Am I on the right path? 3570k w/ Asrock z77 Extreme4

[KillerXtreme]

Hey guys! Been a lurker for a while and last night I decided to re-evaluate my overclock as I've been having crashing issues while playing ArcheAge and streaming with OBS. I've been reading around that OBS tends to crash with an overclock due to the way OBS encodes. So I updated my bios/drivers to the latest and started from scratch. Using this guide as my guideline. I'm mostly coming here just to make sure that I'm headed in the right direction and I'm still within good moving distance to push just that bit further.

First some info...
CPU - i5 3570k
Mobo - Asrock Extreme4 z77
Cooling - Corsair h80
Ram - 8gb Samsung 'Magic' ram
PSU - OCZ EliteXStream 800

So onto the overclock. I've shooting for a 4.6ghz Overclock using Offset mode. As well as LLC at Level 2. So far I've been slowly increasing my 'Additional Turbo Voltage' while leaving my Offset Voltage at +.005v. Currently I'm at +.086v "Turbo Voltage" making my Core Voltage under load 1.298v-1.305v while my temps have been maxing @ 90-95c (while running prime95) This was the last time I tested and wanted to come here and see if I'm still on the right path and should pop it a bit more. The previous Turbo Voltage before +.086v I ran Prime95 for an hour before it errored on me. At +.086 it ran for a little over 5 hours before it errored on me. (Which is a huge leap compared to my tests previous to that) I'm guessing in that the next setting which is +.090v would stabilize even more (if not make it completely stable) but wanted to make sure that I'm still good with my voltages and temps. (obviously temps will never get that high while gaming) Temps are a little high also because I need a new case, and could probably reapply thermal paste to my cooler as its been a while since I have cleaned everything out.

If you need anymore information feel free to ask and I'll provide as needed.

 

[Yuriman]

It looks like you have a good grasp of the process. Lowering temperatures may help with stability, but likely you'll just need to keep increasing voltage until it doesn't crash anymore. ~1.3v is considered a relatively safe voltage, if at the upper bounds of safe, and low 90's C are around the upper bound of a safe temperature, and about what I'd expect at your voltage/clock.

You might consider backing off to 4.5GHz rather than going above 1.3v. Getting every last bit of performance is fun, but ultimately you'll not likely miss that last 100mhz. I assume since you followed the guide, you disabled the proper C-states?

 

[KillerXtreme]

It looks like you have a good grasp of the process. Lowering temperatures may help with stability, but likely you'll just need to keep increasing voltage until it doesn't crash anymore. ~1.3v is considered a relatively safe voltage, if at the upper bounds of safe, and low 90's C are around the upper bound of a safe temperature, and about what I'd expect at your voltage/clock.

You might consider backing off to 4.5GHz rather than going above 1.3v. Getting every last bit of performance is fun, but ultimately you'll not likely miss that last 100mhz. I assume since you followed the guide, you disabled the proper C-states?

Its actually not crashing, only Prime95 is erroring. Always on Core2, never any of the other cores. I've been gaming with a 4.6ghz clock for a few years since I've built the machine. Had to re-overclock it a couple times with updates and one odd glitch that reset everything. I just wanted to clear up OBS + Archeage crashing and verify that it was in fact not because of my overclock.

And yes I've got the proper C states disabled. I followed that guide to the T. Just wanted to verify and make sure that I was still within safe boundaries as I've never gone this far with an overclock before.

 

[BonzaiDuck]

Its actually not crashing, only Prime95 is erroring. Always on Core2, never any of the other cores. I've been gaming with a 4.6ghz clock for a few years since I've built the machine. Had to re-overclock it a couple times with updates and one odd glitch that reset everything. I just wanted to clear up OBS + Archeage crashing and verify that it was in fact not because of my overclock.

And yes I've got the proper C states disabled. I followed that guide to the T. Just wanted to verify and make sure that I was still within safe boundaries as I've never gone this far with an overclock before.

If you're not using it, do everything you can to assure onboard Intel graphics is disabled or inactive.

Try an LLC setting that still leaves about 20 mV of vDROOP. You would be able to see this in a program like HW-Monitor: Run a stress-test for 10 or 15 minutes and record the maximum "turbo" voltage in HWMonitor -- which should show up just at the end of the test -- an "unloaded turbo voltage." Run the stress-test again, reset the "Min and Max" records, and the minimum voltage under load would be recorded. The difference is mostly vDroop. LLC changes the amount of vDroop for voltages under load.

