".

Well, six months later, we re-qualified the OC, and it failed Prime95 at current settings. Had to drop vcore and clocks down a bit, to become "Prime stable" again.

I can't help but imagine that the issue was CPU degradation, due to excessive vcore, over time. Temps weren't bad, but eventually, dust clogged up the tower heatsink we were using, and he was getting BIOS beep tones during gaming. Well, that was the overtemp alarm.

That sounds like my E8600. Did you have it set to the maximum voltage without the energy-saving features? I think I had it clocked to 4.2 Ghz, or maybe it was 4.3. And I had to "re-qualify" as you did. I can't even remember the voltage: it may have been something like 1.42V.

We dropped it back to stock, because -- as Intel says themselves, I think -- it will continue to run properly in its slightly-degraded state at stock settings. Lemme-think: I built that system in 2008, and it went to my brother at stock settings around 2010. There was nothing more wrong with it when we retired it in 2014.

Like I may have said, I've so far had no problem with my 2600K running at 4.6/4.7 since it was built. I won't run it at turbo voltages above 1.36 to 1.38.

So here's what we have with Skylake. There is no "safe voltage" spec. Mobo makers are shipping boards which default on auto to maximum turbo voltages of 1.39V. "Skinny-on-the-street" from some quarters say 1.45V is a prudent limit, and from other quarters -- 1.40V.

Then there was a recent white-paper somewhere, with facsimile excerpts provided in some review. Improvements in chip design and manufacture have made the Skylake "more resilient" to degradation.

So . . . . people will take their chances. But if Prime95 at 4.6 is pegging my voltage at around 1.35V, I shouldn't worry as much as I might for another 100 Mhz.

Tell the truth -- I had simply assumed for some time that a "safe" voltage would follow the die shrink in some proportion. That may be as naïve as thinking that 1.42V is an OK thing.

One could say "gee -- this Skylake is great at 4.2, 4.3 . . . " whatever and stop pushing the envelope. But I don't avoid shopping errands for fear of a terrorist attack. I just don't want to pay for too much gasoline . . . .

And I'd hate to have a CPU die in a year's time. And that's not a pun, either.

 

[superstition]

I can't help but imagine that the issue was CPU degradation, due to excessive vcore, over time. Temps weren't bad, but eventually, dust clogged up the tower heatsink we were using, and he was getting BIOS beep tones during gaming. Well, that was the overtemp alarm.

Higher temp means higher voltage requirement to stay stable. It's not surprising you had to drop the clock. Also, if the temp went higher that could have, along with the voltage, caused degradation.

People building overclocked systems for others should put in a "dust cushion" into their voltage/heat spec because other people should always be assumed to be more lazy about cleaning than you are.

I would also have a "dust clause" in the contract that states that if I find a clogged system when I inspect it later all warranty on the CPU and board is void — and it's up to me how to proceed. That would at least get people to clean their system before asking me to look at it and that might do the trick (fix the stability issue) and spare me the hassle of dealing with it.

 

[BonzaiDuck]

Higher temp means higher voltage requirement to stay stable. It's not surprising you had to drop the clock. Also, if the temp went higher that could have, along with the voltage, caused degradation.

People building overclocked systems for others should put in a "dust cushion" into their voltage/heat spec because other people should always be assumed to be more lazy about cleaning than you are.

I would also have a "dust clause" in the contract that states that if I find a clogged system when I inspect it later all warranty on the CPU and board is void — and it's up to me how to proceed. That would at least get people to clean their system before asking me to look at it and that might do the trick (fix the stability issue) and spare me the hassle of dealing with it.

My remark here is tongue-in-cheek, but perhaps has some relevance.

Intel dropped publishing the spec for "safe voltage," but they still publish TCASE. Some of our illustrious members, far more keen on this subject than I, have dismissed the spec, because it's a "case and cooling" guideline for OEM makers.

