6700K & XMP Voltage Query

[ithehappy]

First of all my system is in my signature. You see that I have a G.Skill RAM (F4-3000C15D) which is basically 3000 MHz but as motherboard can't support it natively I am forced to use the XMP profile. Now my board is Asus Z170 Pro, and when you select XMP from Ai Tweaker you are basically given a pop up message saying whether you want All Core Enhancement, or Stock Intel settings, I was advised to use the former, so was using it without any problems at all for this long, but yesterday thought about checking temps and all, and then I found that the max temp was hitting 75-74ºC, which I found a little bit on the higher side. Then I used the Stock Intel settings from the XMP profile, and the max temps were reduced by ~9-10ºC, which is a lot. But then I disabled XMP altogether and the max temps were even more low, ~5ºC, from even Intel settings. The app used to stress the CPU was something called HeavyLoad.

Now the question, when I enable All Core Enhancement it does something to the CPU as well, I don't know what, but I see that the DRAM voltage is set to 1.35V automatically. Then when I don't use that option and use the Intel stock option, even though the temps are much lower, I still see DRAM voltage at same rate of 1.35V. And when I disable XMP then I think the voltage is 1.2V, but I am not sure about that.

So the question is, I want to use my RAMs at 3000 MHz obviously, but I want to maintain that stock like temp as well, is there any way to do that? I know the most obvious shot is to reduce the voltage, but to what extent? I am ONLY looking to OC my RAMs, that's it, I do not want to touch any CPU settings if that was possible.

Now here are the temperature screenshots from different settings to give you a detailed idea about what nonsense I just written above.

Full Stock Settings (XMP Disabled, RAMs run @2100MHz):

XMP Enabled, Stock Intel Settings:

XMP Enabled, All Core Enhancements On:

These are BIOS screenshots, when XMP profile with Intel Stock option is saved from full stock there are the changes made, I think,

And this is the DRAM voltage of 1.35V I was talking about,

I use a Noctua U12S cooler, and the room temp was 30-30.5ºC while these tests were done.

 

[BonzaiDuck]

I'm off to a good start in building my own i7-6700K with an ASUS Sabertooth "S" model. I'd been watching threads and posts here on this basic Z170 Skylake-K theme for overclocking and cooling -- maybe for a good part of a year before I "executed my Great Plan."

Like you, I chose air-cooling. On the other hand, I think of a system BIOS as my own wading pool of discovery. So based on what I'd read and what I tend to do, I estimated a cooling-solution strategy in terms of the temperatures, risks -- pros and cons. I compared reviews on the coolers so I had a rank-ordering of choices by performance using similar test-beds.

Your U12S is probably reasonably adequate for your system. Or -- your case ventilation could be improved. Intel had begun producing processors without indium solder since Ivy Bridge, so as the dies shrunk, they cheaped out on production. They also say they were worried that indium solder would crack the die as temperatures were raised and lowered. So they started with some cruddy gray stuff, which was replaced in Haswell and therefore Skylake with a polymer.

The polymer is still inferior to indium solder, if it's still better than what they started with.

Intel has a processor spec often sneered at here on the forums, but it was a simple "guideline" to OEM builders for choices of case-design and ventilation. Every processor has a TCASE spec, and with the die shrinks with the same range of thermal design power, the spec has decreased. So for Skylake, TCASE is ~ 68.5C if I'm correct. For this they are saying that this is the recommended maximum for any particular case and ventilation regime assuming the system were running 24/7. But it's not the maximum for either safety or performance with the "Turbo" and "energy-saving" features of the processor.

I have chosen a ThermalRight LG Macho air-cooler, when I'd been "investigating" and plotting a water-cooled adventure for years. I chose to pay maybe $80 to Silicon Lottery to sell me a processor tested at 4.8 Ghz, and another $50 to remove the CPU heat-spreader, replace the TIM with a gallium-based liquid, and then reseat and reseal to processor cap.

So I've run up the ladder of overclocks. I can reach 4.6 Ghz with extreme stress temperatures of 64C, while a reviewer with a CPU straight from the factory and an EXOS water cooling system was reporting ~ 79 to 80C at 4.7 Ghz. And I can GET to 4.7, but haven't had time to get the thermal numbers I think are reliable.

In your case, you've only chosen to accept the temperatures from OC'ing the processor to "on-all-cores" as opposed to its default setting. It shouldn't vary much from what you were seeing at default settings.

IN FACT, the motherboard makers are taking the opportunity to overvolt the processor at default settings. This was confirmed by ASUS technicians, or so it was reported here or there at Anand and other forums by a poster.

So before I ran out and bought a new cooler, or before I sent my processor to Silly Lotts for delidding and CLU, I'd try to reduce the voltage settings for the near-stock speed you prefer. And I'd look into your case ventilation.

But since you mentioned stress-testing software, I wouldn't be in a panic about your temperatures. If this were a matter of normal use -- then, maybe . . .

