Discussion Rusty on OC'ing my Skylake ( & Kaby) Processors -- Continuing the Vaping-Pen Static-Charge Saga

[BonzaiDuck]

So. As I said in that older thread I started about killing my motherboard USB controller with unwise behaviors, I put in the replacement (identical) board with a new Silly-Lots De-Lidded i7-7700K and a two-stick RAM kit of virtually the same TridentZ DDR4-3200 14-14-14 RAM sticks. I bought the 2x16GB=32GB RAM to alleviate stress on the IMC, although it was probably unnecessary. But I know I should be able to run 'em at Command-Rate = 1, so there . . .

I'm scouring my server and now-recovered system for any screen-shots or notes I took when I overclocked the Skylake. I've turned up 18 pages of handwritten notes, and now remember that -- somewhere -- I put it all in an Excel file,, so I'll look for that . . . too . . .

Wanted to get this beast up and running just for stock settings, since I "use" my computer for things, like tax preparation. And I have medical appointments in my Outlook calendar which were made before I zapped the vaping-pen with a dry-windy-day-shuffle-on the carpet static charge on January 15.

But just for preliminaries, I started looking at my real-time HW-Info screens while running a three-pass sequence of LinX. I thought the core temperatures were a bit too high. You'd expect a spread between the four cores of up to 10C here or there, bouncing around in a 5 to 8C degree range more of the time. But I noticed the temperatures on at least one core spiking to 75C, and I'm running stock settings.

Took a closer look. GEEE-sus! the motherboard was giving an "Auto"-determined VCCIO voltage of 1.32V, and I think the VCCSA system agent voltage was around 1.28V. Where does THAT come from? So I down-clocked the RAM to 2,133 Mhz, exited and re-entered the BIOS. NOW it looks normal -- with about 0.968V for the VCCIO and 1.072V for the VCCSA setting under auto. I KNOW this RAM runs under a VCCIO of around 1.15V with 4 sticks of the same specification at the spec speed of 3200 14-14-14.

This is -- to people posting their exuberant threads about their i7-10700K and other processors with Z390/490 motherboards -- an old version. It's a Z170 board. But there are always new BIOS versions, and I can't remember how I started tweaking the Skylake four years ago. Did I run the RAM at a lower speed and then fix the voltages before twisting it up to 3200 Mhz? I can't remember.

I DO remember that I made an early note of the "Auto" Skylake VCORE, also reading in at least two reliable sources that Intel had only released "safe-voltage" info to their motherboard partners. The boards -- even then -- were overvolting the processor. And I found enough chatter on the web today to confirm that boards are over-volting things like VCCIO/IMC and VCCSA.

Are there any insightful observations about the initial overclocking strategy that someone learned under these scenarios? I have a few ideas, because I vaguely remember what I did in 2017, and I have my notes -- if I can read them!

 

[BonzaiDuck]

OK . . . no comments from others so far. Understandably -- the original Skylake and Kaby Lake processors are now "dated". Even so, the 10th-gen units people have been discussing are still built of Skylake cores.

In an earlier thread -- also without comments from others -- I'd asked if anyone had seen voltage-scaling information about the Kaby. Since I've sorted out so far how to proceed with this, I found it -- at Tom's Hardware:




This should save me a lot of time and tedium. I've already disabled "Turbo mode" on my rig to find out the loaded base-frequency voltage at 4.2Ghz -- 1.155V -- when VCORE is set to "Auto". Checking again today, but I think the unloaded value is 1.125V. I try to disable Speedstep in BIOS, but the rig continues to behave as if it is enabled under Windows 10. That makes it harder to get an estimate of the unloaded base-frequency voltage . . .

So the only real question, before any intermediate "Manual" OC settings, is: "Do I have a 'Good' one or a 'Potato'?" So far, I have nearly 500 Mhz of head-room to start finding out, between 4.2 and 4.7 Ghz.