Also, as I recall, the PLL Voltage defaults to ~1.8V. See if you can back it down to 1.65V, and test stability. This actually may improve stability, but it also reduces a voltage and therefore the thermal wattage. Some folks have actually found their "sweet spot" around 1.55V, but the prevailing opinion inclines toward 1.65V.

When you say the temperatures have been "maxing @ 90 - 95c" are you referring to a maximum value trapped by the column "Max" in HWMonitor? Is it only one core? How far does this maximum exceed the average of all cores?

 

[coffeejunkee]

Check for WHEA errors in event viewer as well.

If OBS encoding causes crashes, wouldn't it be better to stresstest with OBS?

Also, if it's not too much, I have an Asus mobo which doesn't have turbo voltage setting, what's the difference between just using the normal offset option instead of the turbo voltage option?

 

[BonzaiDuck]

Check for WHEA errors in event viewer as well.

If OBS encoding causes crashes, wouldn't it be better to stresstest with OBS?

Also, if it's not too much, I have an Asus mobo which doesn't have turbo voltage setting, what's the difference between just using the normal offset option instead of the turbo voltage option?

I'm pretty sure I read through the OC Guide cited by the OP some time ago. I can't be sure if I saw it there, in some other OC'ing guide, or came to the conclusion by reviewing an old Anandtech article on vOffset, vDroop and LLC. I came to the conclusion that it is best to leave the Offset voltage close to 0, or to shoot for a setting of +0.005V.

What I do know -- I discovered this "Extra Turbo Voltage" item in the P8Z68-V Pro Power-Management menu. And it was very quickly evident that you could adjust the Offset and "Xtra" settings together to obtain the same stability effect. Ultimately, with a single setting for Offset, I was then able to adjust the "Extra Voltage" setting to achieve a stable load voltage and pass the stability tests.

Works for me! Otherwise, folks using this or similar motherboards may adjust only the Offset voltage, and the "Extra" item would remain on "Auto." I'm not 100% clear myself as to what the implications of that are. I'm only 100% clear that I have a 100% stable system @ 4.7 Ghz with a combined setting (Offset and Extra) adding up to ~ 50 mV.

Also -- if the guide cited by the OP is the one I recall for this -- there may be demonstration screenies showing the effect of increased Offset voltage on the EIST idle voltage -- maybe suggesting that a minimal setting for Offset is the way to go when using any LLC setting.

 

[Yuriman]

This might be useful to the OP, regarding choosing a LLC setting:

http://forums.anandtech.com/showthread.php?t=2276792

 

[KillerXtreme]

If you're not using it, do everything you can to assure onboard Intel graphics is disabled or inactive.

Try an LLC setting that still leaves about 20 mV of vDROOP. You would be able to see this in a program like HW-Monitor: Run a stress-test for 10 or 15 minutes and record the maximum "turbo" voltage in HWMonitor -- which should show up just at the end of the test -- an "unloaded turbo voltage." Run the stress-test again, reset the "Min and Max" records, and the minimum voltage under load would be recorded. The difference is mostly vDroop. LLC changes the amount of vDroop for voltages under load.

Also, as I recall, the PLL Voltage defaults to ~1.8V. See if you can back it down to 1.65V, and test stability. This actually may improve stability, but it also reduces a voltage and therefore the thermal wattage. Some folks have actually found their "sweet spot" around 1.55V, but the prevailing opinion inclines toward 1.65V.

When you say the temperatures have been "maxing @ 90 - 95c" are you referring to a maximum value trapped by the column "Max" in HWMonitor? Is it only one core? How far does this maximum exceed the average of all cores?

Yea I've already got the onboard disabled.

The voltages that I gave are the average of the under load. Running now at +.09 turbo voltage and the norm is 1.304v with a max of 1.314.

I'll have to try that with the PLL, I've read that lowering it can sometimes increase stability, however I wanted to try to get the vcore as stable as it'll get. I've been reading a lot of people with the same setup getting close to 1.3-1.35 for stability so I figured a voltage of 1.304 was around the ballpark for being stable.

Those temps I'm giving are a mix of all the cores. Some cores max were about 92c, others 95c, only one of them got to 97c as a max, but usually saw hovering around 93c (which was actually Core2, that kept stopping in Prime95, temps might be an issue for that core)

Yea, I'm going to do more testing with OBS to see how things go. I actually played ArcheAge last night for several hours without crashing or anything. Which was a nice change. Now to try it while streaming with OBS.