PS. About the RAM. 2133Mhz is the Intel standard, so the board-makers refer to anything about that as "OC." The XMP settings and speed of 3000Mhz or 3200Mhz will require more voltage from the processor. Your board is apparently providing this, even as you'd set the CPU to "all cores."

 

[BonzaiDuck]

You know, right away, looking at your HWMonitor profiles and the peak VID data reported, I am quite confident you could go into the Tweaker menu, choose "Manual" for initial voltage-tuning, and type in an "over-ride" of ~ 1.25V.

This system isn't set up yet to manipulate photos for posting. However, I just read an article that provided a graph of Skylake voltage scaling:

By Turbo speed:

4.2 Ghz . . . . 1.168V
4.3 Ghz . . . . . 1.200V
4.4 Ghz . . . . 1.232V
4.5 Ghz . . . . . 1.280V

With "Manual" tuning, if you save that setting, reboot to BIOS and check the monitored value in that menu, you'll have an idea of how close they are. That would provide ample voltage for 4.2 Ghz. You could transfer that value to "Voltage for Extra Turbo" in "Adaptive" mode. You could either set the Offset (+) to something between 0.001 an 0.005V, or just leave it on "auto."

Then check your temperatures and see that you have stability with a stress-test. It depends on how comfortable you are wading around in BIOS. But you could reduce your voltage and reduce the temperature. A VID of 1.39V maximum indicates to me that your motherboard defaults to an excessive setting. More volts than the processor possibly needs at that speed.

 

[VirtualLarry]

OP, you bought overclocked DDR4-3000 RAM. Not sure what to tell you, other than you should have expected to have to run XMP mode, and 1.35V for the DRAM voltage, if you wanted to run your RAM at it's full rated speed with your CPU. If you're unhappy with your temps, consider water cooling.

 

[BonzaiDuck]

OP, you bought overclocked DDR4-3000 RAM. Not sure what to tell you, other than you should have expected to have to run XMP mode, and 1.35V for the DRAM voltage, if you wanted to run your RAM at it's full rated speed with your CPU. If you're unhappy with your temps, consider water cooling.

Glad you mentioned that to the OP. I can't tell how experienced he is, other than that he's observant enough to at least notice his CPU thermals. From what I said, I began to worry he would mix up the topics of CPU and RAM voltage. 1.35 V is the standard XMP setting for that speed. XMP will fix it to that value. It should not be more, but for that speed, it should not be less.

The problem that I see is that the OP wants to leave his voltage settings on "auto." His stress test shows a peak VID of about 1.39+V, which I think is about par for the "auto" setting on these boards (like mine). Lemme check my notes here: . . . . yeah -- even my less-than-precise setting for 4.4Ghz stability throws up a maximum VID of 1.255V, VCORE swinging between 1.216 and 1.248V with LLC set to 3 and speed of 4.4 Ghz. He could reduce his temperatures just by tuning the voltage to the CPU speed.

All of these boards overvolt the Skylakes. Not enough to hurt -- but enough to make it seem that there's no OC headroom if the user isn't familiar with the status quo on these chips.

If I weren't planning to OC beyond maybe 4.4, a Noctua NH-U12 would suffice. A 212-EVO would suffice, more or less.

On the issue of water versus air, I'll have to wait and see how my own "waterless" system proves out over the next year, but I think I've proven it's possible to take a different route and send your processor to Silicon Lottery for $50 for the delid-relid. (Or do it yourself, but it seemed to me the $50 was worth it.) The result varies, but I got a 12C improvement from that. The way they now produce these chips with the polymer TIM, it either encourages people to invest in water, or it discourages higher overclocks.

And folks like the OP might worry about stress-test temperatures.

 

[ithehappy]

Thanks for the replies. I am not experienced with OCing at all. One of the main reasons for me to buy the K version of Skylake was its high stock frequency, so that I wouldn't need to OC. In any case I had no idea OCing RAM affects CPU temperatures this much!

Anyway, I have set the DRAM voltage to 1.25V (from auto's 1.35) by opting for the Manual option, and with All Core Enhancement option max temps were reduced by 5ºC (Package ~69ºC), which is good, but I am not gonna keep it anyway. Just for peace of mind I would like to have max temps well within 70ºC for these stress tests So I am gonna enable that Stock Intel option anyway, just don't find any advantage of this All Core Enhancement thing except all cores are turboing (yeah consider that a word) at 4200 MHz, I am more than okay with one core maximising only and others not.

In any case I did notice one thing, while running the stress test on 1.25V, the max voltage did touch 1.372V (shown on CPUID), LOL I don't know exactly what are the steps to achieve a fixed voltage.

Thanks for the responses guys, appreciate it.

 

[coercitiv]

Like BonzaiDuck already mentioned, your CPU voltage is quite high. AFAIK the reason for this motherboard manufacturers prefer to add additional voltage margins... either for safety or simply to ensure people can overclock by just increasing CPU multiplier.