On the Tom's article, they mentioned that resellers would get "RMA Potatoes" only to sell them again. If my retail box has the original Intel sticker-seal, what does that mean? Was my Kaby Lake from original, untested stock? Or is it a "potato"?

 

[Mopetar]

Here's the historical data from Silicon Lottery for Kaby Lake that should give some other points of comparison.

Kaby Lake	All Core SSE Frequency	All Core AVX2 Frequency	BIOS Vcore	% Capable
7600K	4.80GHz	4.60GHz	1.400V	100%
7600K	4.90GHz	4.70GHz	1.412V	Top 89%
7600K	5.00GHz	4.80GHz	1.425V	Top 74%
7600K	5.10GHz	4.90GHz	1.437V	Top 56%
7600K	5.20GHz	5.00GHz	1.450V	Top 30%
7600K	5.30GHz	5.10GHz	1.450V	Top 5%
7700K	4.80GHz	4.60GHz	1.400V	100%
7700K	4.90GHz	4.70GHz	1.412V	Top 96%
7700K	5.00GHz	4.80GHz	1.425V	Top 78%
7700K	5.10GHz	4.90GHz	1.437V	Top 36%
7700K	5.20GHz	5.00GHz	1.450V	Top 12%
7700K	5.30GHz	5.10GHz	1.450V	Top 1%

Based on the data from Tom's it looks like they consider a good CPU to be in about the top third. If you can get to 5.0 GHz it's probably not a complete potato. One thing to note is that for their Kaby Lake results the CPUs were delidded before being tested so your results may be slightly worse for that reason alone.

 

[BonzaiDuck]

Here's the historical data from Silicon Lottery for Kaby Lake that should give some other points of comparison.

Kaby Lake	All Core SSE Frequency	All Core AVX2 Frequency	BIOS Vcore	% Capable
7600K	4.80GHz	4.60GHz	1.400V	100%
7600K	4.90GHz	4.70GHz	1.412V	Top 89%
7600K	5.00GHz	4.80GHz	1.425V	Top 74%
7600K	5.10GHz	4.90GHz	1.437V	Top 56%
7600K	5.20GHz	5.00GHz	1.450V	Top 30%
7600K	5.30GHz	5.10GHz	1.450V	Top 5%
7700K	4.80GHz	4.60GHz	1.400V	100%
7700K	4.90GHz	4.70GHz	1.412V	Top 96%
7700K	5.00GHz	4.80GHz	1.425V	Top 78%
7700K	5.10GHz	4.90GHz	1.437V	Top 36%
7700K	5.20GHz	5.00GHz	1.450V	Top 12%
7700K	5.30GHz	5.10GHz	1.450V	Top 1%

Based on the data from Tom's it looks like they consider a good CPU to be in about the top third. If you can get to 5.0 GHz it's probably not a complete potato. One thing to note is that for their Kaby Lake results the CPUs were delidded before being tested so your results may be slightly worse for that reason alone.

Oh -- did I not say that? I'm even considering the purchase of the Der8auer De-Lid Die-Mate tool, but I never hesitate to send my processors off to Silicon Lottery or even buy them as a binned processor. They -- or "he" -- has done that operation about 10,000 times or more over the last five or six years. For $40, I'd rather not trouble myself getting through a "first time".

There's more to this than the expectations of the chip. Indeed, I think another "over-clocking guide" for Kaby Lake had shown some people got to 5.0Ghz with a Noctua NH-D15 -- virtually no difference than a good AiO cooler and not far off from a custom loop. The information source made clear that the processor had been delidded and prepared with Conduct-o-naut. The voltage they cited was about 1.282 or 1.29V. Of course, that's for a grain of salt: we OC for total stability but many will show their benchmarks based on preliminary settings.

Even the graph data points in the Tom's article may not come up to my standard, but they so far appear to be within the ball-park. I'm currently running AgentGOD's IBT at "Maximum" for a 4.7Ghz speed setting, and the drooped voltage seems to be around 1.184V.