Also, if it's not too much, I have an Asus mobo which doesn't have turbo voltage setting, what's the difference between just using the normal offset option instead of the turbo voltage option?

The difference between turbo voltage and just the normal offset is that the normal offset changes the voltage regardless if your under load or not.
So if your voltage while idling is say .9v and you add a .09v offset your new volt will be .99v while idle (as well as adding that .09v per multiplier when under load). While the turbo voltage only adds the offset*multipler while you are under load so your idle voltage stays low to save power.

Also I've got another thing to ask if anyone has noticed. Oddly when I overclock for whatever reason Firefox, and only firefox, seems to start up a lot slower. It'll come up right away, but with a white screen for about 3 seconds before it pops to the normal firefox welcome screen. (I've got my entire system on an SSD so this shouldn't happen) Every other program I have, chrome what have you, starts up instantly. Hell before I started to overclock firefox would start up instantly. There is a setting or maybe something directly to do with my overclock that is making firefox start up like this, I just don't know what.

This might be useful to the OP, regarding choosing a LLC setting:

http://forums.anandtech.com/showthread.php?t=2276792

Interesting post, from what I've read and understood about LLC is that while under load it is suppose to compensate for the droop. The one thing I don't quite understand is the last thing he says there "What I gather is, LLC is actually counter-productive on my board. Enabling it at higher levels only serves to raise my minimum stable load voltage without doing anything to help my idle and low-load voltages." Higher levels I assume is level 3, 2, and 1, being on other boards 50%, 70%, 100% compensation. They are suppose to overcompensate your voltage on load (because that's where the droop happens) while doing nothing at idle or low-load voltages because its not drooping to compensate. So that post is interesting in showing how each setting compensates. However I don't quite get what he means by being counter-productive. Seems LLC is doing exactly what its ment to by his numbers.

Also I'm using Level2 in case I didn't state it earlier.

 

[BonzaiDuck]

If I remember correctly, the earlier stages of my OC'ing pursuit began or ended with a minimal LLC setting -- I think it was described in BIOS as "25%." If I observed vDroop as 60+ mV, the early 25% setting reduced it to ~40 mV. Whichever setting I ultimately chose, it still allowed 20mV of vDroop.

And the final LLC adjustment ended whatever instability would cause my stress-test to end prematurely. It may not have been the ONLY final adjustment, but anything after that was exclusively an "Extra Turbo voltage" tweak of 2 to 4 mV.

It is also true, however, as to what is explained in the link on the topic. But this was also explained in an article I cite habitually, published around December 2007 or 2008 by Anandtech pertaining to the 9650X chip, or whatever X chip was precursor to the Yorkfield cores. There was an in-depth discussion of vOffset, vDroop, LLC, VID and VCore.

The article's main caution -- using too much LLC in the upper ranges at or above a "sane" CPU VCore -- implied that VCORE would exceed VID, and second, that the harmonic fluctuation, at the point where load stops but before voltage settles back to an idle value, will overshoot VID by some un- measurable level; the same fluctuation without LLC will not spike that high.

In a broader context, this could only be a risk if you have already exceeded a level of voltage considered safe for the processor.

Also, there have been various improvements in voltage regulation with motherboards, components and the CPU since that article was published.

 

[Yuriman]

However I don't quite get what he means by being counter-productive. Seems LLC is doing exactly what its ment to by his numbers.

If you take a close look at my numbers, with LLC at its lowest setting, my lowest stable 4-core load voltage was about 1.250v, while my peak voltage (1 core load) was about 1.300v. Turning up LLC did now allow me to reduce my peak voltage under a 1 core load, it only succeeded in causing my chip to need more voltage under 4-core loads, where droop was previously lowering the voltage. As a result, I needed overall higher voltages for the same clocks.

This was completely unexpected and counterintuitive.

 

[BonzaiDuck]

Back on topic, tho' not discouraging the current focus on LLC -- there IS a way for the OP to reduce load temperatures on his 3570K chip. It has a one-time up-front risk; it's certainly "extra trouble;" but I think the record shows that temperatures can be reduced by 15 to 20C.

