Overclock.net - An Overclocking Community - Reply to Topic

Thread: UPDATED: ASUS Z370 and Z390 Motherboard Series - Official Support Thread Reply to Thread
Title:
Message:

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Password
Please enter a password for your user account. Note that passwords are case-sensitive.
Password:
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:

Log-in


  Additional Options
Miscellaneous Options

  Topic Review (Newest First)
11-12-2019 01:17 AM
mattxx88 i have some issue with my mainboard (Maximus X Formula), regarding RAM settings

i bought a new viper steel 4400 cas19 kit, and few 9900ks to test them.
I did a cmos, then installed kit. then applied xmp settings with all other stuffs @def = NO BOOT

then i tried again, CMOS, all settings @def (yes also ram) i only raised up vdimm 1.45v, vccio 1.200v, vccsa 1225v to see what speed and timing my board assign automatically

the resault was: 3600mhz 14-14-14-33

seems like mainboard automatically gives too extreme settings, thats why i cannot boot @4400mhz

last try i did was to load my oc profile with a ram speed settings @4000mhz cas17. with this profile i could reach 4000mhz cas 16-16-16-32 with new ram, but nothig over.
is there any settings on Asus to make the board load with more "relaxed" timings automatically?
11-11-2019 08:16 PM
Jpmboy ^^ Great post!!



the thing about overshoot (and under shoot), or more accurately, load-transition spikes, is that they occur on the microsec time scale. Intel's spec for virus mode V_ovs is ~ 200mV (eg, p95 -like load) at stock voltage. V_ovs spikes will cause erosion over time. For a gaming rig, it is a good idea to have some droop on the subject rail (115x socket = vcore, x99, x299 = vccin). BUt the susceptibility really varies by the product line.
11-11-2019 05:00 PM
Falkentyne
Quote: Originally Posted by postem View Post
Thanks as I said my comprehension on subject is meager so I got it wrong.

Now about the benefits of having a more stable vcore across loads, and the fact the benefits lessen the higher you go, what are the issues we risk with having overshoot voltage, it's the risk of constantly going over what is safe voltage and getting more degradation?

Also another question, the overshoot is the same regardless of vcore set or it's a function of the actual value dialed in bios?
Overshoot is difficult to explain and it requires a background in Electronics to really understand, but it has to do with capacitors (voltage charging and discharging) not being able to react to sudden current loads changing, which is very hard to explain here.

I'm not fully sure if this is an accurate explanation but you can read this.
https://en.wikichip.org/wiki/load-line_calibration

There was another page which went into FAR more detail about this, explaining it has to do with a "gate" opening and closing a connection, and a capacitor discharging the stored +12v by a certain amount (how much the +12v is modulated downwards and released determines the VRM's input voltage), or how much is discharged, but as you can see, this gets over the head of most people VERY FAST.

The overshoot voltage isn't a big problem because when the overshoot (transient spike) happens, there is barely any current going into the CPU. It happens during a "load release", not during a sustained load.
The issue is if the overshoot gets REALLY high, like something like past 1.65v (this point is unknown) despite lack of current. Because at a certain point, you can just blow through the oxide layer gates, destroying the CPU.

The main issue with a higher LLC is the *higher* sustained voltage. At higher current (amps), this can leapfrog past Intel's safe amps/voltage electrical calibration and slowly degrade processors.
For example, I calculated (if the Intel specification sheets are to be believed) what the maximum safe (on-die sense) measured voltage should be at different loads--this is measured directly from the CPU die, and thus unaffected by "power plane" voltage rise like you saw in Buildzoid's video---the difference between "on-die sense voltage" and "Average voltage from Probinator" probe.

For 9900k:

193 amps: 1.213v
150 amps: 1.280v
125 amps: 1.320v
100 amps: 1.360v
75 amps: 1.40v
50 amps: 1.440v

etc etc...
0 amps (no clock signal sent to processor): 1.520v.

All these values are based on a default Intel specified loadline being used (equal to "LLC2" on your Asus board, or LLC=Standard / Normal on Gigabyte Z390 boards), which is exactly 1.6 mOhms (milliohms) of loadline calibration.
You can calculate the math without even understanding a single thing I said just by using Ohm's law:

Convert volts to millivolts (to match same units as milliohms).
1.520-1520 mv

Vdroop is Amps * resistance (ohm's law), just like watts=amps * volts.
So if you start at 1.520v and plug in the amps, you get:
1520 - ( 193 * 1.6) = 1213 mv
1520 - (150 * 1.6) = 1280mv

And so on. This is self explanatory.

This is based on maximum vdroop of course.