But I've encountered more than just "street-intel" about what is going on with these Kaby Lake chips. Customer reviews complain about heat and throttling, but because of the de-lid and re-lid, I have none of that problem so far. At 4.7Ghz, my maximum single-core temperature is 61C, and as it continues to run, it's average temperature is 57C.

I would not have got this far without my own experience OC'ing since the Q6600, more significantly because of written indications I've gleaned online. It looks as though the SVID values the processor gives to the onboard VRM are all excessive for what is called for.

Running at stock speeds, "Auto" settings that include the "Adaptive mode" BIOS feature are pulling 1.28V or more right away under moderate stress. Everyone complains that it's the motherboard which overvolts these cores, but it seems to be the processor itself. That's why I do all this preliminary stuff to look closely at the stock base frequency, but I never thought to encounter this other factor. I can't even use Adaptive mode because I cannot find settings that bring down the voltage to a reasonable level for the most modest over-clock. Adaptive mode needs to use the processor's SVID, as opposed to the over-ride which controls the motherboard external VRM.

Because the most efficient voltage for the stock turbo and the modest over-clocks is below 1.18+V, you could think that you don't need anything other than "Manual" "fixed Vcore over-ride". EIST seems to provide enough of the lower power-saving clock speeds. It's only the frequency swings that result, since the voltage is already so low, it's only about 0.5V over the idle value under SpeedStep. So it hardly matters if you're going to set a fixed Vcore over-ride at 1.2V for constant use. It's still a mystery to me about why I can't get the Adaptive Mode to work for me, nonetheless. It all seemed to work to my expectations with the Skylake, but then, with the Skylake, I couldn't get to 4.7 with temperatures below 61C.

I bumped up LLC so that I still have between (about) 16mV to 32mV of droop, or so it appears to be the case.

 

[BonzaiDuck]

Here's the historical data from Silicon Lottery for Kaby Lake that should give some other points of comparison.

Kaby Lake	All Core SSE Frequency	All Core AVX2 Frequency	BIOS Vcore	% Capable
7600K	4.80GHz	4.60GHz	1.400V	100%
7600K	4.90GHz	4.70GHz	1.412V	Top 89%
7600K	5.00GHz	4.80GHz	1.425V	Top 74%
7600K	5.10GHz	4.90GHz	1.437V	Top 56%
7600K	5.20GHz	5.00GHz	1.450V	Top 30%
7600K	5.30GHz	5.10GHz	1.450V	Top 5%
7700K	4.80GHz	4.60GHz	1.400V	100%
7700K	4.90GHz	4.70GHz	1.412V	Top 96%
7700K	5.00GHz	4.80GHz	1.425V	Top 78%
7700K	5.10GHz	4.90GHz	1.437V	Top 36%
7700K	5.20GHz	5.00GHz	1.450V	Top 12%
7700K	5.30GHz	5.10GHz	1.450V	Top 1%

Based on the data from Tom's it looks like they consider a good CPU to be in about the top third. If you can get to 5.0 GHz it's probably not a complete potato. One thing to note is that for their Kaby Lake results the CPUs were delidded before being tested so your results may be slightly worse for that reason alone.

Well, Mopetar, here's my take on the various sources of "reliable intel" we've pooled on the Kaby Lake frequency-voltage curves or schedules.

Some in the review-and-testing community may only post voltages that allow certain frequency settings to boot "stably" into Windows. That's one standard.

I suspect that the data from Silicon Lottery uses a more rigorous standard.

Sometimes, one encounters special advice about different testing programs and what they do. For instance, there's a Prime95 "fixed" program-size setting -- I think I'd better hunt it down because it was useful -- which fails any type of instability in less than 15 minutes. If you run it an hour, you might be able to reduce the number of hours spent running other torture-tests.