 

[KillerXtreme]

Back on topic, tho' not discouraging the current focus on LLC -- there IS a way for the OP to reduce load temperatures on his 3570K chip. It has a one-time up-front risk; it's certainly "extra trouble;" but I think the record shows that temperatures can be reduced by 15 to 20C.

According to the guide I linked in the OP, they suggest the min voltage for PLL to be 1.71, however you suggest 1.65 (which actually for my bios/mobo is 1.653) They even say to raise PLL to 1.89 or lowering to 1.709.. I'm going to try your suggestion of 1.653 and see how that goes.

 

[BonzaiDuck]

According to the guide I linked in the OP, they suggest the min voltage for PLL to be 1.71, however you suggest 1.65 (which actually for my bios/mobo is 1.653) They even say to raise PLL to 1.89 or lowering to 1.709.. I'm going to try your suggestion of 1.653 and see how that goes.

Well, I should keep a card-file on articles and topics I come across. I only know that it was an OC Guide with "questions and answers" for both Sandy bridge and Ivy Bridge cores-- skt-1155. I think I first set it to 1.70V, stress-tested for several hours, then changed it to 1.65V and tested again. It had no adverse effect on stability at that level. [And .. lowering it below default may improve stability. I may have seen somewhere that raising it "doesn't help," but cannot guarantee so.]

 

[Yuriman]

Lowering PLL voltage has no impact on stability for my chip at all at the clocks I've tried, even at its lowest selectable setting in BIOS. Certain, it isn't hurting it to lower.

 

[KillerXtreme]

I can say one thing, I streamed for over 6 and a half hours yesterday without a single crash after I lowered it to 1.653 which is new for me, as it tends to crash more often than that. Lowering the PLL definitely does lower the temps quite a bit which is nice. It didn't seem to effect stability.

 

[BonzaiDuck]

I can say one thing, I streamed for over 6 and a half hours yesterday without a single crash after I lowered it to 1.653 which is new for me, as it tends to crash more often than that. Lowering the PLL definitely does lower the temps quite a bit which is nice. It didn't seem to effect stability.

Why does your last sentence there seem contradictory? Maybe you mean that lowering the PLL voltage didn't make the system unstable.

Truth be told, I was at least as much interested in the temperatures.

I was -- and still am -- a bit puzzled. Why did the default of 1.8V prevail? I think that's what you get when you choose "Auto" for PLL Voltage. Perhaps it needs a longer discussion of "what does PLL Voltage 'do'?"

I'm guessing that a higher PLL voltage might serve to stabilize an overclock setting like bCLK. If I'm only running bCLK at spec . . . well, there you are . . .

I think I'm just making shots in the dark here . . .

A little more web-exploration. There was actually a forum exchange about this.

It seems that there's no adverse effect to dropping PLL voltage on an IB-K processor to 1.60V. Some other indications -- it can be lowered as far as 1.5.

But you'd want to test stability with a new setting, anyway.

Also, greater indications of my hunch -- that PLL voltage increase might eliminate "jitter" or instability with higher levels of bCLK. But SB and IB overclockers generally don't fiddle with bCLK anymore. On the other hand, higher PLL voltage may be necessary for extreme RAM overclocking.

http://www.overclock.net/t/1305797/droping-pll-voltage-on-3770k

NOw -- there is SUPPOSED to be a "sweet spot," but it either varies, or it is a myth. Look at it another way: suppose you'd decided that you'd found the highest clock-speed for your processor for which voltage and temperature still left you "comfortable." A drop from 1.8 to 1.5V is 0.3V. For VCORE that's a really big drop in voltage with some serious temperature implications. For PLL voltage, I wouldn't know. I only know I had a net reduction in my highest temperatures after dropping down by only 15 mV on that setting.

Here's an OC'ing guide for Ivy Bridge that even explores LN2 cooling and clocking:

http://www.overclock.net/t/1247413/ivy-bridge-overclocking-guide-with-ln2-guide-at-the-end

I think the implication for bCLK stability from higher PLL Voltage settings is somewhere in that guide.

ALSO -- speaking only from Sandy Bridge experience, but it should apply equally -- try to avoid enabling "PLL Overvoltage." It will also increase some voltages and therefore thermal wattage. A lot of OC'ing guides tell people "right away" to enable this setting at clocks of >= 4.5. I discovered it was NOTHING to keep stability at 4.7 and disable it!! Not relevant to your Z77 BIOS, but enabling PLL Overvoltage on my Z68 made "wake from sleep" problematic until there was another BIOS revision.