Using "Auto" Vcore makes things complicated because while you're "supposed" to NOT use loadline calibration (keep it at intel spec) on auto voltage, the AC Loadline mOhms value is supposed to help counter the super high vdroop by boosting the CPU's input voltage to the VRM (before vdroop is processed). This has nothing to do with loadline calibration's mOhms by the way.

Ok, so what about loadline calibration on fixed voltage?

Well the settings above are based on "absolute maximum", which is if you used a BIOS voltage of 1.520v with LLC2 (AND NO HIGHER), which would run BOTH your idle voltages and your load voltages at the edge of what Intel specifies as safe. I dont think anyone wants to do that even if your processor wouldn't degrade.

Well, by using a stronger LLC, you reduce the vdroop (lowering the mOhms) for loadline calibration, which then throws off the entire volts/amps voltage curve for what is a safe LOAD voltage.
Take this example here.


1.340v BIOS set = LLC6 = 0.4 mOhms of VRM Loadline (loadline calibration). Instead of 1.520v, we now use a starting value of 1340mv. Let's do some math.

Is 100 amps safe here?
1340 - (100 * 0.4) = 1300mv

(Cross reference the Intel spec):
1520 - (100 * 1.6) = 1360mv --- 1300 < 1360 = YES YOU'RE SAFE.


Is 150 amps safe here?

1340 - ( 150 * 0.4 ) = 1280mv.

(Cross reference the intel spec):
1520 - (150 * 1.6) = 1280mv ---1280 = 1280= YES YOU'RE SAFE.

Is 193 amps safe here??
1340 - (193 * 0.4) = 1262mv

(Cross reference the Intel spec):
1520 - (193 * 1.6) = 1213mv --1262 greater than 1213 <--NOPE you're not -you're 50mv too high!

So you can see how changing the loadline calibration causes you to have to think about how much load voltage / amps you're putting into your CPU.

And the second issue as we already talked about, is the "sustained to transient -drop-" increasing at higher LLC, since it's the lowest voltage the processor will see that determines if you get "random" instability or not. And unlike "load release" spikes. the "Load activate" dips actually happen at heavy current, so you can get 'random instability if it goes too low.

The main benefit from a lower LLC is thus, a lower "average to minimum" dip. But at the cost of higher vdroop.
11-11-2019 04:23 PM
postem
Quote: Originally Posted by Falkentyne View Post
That's not what buildzoid said at all.

He said that going past LLC5 on the Asus board isn't recommended because the peak to peak increases more going from 5 to 6 than 4 to 5, and 7 and 8 aren't worth it at all.
He also specifically said that it was the 'minimum' which determines your stable voltage, and he said that the minimum doesn't decrease as much as the maximum increases.
He also said that the undershoot (transient dip) gets worse the higher the LLC you use, but not as severe as the transient overshoot increases.

He also compared to the eVGA Dark, where the transients on the Dark remained splendid up to the -75% vdroop level, which was the point he recommended using there and not going up to -90% vdroop.

Here might be a more to the point video.

https://www.youtube.com/watch?v=IEr8-4aBtCU

Here you can see the IX Gene (same power delivery as the Extreme; the Apex is better than both but he doesn't have an Apex) that the difference between "Minimum" to "average" increases more at LLC6 vs LLC5, and the average to max increases even more, so he recommends 5 as the best option. It was 40mv on LLC5 (minimum to average) and 60mv on LLC6 min to average). Notice that going to LLC7 and LLC8 simply made the average to minimum even larger? 80mv on LLC7 and 90mv on LLC8.

The Gigabyte Master was similar E.g. the Master had -40mv on LLC High (from average to minimum) but -60mv average to minimum on LLC Turbo, which means LLC High would have a lower drop from average to minimum (20mv improvement).
Thanks as I said my comprehension on subject is meager so I got it wrong.

Now about the benefits of having a more stable vcore across loads, and the fact the benefits lessen the higher you go, what are the issues we risk with having overshoot voltage, it's the risk of constantly going over what is safe voltage and getting more degradation?

Also another question, the overshoot is the same regardless of vcore set or it's a function of the actual value dialed in bios?
11-11-2019 04:02 PM
Falkentyne
Quote: Originally Posted by postem View Post
As I understand there are conflicting views. I saw the video yesterday, buildzoid basically advises to go full LLC, or higher you can (consider anyway he is always doing extreme overclock), while elmor seen to be telling going higher LLC values mean overshoot that could be an issue.

I think both proposals need to be analized in lieu of what kind of vcore you are dealing with. On the video buildzoid sometime mention he doesn't even consider 1.4v safe for long use, so do I, even considering Intel guideline of 1.5v max, anything over 1.35 will probably induce some degradation.