I never made a note of it four years ago, but that's when I discovered OCCT's "CPU" or "OCCT:CPU" test, which can be compared to a Linpack test which OCCT includes in its menu.

I think it may conform to the Poisson-distributed patterns cited, for instance, as distributions of errors in a bolt of cloth -- changing by length or amount. High failure rates occur early in the processing of the fabric -- or in the electronic failure rates sometimes cited as "infant mortality". In the OCCT:CPU test, an instability that will escape 10 iterations (likely an hour or so) in IntelBurnTest will emerge in the first three or four minutes.

Focusing on that test alone, my numbers are quickly approaching those from the Silicon Lottery block-sample. That is, for 4.9 Ghz, I need a LLC settings that keeps droop to within 12 to 16 mV, and probably less than or equal to the 4.9 Ghz cohort of any similar block sample. And I need to set the "VCORE Voltage-Override" to 1.360V, which then gives a monitored/recorded value of . . . . 1.36V "maximum", 1.32 "minimum" and 1.344V "prevailing. It's going to pass the 1-hour OCCT:CPU test. And I'll file it away in the BIOS "Saved Profiles" as "49_Working" for a future visitation with various longer tests. But this voltage can only increase, for either 4.9 or 5.0 Ghz speeds. I will soon be at 1.40V.

I'll say this. If I got a retail-box "potato", it's still a "good" one. Good enough.

Sooner or later, I need to settle on which speed between 4.5 and 5.0 to use for 24/7 no-nonsense running. And I already figured that 4.5 seems good enough for now, but I could choose another speed and twist up the voltage another 10mV or so.
.
Somewhere between a 'self-imposed" rule and one justified by an estimate, I'm thinking if you take vDroop partly out of the equation by about 50mV with a moderate/mid-range LLC choice, the transitory spike that occurs just after the CPU unloads itself would be about the same in the opposite direction. So if I set the VCORE override to 1.40 and get back 1.392V monitored value with the same 16 mV vDroop, the spike could take it to 1.45V. For me, that's right on the edge of ruling out a 5.0 for 24/7 . . . But my assumptions about this are still based on the speculations of a "non-electronics" person. . . . [Well, it looked like OCCT:CPU failed after about 12 minutes. Add another 10mV . . . then . . . ]

 

[Mopetar]

What do you need it to do while running 24/7? If it's something where the speed isn't necessarily important, I'd go for a more energy efficient tuning.

 

[VirtualLarry]

There's a definite reason that overclocking is mostly "dead" on newer platforms. Because the newer chips do it for you, automagically! Speed-margining, to the edge of the silicon envelope, subject to voltage/power/thermal/cooling limits.

Honestly, you must have a lot of spare time available, BonzaiDuck, because you're kind of almost jousting at windmills, it seems, or chasing ghosts.

I'm not saying that Skylake, Coffee Lake, and friends can't be overclocked, but heck, all of this work, to stick with lower core-count CPUs? When you could simply do a platform upgrade (if you're going to buy a new CPU anyways, delidded premium on top of that), and get MODERN "benefits" from a modern platform.

Are you for some reason wedded to a pre-Win10 OS installation or something? Is that why you're going to all of this trouble? Or some sort of sentimental reason?

I get the idea of being able to say that your rig is a "Hot Rod". I get the need for customization, to make it your own. I guess I just don't get the need to do it with, what is now "old tech". Unless, of course, back to the older OS installation or whatnot.

PCI-E NVMe 4.0 is a real thing, you know. Waaay better than bothering with PrimoCache for SATA drives.

Edit: I guess, if it were me, and I was still "stuck" on Skylake, if I wasn't also "stuck" on Windows 7 64-bit or something archaic like that, then I would move to an inexpensive, yet, modern, platform, while waiting for AMD and Intel's DDR5 / PCI-E 5.0 platforms to drop.