 

[KillerXtreme]

PLL Overvoltage is disabled. From what I've read around was that PLL can help an overclock by either lowering or highering it. At auto for me at least I know my overclock was unstable by running Prime95 for ~5h and a core stopping. When I lowered it to your suggested 1.653, I switched over to a real world test. Playing ArcheAge and streaming to twitch while having all my usual programs running I was able to stream over 6 hours before OBS decided to crash with its usual H.264 error (common crash is because of overclocks) So I highered the voltage to 1.701, again doing my realworld testing. I was able to stream nearly 12 hours (before I had to go to bed) The only crashing that was happening was the game but only after 11 hours of gameplay, and at that point it could have just been the game itself with issues.

I must say at this point I'm quite happy with my overclock, my voltages, and its temps. (~65-70c while gaming) I do have to say I do need to get a new computer case, one that will give me better airflow than my current one with is a InWin F430, that I need to keep the side door off because it blocks airflow even more with the H80's Fan in the back of the case.

So if you guys got a good recommendation for a case with good airflow thats reasonably priced do tell... I was looking at the HAF cases, but still looking around.

 

[monkeydelmagico]

.

I must say at this point I'm quite happy with my overclock, my voltages, and its temps. (~65-70c while gaming) I do have to say I do need to get a new computer case, one that will give me better airflow than my current one with is a InWin F430, that I need to keep the side door off because it blocks airflow even more with the H80's Fan in the back of the case.

So if you guys got a good recommendation for a case with good airflow thats reasonably priced do tell... I was looking at the HAF cases, but still looking around.

As well you should be. I had a similar setup 3570k w/asrock pro-4 and could run 4.6 @ 1.28v prime95/IBT. I found the rig to be more stable at 4.4 though.

It's good you got the temps under control. 90c+ is high.

Lots of great cases, HAF is nice but can be a bit noisy and dusty. Flows the most air though. I picked up an N400 recently and can tell you to avoid it. Antec 300, corsair carbide, and nzxt source are all good. Fractal R4 is my favorite....I shoulda bought it.

 

[BonzaiDuck]

PLL Overvoltage is disabled. From what I've read around was that PLL can help an overclock by either lowering or highering it. At auto for me at least I know my overclock was unstable by running Prime95 for ~5h and a core stopping. When I lowered it to your suggested 1.653, I switched over to a real world test. Playing ArcheAge and streaming to twitch while having all my usual programs running I was able to stream over 6 hours before OBS decided to crash with its usual H.264 error (common crash is because of overclocks) So I highered the voltage to 1.701, again doing my realworld testing. I was able to stream nearly 12 hours (before I had to go to bed) The only crashing that was happening was the game but only after 11 hours of gameplay, and at that point it could have just been the game itself with issues.

I must say at this point I'm quite happy with my overclock, my voltages, and its temps. (~65-70c while gaming) I do have to say I do need to get a new computer case, one that will give me better airflow than my current one with is a InWin F430, that I need to keep the side door off because it blocks airflow even more with the H80's Fan in the back of the case.

So if you guys got a good recommendation for a case with good airflow thats reasonably priced do tell... I was looking at the HAF cases, but still looking around.

The matter of "PLL Voltage" spurred me to some more web-searches, and I came across the same material that led me to lower the default 1.8 or "Auto" value. Apparently, only for extreme overclocking does an INCREASE in PLL voltage help. Lowering it seems to improve stability.

This PLL voltage seems related to the components that have an input signal and an output signal that must be kept in phase, even as there may be a frequency change. It affects "bus" stability -- QPI, bCLK and memory. Or so I glean from my light reading. If one were going to increase bCLK, then a higher PLL voltage might help. If not -- then lowering the PLL voltage seems to offer more stability.

The "PLL Overvoltage" actually over-rides the PLL voltage setting. Lowering PLL voltage may result in slightly lower power requirement -- therefore thermal wattage -- to the tune of about 5W -- not a lot, but something.

Well, here's the result of an experiment of sorts.

My OC settings were for 4.7 Ghz, with RAM set at DDR3-1866 spec -- voltage, timings and speed. I'd been experimenting with the XMP settings, which -- in the XMP motherboard item, showed "2N" for the command rate.