One way to set voltage very close to stock settings based on values from Intel should be to change CPU Core/Cache Voltage from "Auto" to "Offset Mode" on your board, then add a nearly insignificant increment of +0.005V (done by selecting Offset Mode Sign [+] and CPU Core Voltage Offset [0.005]). Normally that would increase voltage even further, albeit by a very small value, but in the case of my BIOS it resulted in calculating that offset from "real" stock voltage, leading to a drop of about 0.05V. I kinda expect to do this for you as well, the effects will be visible after reboot. If this doesn't work you can safely revert to [Auto].

The advantage of this method is your are still technically on stock settings, no manual values and no guessing.

On the subject of RAM voltage, your memory kit is rated for 3000Mhz at 1.35V. I would advise you to keep that voltage in place. Correct adjustments to CPU voltage will get you lower temps anyway.

In any case I did notice one thing, while running the stress test on 1.25V, the max voltage did touch 1.372V (shown on CPUID), LOL I don't know exactly what are the steps to achieve a fixed voltage.

You don't want a "fixed" voltage, part of the reason voltage values jump around is to compensate for high load and/or different chip temps, what you want is a correct base value for the voltage from which the system can enforce correct "variable" voltage values.

 

[ithehappy]

T

Like BonzaiDuck already mentioned, your CPU voltage is quite high. AFAIK the reason for this motherboard manufacturers prefer to add additional voltage margins... either for safety or simply to ensure people can overclock by just increasing CPU multiplier.

One way to set voltage very close to stock settings based on values from Intel should be to change CPU Core/Cache Voltage from "Auto" to "Offset Mode" on your board, then add a nearly insignificant increment of +0.005V (done by selecting Offset Mode Sign [+] and CPU Core Voltage Offset [0.005]). Normally that would increase voltage even further, albeit by a very small value, but in the case of my BIOS it resulted in calculating that offset from "real" stock voltage, leading to a drop of about 0.05V. I kinda expect to do this for you as well, the effects will be visible after reboot. If this doesn't work you can safely revert to [Auto].

The advantage of this method is your are still technically on stock settings, no manual values and no guessing.

On the subject of RAM voltage, your memory kit is rated for 3000Mhz at 1.35V. I would advise you to keep that voltage in place. Correct adjustments to CPU voltage will get you lower temps anyway.



You don't want a "fixed" voltage, part of the reason voltage values jump around is to compensate for high load and/or different chip temps, what you want is a correct base value for the voltage from which the system can enforce correct "variable" voltage values.

Thanks. Applied this setting, but unfortunately it didn't reduce the max temps, actually it was around 1-2ºC higher than previous Stock Intel settings with all on Auto. However I do notice that CPUID now shows max voltage of 1.31V only instead of, which is a tiny bit lower, but don't know how it didn't have any effect on temps!

Also please tell me which stress test app would be the most efficient for Skylake, I know latest Prime95 shows abnormal temps, and I know the one I am using, HeavyLoad, well its not mentioned anywhere, LOL. And also is this CPUID HWMonitor a reliable monitoring tool too? If not which should I use?

 

[wingman04]

What are you using for stress test? Just set XMP it will only overclock your memory and raise the Vccio Vccsa voltage in the IMC a little.

 

[BonzaiDuck]

What would the rest of you say? I think since he's not all too keen on getting his feet wet with overclocking masochism, he should simply avoid Prime95 testing and use XTU, AIDA64 or some milder tests.

I say this because Intel guarantees the processor to run at 4.0 turbo 4.2, the mobo makers and RAM makers spec their products to run DDR4-3000 or whatever at that speed at the spec voltage and timings. So set the BIOS to XMP.

The other tests are likely to show temperatures well within the TCASE spec for the processor, which I may have said is simply a case and ventilation design-guideline to OEMs to assure the machines don't run over 68.5C of that spec in continuous 24/7 operation. I use it as a sort of goal and objective for cooling my systems -- no less, no more. A stress test on a Skylake that runs temperatures into even the low 80s isn't needed for stock speed and settings.

The skinny on the street is that Prime95 has outlived its use in marathon testing beyond someone's inclination to assure themselves, and isn't really desired for marathon runs. Instead, it still offers a very quick test for stability in a "custom" setting which you could run for a half hour, an hour, a couple hours. I fell asleep and woke up to find it had run 5 hours, but my cooling strategy keeps the temperatures at or below 70C for Prime95 and the higher accepted overclocks people either accept or exceed.

His Noctua cooler or a 212 EVO would likely heat up to as much as 20C above that -- if -- IF! -- he chose my clocks at 4.5, 4.6 and 4.7. To me, that's unacceptable, and even the 80s C would be too high. He only wants to run at turbo 4.2. He can expect to get the temperatures he gets with the cooler, and the mid-70s are acceptable. It's a stress test! The system doesn't run at those temperatures forever, because it doesn't regularly do the things the test is doing in overnight stability test runs.

Correct me. But I don't see anything wrong with his temperatures, the stress tests or the Stock and XMP settings. The question is -- does the system fail the tests with those settings? Then he needs to RMA a part after he troubleshoots to find it. But I don't think he mentioned any problem like that.