Something like a Ryzen R5 3600 CPU (6C/12T, 4.0Ghz all-core turbo), on a B550 mobo (if you want PCI-E 4.0 and NVMe 4.0 for your GPU and NVMe SSD, respectively), or a B450M TUF Gaming PRO-S, if you didn't need PCI-E 4.0/NVMe 4.0 just yet, because it has the quality-of-life improvements like "push-button USB flash", and 2.5GbE-T ethernet, and I think maybe Wifi too. Plus, it's an Asus, and it has beefy VRM heatsinks on the board. (That board runs around $129.99 when in-stock @ Newegg. I'm going to be building a few soon.)
Or if that's too much to pay, then an ASRock B450M-HDV or something, for $70 for the mobo. Basic abilities, but should be a solid board.

Then just worry about RAM (get DDR4-3600 for a Ryzen R5 3600, prefer 32GB), and a GPU, and a PSU to power it all, and a "designer" case, perhaps with some RGB bling to bring it all together.

I would be willing to wager a small sum of money that such a rig, would:
1) Not have the hassle of manual overclocking, although it still retains that ability, and
2) Has better performance, in productivity, and equal performance in games, if not better performance in games like 'The Division 2' and 'Shadows of the Tomb Raider' and the 'Assassin's Creed' games, all games known to prefer "more than four cores".

The cost shouldn't be more than $500 in total, give or take a few, minus the GPU.

 

[BonzaiDuck]

There's a definite reason that overclocking is mostly "dead" on newer platforms. Because the newer chips do it for you, automagically! Speed-margining, to the edge of the silicon envelope, subject to voltage/power/thermal/cooling limits.

Honestly, you must have a lot of spare time available, BonzaiDuck, because you're kind of almost jousting at windmills, it seems, or chasing ghosts.

I'm not saying that Skylake, Coffee Lake, and friends can't be overclocked, but heck, all of this work, to stick with lower core-count CPUs? When you could simply do a platform upgrade (if you're going to buy a new CPU anyways, delidded premium on top of that), and get MODERN "benefits" from a modern platform.

Are you for some reason wedded to a pre-Win10 OS installation or something? Is that why you're going to all of this trouble? Or some sort of sentimental reason?

I get the idea of being able to say that your rig is a "Hot Rod". I get the need for customization, to make it your own. I guess I just don't get the need to do it with, what is now "old tech". Unless, of course, back to the older OS installation or whatnot.

PCI-E NVMe 4.0 is a real thing, you know. Waaay better than bothering with PrimoCache for SATA drives.

Edit: I guess, if it were me, and I was still "stuck" on Skylake, if I wasn't also "stuck" on Windows 7 64-bit or something archaic like that, then I would move to an inexpensive, yet, modern, platform, while waiting for AMD and Intel's DDR5 / PCI-E 5.0 platforms to drop.

Something like a Ryzen R5 3600 CPU (6C/12T, 4.0Ghz all-core turbo), on a B550 mobo (if you want PCI-E 4.0 and NVMe 4.0 for your GPU and NVMe SSD, respectively), or a B450M TUF Gaming PRO-S, if you didn't need PCI-E 4.0/NVMe 4.0 just yet, because it has the quality-of-life improvements like "push-button USB flash", and 2.5GbE-T ethernet, and I think maybe Wifi too. Plus, it's an Asus, and it has beefy VRM heatsinks on the board. (That board runs around $129.99 when in-stock @ Newegg. I'm going to be building a few soon.)
Or if that's too much to pay, then an ASRock B450M-HDV or something, for $70 for the mobo. Basic abilities, but should be a solid board.

Then just worry about RAM (get DDR4-3600 for a Ryzen R5 3600, prefer 32GB), and a GPU, and a PSU to power it all, and a "designer" case, perhaps with some RGB bling to bring it all together.