Back in the Conroe/Kentsfield days, you could expect TCASE -- theoretically the temperature at center of the processor IHS -- to be 10 to 15C less than the core temperatures. I think the discrepancy narrowed over the later chip generations. Let us say that it should be <= Tj core temperature values. Or ~ Tj, but not greater than. You can't measure it yourself unless you cut a slot in the IHS.

Anyway, something of that item should play into the "dust cushion."

 

[monkeydelmagico]

Very nice rig and write-up BonzaiDuck. Thank you for posting it.

 

[BonzaiDuck]

Very nice rig and write-up BonzaiDuck. Thank you for posting it.

I'll be an unabashed chatterbox.

After I make up my mind to continue with the original plan and reinstall a CM Crossflow barrel fan and a Lexan duct I built (now needs modification), I'll post a picture. The interior is clean -- bare.

Maybe you saw the South Park episode where some Good Fairy Godmother tells the South Park boys that they must "find the clitoris." I think I've found the modestly-expected sweet-spot at 4.6 Ghz.

The problem with all the literature you find -- including geek forum posts -- is a lack of clarity about how the poster or writer defines a vCORE reading, whether they've confused it with VID, whether they're referring to a "set" value in BIOS or the monitored value, whether they mean unloaded maximum voltage or loaded and drooped voltage. But for comfort, my "set" value is 1.350V Adaptive "for Extra Turbo" voltage. the multiplier is 46, BCLK 100, DDR 3,200Mhz. This results in a monitored value maximum closer to 1.39, minimum 1.344 and prevailing-at-load 1.376V. VID is always maybe a dozen or so millivolts higher than monitored values.

If someone tells you that they binned your chip for 4.8 Ghz and <= 1.424V (quoting them: "maybe less!"), that they verified it with 1 hour of ROG RealBench on a Maximus board, that means a modicum of stability achieved with that mid-range stress-test. To validate with LinX, and Prime95, it's best to attend to cooling -- any whicha-way you can, with the well-discussed tradeoffs. I think the fine-tuning will result in higher voltages than required for RealBench. But the skinny on the street -- and there's lots of it to compare -- has a rumor that "Intel said" the maximum safe voltage is 1.45V. Several other sources simply advise 1.40V.

The chip is supposed to be more resilient to the electrical degradation than prior generations, yet the traces are smaller and closer together. The possible effects of thermal degradation would seem to loom greater above some temperature. And oddly, since it was only my novice opinion, I also found yesterday a suggestion to avoid running much above the TCASE spec. Which -- I think -- was 68.5C.

All I've done -- for myself or anyone who gave their attentions -- is to prove you can do this with a heatpipe cooler if the main objective is simply to improve cooling in any way. So you'll either make with a razor-blade and mallet, not knowing for sure what you're getting into, or pay somebody well-practiced in doing it.

Unless you use "Big Water," or find space in the case for two radiators, water options may mean only a 5 to 10C improvement. By comparison, a Predator 240 should run 5C cooler than the TR LGM. But there's a range of possible results from a CLU-relid, and mine meant a 12C improvement. I'd spend another $100 over the LGM's price to make that 17C.

The voltage is a matter of risk-averse behavior or accepting a possibility that you'll spend money on a chip replacement in a year's time. I've read enough, tried enough and discussed enough to think that 4.6 Ghz is as good as it gets for me.

A Testing Note: to see what stressing will do temperature-wise, here are the thermal results for a few that are more than familiar: Affinitized LinX largest problem size and memory -- 73C; Intel Burn Test -- 67C; Prime95 Blend -- 68C; OCCT:Linpack -- 68C. OCCT:CPU was a little lower, I think.

XTU and Aida64 are limp. Temperatures of 60C or lower; uncertain error detection. RealBench is somewhere between the two groups of tests, I think. I'll have to check again. . . .

Hmmm. RealBench, 68C.

And Room Ambient -- 80F. That's what the thermostat says, anyway.