The reason buildzoid tells to use a higher LLC is that he thinks or at least I understood on video, it's better to use a higher LLC to achieve less undershoot under load than to use a higher vcore, but consider that if you are running conservative values like 1.2, 1.3 vcore, either way, the worse would be a higher LLC, due to overshoot. If with a median LLC you can manage to run stable at sane vcore, it would be the better approach.
That's not what buildzoid said at all.

He said that going past LLC5 on the Asus board isn't recommended because the peak to peak increases more going from 5 to 6 than 4 to 5, and 7 and 8 aren't worth it at all.
He also specifically said that it was the 'minimum' which determines your stable voltage, and he said that the minimum doesn't decrease as much as the maximum increases.
He also said that the undershoot (transient dip) gets worse the higher the LLC you use, but not as severe as the transient overshoot increases.

He also compared to the eVGA Dark, where the transients on the Dark remained splendid up to the -75% vdroop level, which was the point he recommended using there and not going up to -90% vdroop.

Here might be a more to the point video.


Here you can see the IX Gene (same power delivery as the Extreme; the Apex is better than both but he doesn't have an Apex) that the difference between "Minimum" to "average" increases more at LLC6 vs LLC5, and the average to max increases even more, so he recommends 5 as the best option. It was 40mv on LLC5 (minimum to average) and 60mv on LLC6 min to average). Notice that going to LLC7 and LLC8 simply made the average to minimum even larger? 80mv on LLC7 and 90mv on LLC8.

The Gigabyte Master was similar E.g. the Master had -40mv on LLC High (from average to minimum) but -60mv average to minimum on LLC Turbo, which means LLC High would have a lower drop from average to minimum (20mv improvement).
11-11-2019 03:19 PM
postem
Quote: Originally Posted by Falkentyne View Post
As I understand there are conflicting views. I saw the video yesterday, buildzoid basically advises to go full LLC, or higher you can (consider anyway he is always doing extreme overclock), while elmor seen to be telling going higher LLC values mean overshoot that could be an issue.

I think both proposals need to be analized in lieu of what kind of vcore you are dealing with. On the video buildzoid sometime mention he doesn't even consider 1.4v safe for long use, so do I, even considering Intel guideline of 1.5v max, anything over 1.35 will probably induce some degradation.

The reason buildzoid tells to use a higher LLC is that he thinks or at least I understood on video, it's better to use a higher LLC to achieve less undershoot under load than to use a higher vcore, but consider that if you are running conservative values like 1.2, 1.3 vcore, either way, the worse would be a higher LLC, due to overshoot. If with a median LLC you can manage to run stable at sane vcore, it would be the better approach.
11-11-2019 02:16 PM
Falkentyne
Quote: Originally Posted by KotOr View Post
so it's wise to lower LLC then? and do new stability testing ? thanks
11-11-2019 01:35 PM
KotOr so it's wise to lower LLC then? and do new stability testing ? thanks
11-11-2019 01:17 PM
Falkentyne
Quote: Originally Posted by KotOr View Post
hey guys i have question. i have 9900KF + M11 apex and i overclock it to 5Ghz all cores with LLC6 adaptiv 1.3V, offset -0.080V, Max 1.23V, AVX load 1.185V ,NonAVX 1.2V 88c test with cinebench r15 and aida64 . And my question is about LLC. On lots of forums i read that is recommended use of LLC 5 . but i found review that shows LLC 6 is still Vdroop and LLC 7 is overshoot. so why everybody recommend LLC 5 and should i change my settings? thank you
No LLC will overshoot "RMS" voltage.
People only thought the vcore 'overshoots' on RMS just because the Super I/O sensor totally sucks.
Every LLC besides LLC 1,2 and LLC3 will overshoot via TRANSIENTS. This is NOT what you are seeing on sensors. No sensor has enough resolution or update speed to see transient spikes or dips. That's why you need an oscilloscope.
Even multimeters that cost less than $1,000 can't pick up microsecond voltage swings.

https://elmorlabs.com/index.php/2019...ne-visualized/
11-11-2019 12:28 PM
KotOr hey guys i have question. i have 9900KF + M11 apex and i overclock it to 5Ghz all cores with LLC6 adaptiv 1.3V, offset -0.080V, Max 1.23V, AVX load 1.185V ,NonAVX 1.2V 88c test with cinebench r15 and aida64 . And my question is about LLC. On lots of forums i read that is recommended use of LLC 5 . but i found review that shows LLC 6 is still Vdroop and LLC 7 is overshoot. so why everybody recommend LLC 5 and should i change my settings? thank you
This thread has more than 10 replies. Click here to review the whole thread.

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off