I would be willing to wager a small sum of money that such a rig, would:
1) Not have the hassle of manual overclocking, although it still retains that ability, and
2) Has better performance, in productivity, and equal performance in games, if not better performance in games like 'The Division 2' and 'Shadows of the Tomb Raider' and the 'Assassin's Creed' games, all games known to prefer "more than four cores".

The cost shouldn't be more than $500 in total, give or take a few, minus the GPU.

I appreciate your thoughtful input. I may build a new computer every three or four years -- a more frequent exercise back in the millennium's first decade when I had five other family members to take care of. But now, one brother has moved away with my sister-in-law -- that was about four computers-worth; Moms at 95 can't even work the remote for her TV and gripes about having to sit up straight to eat dinner. My other brother who "lives" upstairs doesn't even want to work with his desktop anymore -- he's become an Apple cell-phone junkie. Can you believe this? He once learned how to program in the '90s. I'd upgraded his PC to Windows 10 four years ago -- on a separate boot-volume -- expecting him to reinstall his software -- e-mail etc. -- as he became familiar with it. All this time, he'd been expecting ME to do all of that for him, so now . . . I've got to install Win 10 again to his system as an "upgrade" to the Win 7 partition/volume. Extra work for BonzaiDuck! I feel like Johnny Depp as "Gilbert Grape".

Every time I build a system, I take six months (SIX MONTHS!) to research and plan. It's a "special project" kind of thing. The original Skylake was the best-planned system of all.

So when I zapped the USB controller with a vaping-pen static-charge, planning to have my taxes done a month or two ahead of deadline, I was in a panic. Turbo-Tax only installs on Windows 10 -- period. Those anxieties drove all my actions since losing the motherboard (ASUS, by the way, will have a replacement board returned to me by Thursday this week, according to their e-mail update). Truth be told, I could've solved my problem with an outlay of less than $200. And -- get ready for this! -- I finally sat down with Turbo-Tax last week, and it was all done to my anal-retentive satisfaction in . . . . . THREE HOURS.

I don't even "need" to have the original Skylake system over-clocked. It -- or the new Kaby processor either one -- is fast enough. But for 15 years or so, this has been the one, single addiction that won't shorten my life, even if it takes up a few extra hours.

The Z170 boards and earlier -- recalling my Z68's -- all provide that "auto-over-clocking" ability, but I didn't like the results. They all come out over-volted and running hotter than needed, even if it's acceptable. And -- I like to have "control" of everything, a corollary to "knowing what's actually going on or happening". If I wanted to "find out" what 9-gen or 10-gen processors and board would do for me, it would still have a degree of uncertainty which I didn't PLAN on for this year.

Since I've explained all this, and since you know that I'm not a noob or just another mainstreamer, let me tell you what I know about "users at large".

My dentist -- also a friend -- would maintain his office and home computers himself, but he would never over-clock, and he never built one from scratch. Too much trouble, he'd say, and he's right. He's got a "practice"; he plays tennis; he's in the middle of a marital breakup; he has to walk his dog. He doesn't have time for this. Paul buys corporate-surplus-asset turnovers -- Dell Xeons -- which he says will idle at 50C. Maybe -- he buys better power supplies for them, but that's the extent of his needs and desires.

My friend in Virginia is a retired Navy electronics technician. He's always doing the same thing as Paul, and the reason I acquired my laptop which was originally released 13 years ago. He "keeps up" with tablets and laptops, but he's not going to get a Comet Lake anytime soon. Like me, his eyes are going bad.

My high-school sweetheart -- a widow who lives over in Yorba Linda -- was telling me via e-mail last week that her computer kept reporting the wrong time, or would get out of whack very soon after she reset the time in Windows. I figured it was her CR2032 battery. I'll bet her computer is 10 years old. BUT! She uses it daily!