 

[BonzaiDuck]

Also -- acquire these:

Intel Extreme Tuning Utility or XTU
AIDA64 30-day trial (unless you want to buy the license)
(ASUS) ROG RealBench
HCI Memtest 64
OCCT

OCCT would present the most thermal stress of all those, but mostly in the Linpack option. OCCT: Linpack still runs acceptably cooler than LinX, Prime95 or Intel Burn Test, and it uses the Linpack library that at least two of those incorporate.

Even so, testing for stability with OCCT:CPU will find errors more quickly and runs cooler than the Linpack tab.

 

[wingman04]

I say throw everything you can at the CPU with prime95 blend. Prime95 uses pure Arithmetic for all the binary transistors in a CPU, the old 8 bits to a 1 bite x86 architecture is still being used like the old prime95 that has modern updates. With prime95 they found the skylake defect that Intel missed.

 

[BonzaiDuck]

I say throw everything you can at the CPU with prime95 blend. Prime95 uses pure Arithmetic for all the binary transistors in a CPU, the old 8 bits to a 1 bite x86 architecture is still being used like the old prime95 that has modern updates. With prime95 they found the skylake defect that Intel missed.

that's true -- all true. Especially viable for a stock-speed system. But I was talking about overclocking with a retail-box unmodified chip and a single tower heatpipe cooler.

At his setting, he should be able to run Prime95 Blend day and night. I would think that would toast up his processor probably to the level his Heavy Load did. But his temperatures are nothing to get in a panic about now, and they probably wouldn't with Prime95.

I just don't see the purpose in all the stress testing unless he's going to run up the multiplier from its default. You'd still want to stress test with the other software on new parts, and if changing the Tweaker page to "Sync all cores" or "for all cores," OCCT should be enough.

There's just been enough bad press in the OC community lately about Skylake and Prime95. Like I said -- my cooling is robust enough that Prime95 Blend, LinX and Intel Burn Test will take me to 70C between 4.6 and 4.7 Ghz.

I was relieved yesterday that I could get through several hours of Prime and finally see no errors the other day, after some cautious voltage tweaks in increments of 2 and 4 mV.

Which -- was a good exercise, so I now discover that LinX doesn't present a problem either. Myself -- I'm good to go.

I'm not temperature-limited, but I'm self-limited about the voltage. I'd rather see the VID under 1.42 and the VCORE unloaded maximum below 1.40V. Apparently, LLC=5 seems to keep VID and VCORE close without VCORE exceeding VID. If I can't tune to Prime95 and LinX and meet that objective, then I'll drop the multiplier one notch for 24/7 operation.

 

[wingman04]

that's true -- all true. Especially viable for a stock-speed system. But I was talking about overclocking with a retail-box unmodified chip and a single tower heatpipe cooler.

At his setting, he should be able to run Prime95 Blend day and night. I would think that would toast up his processor probably to the level his Heavy Load did. But his temperatures are nothing to get in a panic about now, and they probably wouldn't with Prime95.

I just don't see the purpose in all the stress testing unless he's going to run up the multiplier from its default. You'd still want to stress test with the other software on new parts, and if changing the Tweaker page to "Sync all cores" or "for all cores," OCCT should be enough.

There's just been enough bad press in the OC community lately about Skylake and Prime95. Like I said -- my cooling is robust enough that Prime95 Blend, LinX and Intel Burn Test will take me to 70C between 4.6 and 4.7 Ghz.

I was relieved yesterday that I could get through several hours of Prime and finally see no errors the other day, after some cautious voltage tweaks in increments of 2 and 4 mV.

Which -- was a good exercise, so I now discover that LinX doesn't present a problem either. Myself -- I'm good to go.

I'm not temperature-limited, but I'm self-limited about the voltage. I'd rather see the VID under 1.42 and the VCORE unloaded maximum below 1.40V. Apparently, LLC=5 seems to keep VID and VCORE close without VCORE exceeding VID. If I can't tune to Prime95 and LinX and meet that objective, then I'll drop the multiplier one notch for 24/7 operation.

I run prime95 blend with a 4.5GHz i5 6600k vcore 1.33v 15 hours hyper 212 cooler at a peak temp of only 84c. I wonder why you say Prime95 is defective, if you do distributed computing or complex workload conditions, it is the same thing as running Prime95. Gaming does not use the CPU to it's full potential.

 

[BonzaiDuck]

I run prime95 blend with a 4.5GHz i5 6600k vcore 1.33v 15 hours hyper 212 cooler at a peak temp of only 84c. I wonder why you say Prime95 is defective, if you do distributed computing or complex workload conditions, it is the same thing as running Prime95. Gaming does not use the CPU to it's full potential.

Wingman!! Wingman!! Peak temperature of ONLY 84C?

I ordered my parts on September 2. I've taken unforeseen changes in direction, but this project has emerged over several months of adapting two systems to two different cases, and I decided "big air" instead of "big water" at the last minute.

But I'd been studying forum information, reviews, articles and white papers here and there over the last year.