You always get a little boost with a 1N command rate, and the "go-ahead" was given by G.SKILL tech support when I asked: "Sure. You can run them at CMD=1 . . . " A lot of other folks, some with Samsung, some with G.SKILL, had no trouble with this. And I didn't have trouble with it at lower speeds (4.6 Ghz), so I took the approach of "set it and forget it" for the current 4.7 overclock, after changing my manual RAM settings to the XMP profile.

First thing I discover, the ASUS board may "show" 9-9-9-24-2N for the XMP profile. But setting the command rate to "Auto" instead of 1 or 2 shows it operating at 1N after rebooting to BIOS!

I also decided to see what would happen if I dropped the PLL voltage from 1.653 to 1.60V.

Then, I ran a stress-test at the 4.7 Ghz clock speed. I had been through this to certify stability: 50 interations of either IBT or "affinitized" LinX -- with the higher PLL voltage settings.

So I gave it an hour of Prime95 sFFT -- no problem. Immediately following, an hour of Prime95 Blend -- no problem. Immediately after that -- 20 iterations of affinitized LinX. With the LinX, the system BSOD'd after less than 5 iterations!

Stop code: 09C -- suggesting an increase in QPI/VTT or VCCIO voltage (for the IMC).

Returned to BIOS, reset PLL voltage to 1.653V. The LinX crashed after 18 iterations.

Stop code: 09C.

At that point, I dropped the command rate back to a manual setting of 2.

Eventually, I find that raising the VCCIO from 1.09 to 1.13V allows running the RAM at CMD=1 or 1N, setting the RAM timings to all "auto" under XMP.

I have more tests to do, but all this seems intuitively consistent with what you might expect. Other sources suggest that one could lower the PLL voltage to 1.50+, but I suspect you'd still find yourself adjusting the VCCIO upward. And, of course, the limit to safe VCCIO is 1.2V, or so it is recommended.

And I can see why setting PLL-Overvoltage to either "auto" or "enabled" might stabilize things. But I can also see that you could likely get by without it, if you accept "comfortable" overclock settings. I KNOW I can "get by without it" at 4.7, command rate=1 or 2 and PLL voltage = 1.653V. It's also useful to know how close one might come to instability without knowing it.

 

[BonzaiDuck]

I don't want to seem as though I'm promoting anything. Around 2008, I purchased my first HAF 922 midtower. Over then next few years, I bought two more of them.

I always think folks are snickering if I speak of the virtues of 200mm fans. First -- they have lower static pressure -- not coveted by some. Second -- if you run them at full bore, you will either get dust (eventually) or you will find the airflow you wanted is restricted with filtering. Third -- some of them use sleeve bearings, and don't have as long of an MTBF.

But! They can be quiet. Further, you can really pressurize a case like the HAFs with just a couple 200mm fans, provided you block off the unused ports, or use all of them for intake.

If you can control all the fans in your case from the motherboard such that they spin up under higher CPU temperatures and otherwise run ~300RPM at idle, you're going to get less dust -- even without filtering.

The HAF is a pretty cheap case -- I mean "inexpensive." It doesn't have a lot of "bling" despite CM's efforts to give you an LED fan in the front port. Moreover, I'd replace the CM fans with BitFenix Spectre Pro's, if you don't mind the trouble of a little modding to mount them. Currently, my sig-rig has a (non-LED) NZXT 200mm spec'd at 166CFM for the side-panel, and the CM LED fan in the front. I'm soon going to replace the latter with a Spectre Pro -- spec'd at 144 CFM.

With thermal control of the front-panel fan, I'd even remove the mesh screen filter that comes with the case.

What sort of options you have for water-cooling, I'm still trying to work out in my mind before ordering a different case. I'd rather water-cool with a HAF if I can. I'd like to try square radiators with 180mm fans. I'll have to think about it some more . . .

 

[BonzaiDuck]

Just for anyone interested. Like I said -- "not an electronics engineer." So I pick up information from various sources.

Apparently, there is a standard reference across other electronics gadgetry and beyond Intel CPUs: 1.65V.

The idea communicated from the same source: "Better keep it at 1.68V -- for certain reasons" -- particularly the (apparently standard) increments of adjustment and their effect on the actual voltage.

There is obviously a "sweet spot," but this voltage seems to correspond to what some had "found" on their own. If someone can lower it to 1.5V, and they apparently can -- you don't have any idea about "effects" that they failed to uncover.