So I can see how I might lean toward building anew every year or two, but I've got my garden, I'm stealing recipes from Stanley Tucci and running a full-service kitchen, I have to pay all the bills, get a 135 lb old lady to my car for every one of about 10 doctor appointment annually, make my own doctor appointments, do all the grocery shopping, manage the car repairs for myself and Bro -- you see? I look at it this way: If I have "lots of time", I don't have a "plan" this year for a latest-gen computer. Of course, with the tax returns, the refund on a mortgage just paid off, and that $1,400 stimulus, I could certainly afford it, even for the panic spending to get this one back in service. But I hadn't planned it!

On the NVME PCI-E version 4 improvements, I did indeed look into that. I checked a review here at Anandtech, and unless I misread it or missed something with my speed-reading quick-scan approach -- the touted 7GB/second sustained throughputs weren't showing up very well in the benchmarks. Was I wrong?

You tell me . . . .

Anyway -- the clocks. The clocks. I must be getting better at this, even if the skill-set is becoming obsolete, as you say. I've got frequency and voltage pairs that are stable for 4.6, 4.7, 4.8, 4.9 and 5.0 now. For the time being, 4.7 is fine, and I did the tweaks so that the fixed-voltage setting -- for now -- is rock solid. I think I figured out how I got confused for making the Adaptive Mode work, and with a single experiment on some clock speed like 4.6, I can go back and convert all those fixed voltages. Anyway, 5.0 just doesn't seem feasible for 24/7. It pushes the voltage requirement closer to 1.42V and the temperatures peak out at 92C. I don't like it.

What's the maximum clock speed on these Octo and Deca-core Comet Lake jobs, anyway? I see the stock turbo speed for an i7-10700 is 4.8Ghz. OK -- your point is taken. But I don't NEED 8 cores! I don't keep up with the latest games! And 4.8 is still easy on a gen-7 processor. 4.9 promises to be easy. It was just a little harder on the Skylake getting to 4.8 so that I felt comfortable with it.

Of course, posters to this or other threads seem to agree. It doesn't make a lot of difference. I just build my systems as though it could, would, or might make a difference, even if I find out that it doesn't.

Like I said -- here, there or on other threads -- this is a harmless addiction. I can't help it, whether I refurb an old-gen system or build a Comet or a Rocket Lake.

I can't HELP MYSELF. Surely -- you can see that? I'm an old dog with some stale tricks, and I tend to stick with what's known as opposed to keeping up with technology which requires fewer tricks. Don't worry! I'll do it -- this year or the next! But -- not the "latest" processor and board. Probably "last year's" processor and board. I may follow the myth about "mature BIOS" histories, or I like to wait and see what sort of troubles or difficulties occurred with more eager PC builders.

 

[BonzaiDuck]

Well, like I said -- "I'm back,, I'm bad, I'm on top . . . "

I can SEE how I can run this at 5.0Ghz, especially with the "Negative AVX Offset" feature.

But bragging rights weren't my goal with this, and I can save tuning the highest clock settings as something to do any time later. 4.8Ghz runs cooler than my friend's Xeons with their stock settings. The AVX Offset down-clocks it to 4.7.

All those poor suckers and wannabees in the customer reviews! Oh! It runs too hot! I can't even get it above the stock spec speed! Just Teww-ible! Teww-ible, I say!

With the De-lid and Re-lid, it barely breaks 65C. And apparently, that means I can volt it at between 1.25 and 1.28V, as opposed to maybe 1.32 or more. Schweet!

Gotta hand it to the OCCT developer. That program will trap errors in a half hour that you wouldn't find with LinX . . . .

 

[ThirteenthDominion]

But bragging rights weren't my goal with this, and I can save tuning the highest clock settings as something to do any time later. 4.8Ghz runs cooler than my friend's Xeons with their stock settings. The AVX Offset down-clocks it to 4.7.

All those poor suckers and wannabees in the customer reviews! Oh! It runs too hot! I can't even get it above the stock spec speed! Just Teww-ible! Teww-ible, I say!