I'd seen all sorts of myths propagated about Prime95, and just as much about LinX. I think these myths -- cautions, many of them -- began with the Haswells and Devils Canyon, polymer TIM be damned.

There's no doubt you can run up to 90C degrees with these chips, but one, probably more reviews of Skylake overclocking mentioned the noise created by the heat, causing instability for the high clocks.

Forums and the online literature was steering people to milder tests.

The only problem with Prime95 and LinX both using the AVX instruction set dovetails with the change in Intel's practice of using Indium solder between die and I-H-S. It's harder to cool the chip, just for that reason. YET! Intel had reduced the Skylake TCASE spec from prior gen chips by 5C. Even for a spec that seems meaningless for overclocking to the limit, one has to ask "why?"

the chip is capable of running well above that temperature, but the hotter the chip, the closer to instability.

But with these temperatures? Right now? With my Skylake-in-progress topping out at 68C with IBT Maximum? I could run Prime95 for two days.

It's been discussed quite a bit around here for the last couple years: you shouldn't have to stress the processor for 24 hours to assure stability with some confidence. that was the selling point for OCCT's CPU test.

And I think the errors are Poisson distributed -- each milestone hour with no errors increases the chance that no errors will occur for a subsequent and equal length of time.

Instead, what's impressed me this week about Prime95 were the custom tests that will find a minimum stable voltage beginning with a half-hour test. You can tune the VCORE to within a couple mV of that level and then add maybe another 4 mV for good measure.

That leaves the question of "maximum safe voltage," for which there is no longer a spec. Some of the wisdom says 1.40V. Two things to consider: first, the die shrink would suggest a lower VCORE threshold than prior 22nm and 32nm processors; but second, other information suggests they've improved the "resilience" of the chip with Skylake, pertaining to slow degradation -- electromigration.

I just figure it's anyone's gamble, once you arrive at the 1.40 threshold. And you see all sorts of cautions in forums like "Oh, no! Don't go over 1.35V unless you use water-cooling!"

Or -- you replace the original TIM of the chip with CLU.

 

[ithehappy]

I have been reading and reading about Skylake OC and voltages and all, and this is what I have gathered so far. Please tell me if I am wrong, when I am wrong.

Firstly as my RAMs are designed to run at 3000 MHz at 1.35V, I shouldn't alter the DRAM voltage of 1.35V.

Now I need to lower the CPU Core/Cache voltage, but to what value? When I disabled everything and run the system on full stock (RAM at 2100), I see in BIOS that CPU voltage keeps changing between 1.264 and 1.280V. So I am guessing whatever value I set under Manual mode for CPU Voltage it has to be more than 1.28V? Am I wrong here?

Thirdly, I think my system might/will draw more power than other Skylake systems because I am using a very old PSU, my TX650 is nearly six years old, I was just thinking about it yesterday, the more used up a computer product is, more voltage it would need to as time goes by right? So that could be a reason for higher than normal voltage.

Basically I am VERY HAPPY with the stress test max temps result when its run on full stock mode, less than 60ºC for my high ambient is excellent result, but when I use the XMP profile that 59ºC moves up to 64ºC (with stock Intel setting, not All Core Enhancement) , again 64C is a great temp too, but it mentally bothers me a little as I am only OCing the RAM, but the CPU. However now I understand that CPU is getting a little bit OCed too when XMP is being applied, that I didn't know and I don't like it LOL.

PS: I read at least 50 topics now about this Skylake OC thing, and more or less in every topic on Tom's its mentioned that Prime95 shows abnormal temp for these newer chips, so its recommended to not use Prime, and even if its used some older version is recommended, they say IBT is much better and more useful for Skylake than Prime, and they also said to use RealTemp instead of CPUID HW Monitor, because RealTemp was specifically designed for Intel CPUs. LOL so many contradictory comments everywhere, for new users and idiots like me it only confused the heck outta us.

 

[wingman04]

Wingman!! Wingman!! Peak temperature of ONLY 84C?

I ordered my parts on September 2. I've taken unforeseen changes in direction, but this project has emerged over several months of adapting two systems to two different cases, and I decided "big air" instead of "big water" at the last minute.

But I'd been studying forum information, reviews, articles and white papers here and there over the last year.

I'd seen all sorts of myths propagated about Prime95, and just as much about LinX. I think these myths -- cautions, many of them -- began with the Haswells and Devils Canyon, polymer TIM be damned.

There's no doubt you can run up to 90C degrees with these chips, but one, probably more reviews of Skylake overclocking mentioned the noise created by the heat, causing instability for the high clocks.

Forums and the online literature was steering people to milder tests.

The only problem with Prime95 and LinX both using the AVX instruction set dovetails with the change in Intel's practice of using Indium solder between die and I-H-S. It's harder to cool the chip, just for that reason. YET! Intel had reduced the Skylake TCASE spec from prior gen chips by 5C. Even for a spec that seems meaningless for overclocking to the limit, one has to ask "why?"

the chip is capable of running well above that temperature, but the hotter the chip, the closer to instability.

But with these temperatures? Right now? With my Skylake-in-progress topping out at 68C with IBT Maximum? I could run Prime95 for two days.

It's been discussed quite a bit around here for the last couple years: you shouldn't have to stress the processor for 24 hours to assure stability with some confidence. that was the selling point for OCCT's CPU test.

And I think the errors are Poisson distributed -- each milestone hour with no errors increases the chance that no errors will occur for a subsequent and equal length of time.

Instead, what's impressed me this week about Prime95 were the custom tests that will find a minimum stable voltage beginning with a half-hour test. You can tune the VCORE to within a couple mV of that level and then add maybe another 4 mV for good measure.

That leaves the question of "maximum safe voltage," for which there is no longer a spec. Some of the wisdom says 1.40V. Two things to consider: first, the die shrink would suggest a lower VCORE threshold than prior 22nm and 32nm processors; but second, other information suggests they've improved the "resilience" of the chip with Skylake, pertaining to slow degradation -- electromigration.

I just figure it's anyone's gamble, once you arrive at the 1.40 threshold. And you see all sorts of cautions in forums like "Oh, no! Don't go over 1.35V unless you use water-cooling!"

Or -- you replace the original TIM of the chip with CLU.

I overclock with a safety margin 4.5GHz 84c. Intel throttles Turbo then base clock at 100c then shuts down the PC 125c to prevent damage, so folks can safely run at 99c. 84c is very safe with the prime95 program or others, If folks don't overclock on the ragged edge of stability where temperature expansion of the Die makes a difference.

Prime95 and other programs 24 hours stability testing is for people that really use there PCs a lot, some people make it to 1-23 hours then have a CPU calculation failure. The problem is when folks overclock on the bleeding edge the transistors don't have much tolerance, the only thing you can do is a final test for a long period time. Intel and other Chip makers uses Arithmetic to calculate voltage+AMPs and transistor speed then they add a safety margin for a transistor zero failure switch time rate. For 4 months my PC has run 24/7 gaming also work with zero lockups or problems at 4.5GHz.

There is no such thing as safe overclocking voltage. When you increase the clock speed and voltage you also increase the Amps+voltage that causes degradation. A very simple explanation Intel uses Arithmetic to figure out resistance of the high-k dielectrics in conjunction with a metallic gate then they know the melting point of of the material, with all that they use ohm's law + TDP+ melting point to figure out safe stock voltage+Amps. I volunteer on the Intel site and I see many CPUs damaged by overclocking. When they make CPU Dies no two are the same there are imperfections on all of them that can or will fail with overclocking.

 

[wingman04]

I have been reading and reading about Skylake OC and voltages and all, and this is what I have gathered so far. Please tell me if I am wrong, when I am wrong.

Firstly as my RAMs are designed to run at 3000 MHz at 1.35V, I shouldn't alter the DRAM voltage of 1.35V.

Just run XMP and be done with it. It will set all your memory settings to optimum. link http://www.intel.com/content/www/us/en/gaming/extreme-memory-profile-xmp.html

Now I need to lower the CPU Core/Cache voltage, but to what value? When I disabled everything and run the system on full stock (RAM at 2100), I see in BIOS that CPU voltage keeps changing between 1.264 and 1.280V. So I am guessing whatever value I set under Manual mode for CPU Voltage it has to be more than 1.28V? Am I wrong here?

Why do you need to change that voltage?

Thirdly, I think my system might/will draw more power than other Skylake systems because I am using a very old PSU, my TX650 is nearly six years old, I was just thinking about it yesterday, the more used up a computer product is, more voltage it would need to as time goes by right? So that could be a reason for higher than normal voltage.

Power supply unit has nothing to do with power draw from the components running connected to it.

Basically I am VERY HAPPY with the stress test max temps result when its run on full stock mode, less than 60ºC for my high ambient is excellent result, but when I use the XMP profile that 59ºC moves up to 64ºC (with stock Intel setting, not All Core Enhancement) , again 64C is a great temp too, but it mentally bothers me a little as I am only OCing the RAM, but the CPU. However now I understand that CPU is getting a little bit OCed too when XMP is being applied, that I didn't know and I don't like it LOL.

Just use the XMP for memory in Bios and it won't overclock the CPU. The safe temperature for silicon CPU is 100c then Intel will throttle the CPU speed 100c to keep it there and safe. laptops run that way all the time.

PS: I read at least 50 topics now about this Skylake OC thing, and more or less in every topic on Tom's its mentioned that Prime95 shows abnormal temp for these newer chips, so its recommended to not use Prime, and even if its used some older version is recommended, they say IBT is much better and more useful for Skylake than Prime, and they also said to use RealTemp instead of CPUID HW Monitor, because RealTemp was specifically designed for Intel CPUs. LOL so many contradictory comments everywhere, for new users and idiots like me it only confused the heck outta us.

Depends what you use your PC for, some apps or programs use 100% of the CPU the some don't, prime95 program uses 100% of the CPU. Temperature is related to CPU utilization. Intel and other chip makers build them to use 100% of everything otherwise we would have to restrictions of usage.