The silicon lottery is a very real thing, no two cpus are the same. You claim to have a chip that surpasses even a higher binned chip (xeon), so it's no surprise you don't have the experience of lottery losers.
Not to mention that you've delidded and LM'd which has quite considerably augmented the thermal characteristics of your cpu, relative to a stock 6700k.

 

[BonzaiDuck]

The silicon lottery is a very real thing, no two cpus are the same. You claim to have a chip that surpasses even a higher binned chip (xeon), so it's no surprise you don't have the experience of lottery losers.
Not to mention that you've delidded and LM'd which has quite considerably augmented the thermal characteristics of your cpu, relative to a stock 6700k.

Well, I was mildly disparaging the folks who were so disappointed with the Kaby Lake thermal performance. We've been unsatisfied with straight-from-the-factory performance since the Ivy Bridge model-line. I suppose if people were so focused on over-clocking prospects, they would have been serious enough previously to know all this by now. I'm really just patting myself on the back for being willing to take the trouble for making the chip see its greater potential.

I decided to leave my XMP settings as-is for the TridentZ's and attempted booting to 4.9Ghz after changing the AVX Negative Offset to 2 -- it would then down-clock to 4.7 as it had with the 4.8 core ratio. I've been fiddling with this daily for a week or so, and the OCCT:CPU [SSE] test didn't trap the error after maybe 30 minutes; I got a reboot with critical error code 41 -- first time since I started.

The data-plot in the graph I put in post #2 shows a disparity between "best and worst" of about 0.07V for the effective VCORE, and I would think the observable drooped value under load is more worthy of any comparison. I didn't want to mess around, and punched in a total of 20 1-millivolt notches just for starters. The loaded value now sits at the 4.9Ghz "potato" voltage, but it's rock-stable so far. My guess is that it already has headroom in my twenty-clicks to show itself lower than what the Potato number gives, but at this rate -- a maximum voltage about 0.1V below the accepted limit for these 14nm chips, I'm very pleased. The temperatures under the ongoing OCCT:CPU/SSE test are all in the range of 61 to 65C. Since the AVX and AVX2 tests will drop the processor back to 4.7, I don't expect to see anything wrong there, either.

So -- yeah -- after all this trouble replacing primarily the motherboard but adding the refresh chip and a swap for a two-stick kit of high-density RAM -- I think it's just wonderful. Better than before . . .

 

[BonzaiDuck]

I'm pretty sure I've developed a reputation for TLTR, prolix screeds and rants -- here, on P&N, Garage and so forth. So . . . blah, blah blan.

VirtualLarry has a point, but my main objective was to get my failed mobo back up and running, and I succumbed to the temptation born of curiosity about a refresh chip. Last year, I upgraded my 26-year-old SUV for rear-view night-vision camera, voice navigation and recognition, new Polk-Audio replacements and double-Blue-tooth receivers for two sources of "massive" MP3 record-album sources. I'm an old 20th century rocka-roller, and I "just had to catch up" on music through the 90s. In January, I must've spent $300 on music downloads. Call it "pent-up-demand" spending -- this episode with the dead motherboard follows the same pattern.

Anyway, for being sidetracked over a week or so for the tweaking, I now have my taxes done and and "I'm back and running". What to do with the Kaby Lake? Well, I think I can run it at ether 4.9 or 5.0 with the AVX-Offset merely set to "1". To do that, I have to run the VCCSA at 1.30V and the VCCIO at 1.25V, or so the current ongoing OCCT AVX2 stress-test seems to prove. 4.8, 4.9 and possibly even 5.0 appear feasible for 24/7 running. If the IMC and Sys-Agent voltages seem a bit high, I suppose I can let the chip follow whatever life-span that gives, then do as VirtualLarry suggests.

If it passes 2 hours with OCCT:CPU/AVX2, minor tweaks will get it past a Prime95 and LinX day-long run.

Jousting at Wind-Mills? Well -- not for too long. Not nearly long enough for Cervantes. Just a detour in time for me.