Just 1% of the PC users even looks at the temperature.

 

[ithehappy]

I like your clear cut replies.

 

[ithehappy]

Why do you need to change that voltage?

Umm, to have lower temp? I was told that my peak voltage was high. On first post you could its peaking up to 1.39V.

Power supply unit has nothing to do with power draw from the components running connected to it.

I didn't know this at all, thanks.

Just use the XMP for memory in Bios and it won't overclock the CPU.

If it is not OCing my CPU with XMP, then why do I have extra ~5ºC temp with it being enabled in comparison with no XMP? Honestly if I noticed same temps with both XMP enabled and disabled, which was below 60C then I would not have created this topic.

59C- Full stock
64C- XMP with stock Intel option
74C- XMP with All Core Enhancement

So basically if someones uses the last one the temp difference is 15C between XMP enabled and disabled, that's huge.

Depends what you use your PC for, some apps or programs use 100% of the CPU the some don't, prime95 program uses 100% of the CPU. Temperature is related to CPU utilization. Intel and other chip makers build them to use 100% of everything otherwise we would have to restrictions of usage.

Thanks for this info.

Just 1% of the PC users even looks at the temperature.

I think its even less and I envy them for living more practical lives

PS: What's CPU VCCIO voltage?

 

[wingman04]

Umm, to have lower temp? I was told that my peak voltage was high. On first post you could its peaking up to 1.39V.

I was told Vcore of 1.4v in the overclocking community was ok. I don't mess with Cache voltage.

There is no such thing as safe overclocking voltage. When you increase the clock speed and voltage you also increase the Amps+voltage that causes degradation. A very simple explanation Intel uses Arithmetic to figure out resistance of the high-k dielectrics in conjunction with a metallic gate then they know the melting point of of the material, with all that they use ohm's law + TDP+ melting point to figure out safe stock voltage+Amps. I volunteer on the Intel site and I see many CPUs damaged by overclocking. When they make CPU Dies no two are the same there are imperfections on all of them that can or will fail with overclocking.

If it is not OCing my CPU with XMP, then why do I have extra ~5ºC temp with it being enabled in comparison with no XMP? Honestly if I noticed same temps with both XMP enabled and disabled, which was below 60C then I would not have created this topic.

59C- Full stock
64C- XMP with stock Intel option
74C- XMP with All Core Enhancement

So basically if someones uses the last one the temp difference is 15C between XMP enabled and disabled, that's huge.

Some Settings with XMP software or Bios XMP overclocks the CPU. Also with faster memory there is more CPU utilization = increased temperature.

PS: What's CPU VCCIO voltage?

With XMP speed ram 3000 it raises the voltage of the (IMC) integrated memory controller in the CPU. That can also increase temperature.
Memory controller voltage Vccio original 0.960v changed to 1.165v for 3000 speed XMP.
Memory controller voltage Vccsa (System Agent) original 1.060v changed to 1.275v 3000 speed XMP.

 

[ithehappy]

Understood, and I am keeping this Auto settings for now. I tried Manually setting CPU Core to 1.25V and funnily enough even with low peak voltage than before (1.33 vs 1.39 of previous test) I didn't see any lowered max temps, actually 1-2ºC higher! I then changed the VCCIO to stock like setting too, but same. This is with the stock Intel setting. If I reduce the voltage on All Core Enhancement mode however the max temps reduce by ~5ºC, but I find that mode kinda pointless anyway.

So keeping this XMP stock Intel settings with everything at Auto.

PS: One more question, a friend who has 6600k told me there's an option in BIOS where you could have the system to run at really low power, I tried to find a settings like that, but couldn't, it would be helpful for me because quite often I keep this system on for downloading only for prolonged period, so lower power usage would be useful for those occasions.

 

[wingman04]

Understood, and I am keeping this Auto settings for now. I tried Manually setting CPU Core to 1.25V and funnily enough even with low peak voltage than before (1.33 vs 1.39 of previous test) I didn't see any lowered max temps, actually 1-2ºC higher! I then changed the VCCIO to stock like setting too, but same. This is with the stock Intel setting. If I reduce the voltage on All Core Enhancement mode however the max temps reduce by ~5ºC, but I find that mode kinda pointless anyway.

So keeping this XMP stock Intel settings with everything at Auto

Just for your information XMP CPU voltage Vccio and Vccsa varies a little form mother board manufactures it is not a Intel specification. Also All Core Enhancement is a option on your motherboard and is not a Intel specification. This is kind of confusing Intel does not condone running the CPU out of specification they just advertiser it sometimes.
Have a look at this link. Performance Tuning Protection Plan https://click.intel.com/tuningplan/

PS: One more question, a friend who has 6600k told me there's an option in BIOS where you could have the system to run at really low power, I tried to find a settings like that, but couldn't, it would be helpful for me because quite often I keep this system on for downloading only for prolonged period, so lower power usage would be useful for those occasions.

Most boards save power automatically they cut the CPU running phase count down and shut down PC-E lanes not used, It saves much less than a watt. A LED light bulb 60watt equivalent from sylvania uses 8.5w your PC only downloading uses about 50watts, playing PC games about 300w-400w. Air conditioning uses 3500 watts. Stove uses up to 3000 watts Clothes dryer 1000W 4000W Coffee Maker 800W 1400W. I use a killawatt to measure 120v usage.

If you have any other questions just ask.

 

[ithehappy]

Just for your information XMP CPU voltage Vccio and Vccsa varies a little form mother board manufactures it is not a Intel specification. Also All Core Enhancement is a option on your motherboard and is not a Intel specification. This is kind of confusing Intel does not condone running the CPU out of specification they just advertiser it sometimes.
Have a look at this link. Performance Tuning Protection Plan https://click.intel.com/tuningplan/


Most boards save power automatically they cut the CPU running phase count down and shut down PC-E lanes not used, It saves much less than a watt. A LED light bulb 60watt equivalent from sylvania uses 8.5w your PC only downloading uses about 50watts, playing PC games about 300w-400w. Air conditioning uses 3500 watts. Stove uses up to 3000 watts Clothes dryer 1000W 4000W Coffee Maker 800W 1400W. I use a killawatt to measure 120v usage.

If you have any other questions just ask.

You have been a great help really, for saving my time actually.

Well about that power saving thing, mine is averaging at 85-90 watts, on Idle, (with monitor off that is), I would love to lower it down even more, but its all right. I just found those options on BIOS, you know Performance, Balanced and Power saving mode, given by Asus obviously, don't know if they really work, on default Balanced one anyway, but my friend told me something else, not these profiles, whatever, not a big deal at all, and thanks once again.

 

[BonzaiDuck]

I can't find points of disagreement with Wingman.

There's just no spec anymore for "maximum safe VCORE." If I'm not mistaken, the last published spec was for a 32nm Intel Nehalem processor -- around 1.38V. What Wingman says about Intel's own "calculation" should also be seen in context of their "cost accounting." You would think -- if there were a published spec (and there is no more) -- they would attempt to guarantee a minimum of 3-year warranty returns just from people running the processor at stock speeds. On the other hand, we all agree that with no spec, the motherboard makers ship boards with "auto" settings that can "turbo-idle" to 1.39V, as does the OP's, shown from his screenies. It's the same thing I discovered about my ASUS board the first time I booted the system.

Here's a graph similar to the work IDontCare had published here at the forums (IDC is an illustrious member, who -- I think -- worked for TI). At first, I thought I recognized IDC's work, but the comma-decimal is a European convention, I think:

Consider that I have a "binned" chip, and there was no such distinction for the testing for the graphed results. This is the same thing I'd noticed about IDC's work on the Sandy Bridge and possibly Ivy (in his de-lidding thread). Either typing in these voltage milestones for "Manual mode" fixed voltage or for "Adaptive mode's" "Voltage for extra turbo" on top of a low or 0.000V Offset can likely give you panic-free stress-testing, because the voltage is so close that stressing programs will catch the error and simply stop without a BSOD. Of course, what you type in isn't "what you get," so I do a double-check after BIOS "save and reboot" into BIOS to see what the VCORE monitor shows under "Manual Mode." I adjust to match the reported VCORE against the schedule.

Cheap trick? The only other option is to edge up the voltage from some seeming stability-minimum and suffer through BSODs. I've managed to tune in 4.6 and 4.7 Ghz with only three or four BSODs during the first few hours after initial boot-up.

Wingman notes there's "no such thing" as "safe overclocking voltage," but . . . "we been aroun', you know?!" through some several generations of processors. I've personally never damaged a processor, because I make my own rules based on the incomplete information I have, and I'd rather build a "great computer" than beat an "LN2 competition." My oldest Sandy Bridge has been running 24/7/365 @ either 4.6 or 4.7 -- variously. That's 5 years, running OC'd at those speeds.

The die-shrinks mean less surface area to transfer heat to the processor-cap and cooling device. Voltage defaults have declined with wattage defaults, but this polymer TIM thing -- it's a setback if you OC.

But there's nothing worrisome about the OP's temperatures under stress at stock speeds. With the Noctua single-tower cooler, he's fine to run the memory at XMP spec, and "sync all cores." Even that -- is slight overclock. Let me tell you about the WWII sergeant who trained his men in short-wave radio. He put some 4"-wide paint-filters on the radio adjustment knobs and painted nipples on them.

"Be patient. Be very, very gentle. . . . " [Where are the icons and smilies with this new interface? I needed to make a big toothy grin here.]

CORRECTION: Just to avoid misleading anyone, you have to recheck the reported voltage IN WINDOWS once you try typing in a fixed setting in BIOS, and watch for indications at turbo speed. You can get by with lower LLC -- maybe level 3 works for 4.5 Ghz. Ultimately, level 5 (on my chip anyway) seems to match the reported value under stress-load (little or no vDROOP), and is close to the VID needed to raise that voltage for the fixed setting.