[Sammelthread] Intel DDR5 RAM OC Thread

[Veii]

@Veii whats the ideal SVID mode in bios? There's auto, Intel fail safe, typical, best case, trained.

All presets ontop of fused curve
At best "non" , but its unknown what Auto does.

Trained.

👆
Best case to run "Trained".
V/F already will be slightly skewed if mount is bad or cooling changes drastically. Or CPU degrades strongly.
Good to have a trained curve. Higher chance of consistency vs ever changing ME presets.

Beitrag automatisch zusammengeführt: 14.03.2024

Edit:
Eine Frage:
Statt PC aus, ist es gut?
Änderung, F10--->Startet, durch boot Retry Button.
Oder besser stromlos, ohne Strom einschalten, kein LED, nichts, dann PSU an, und starten?

Coldboot vs Warmboot.
Nicht jede Einstellung führt ein frisches retraining durch.
Manche Boardpartner haben Bugs in dem Bios.
MemoryContextRestore würde RAM nicht neutrainieren da es ansonnsten für 60 Links/Traces (von CPU zu MEM) sehr sehr lange braucht (siehe AMD)

Zwar ist ein komplettes Training besser als ein potential instabiles, aber ja.
Bei großen Änderungen bitte Cold Boot.
PSU off, powerbutton (discharge) , dann PSU an und start.
// Bei Intel dauert das Training ein Bruchteil von dem Standart bei AMD. Fastboot variablen braucht man nun wirklich nicht. Gehört immer aus.
// 192GB (4* Micron Rev.D 48GB DS UDIMM) braucht 10min jeden Boot. Auch beim CMOS reset. Was sind hier auf Intel schon 20 vs 25-30 sek.

Retry button, ist ein soft / warm retraining.
Manchmal komplett manchmal nicht.
Das hängt von ASUS ab, aber wie sich die CPU benimmt hängt nicht von ASUS ab.
Cold boots, bei VDDQ Änderungen, bei training preset Änderungen, bei ODT/RTT änderungen.

Immer auf Nummer sicher gehen.

 

[Furyos]

10 layers~

Falls es jemand versuchen möchte, bei eBay für 250€:

ASRock Z790 PG-ITX/TB4 LGA 1700 Mini-ITX Intel Mainboard 4710483941108 | eBay

ASRock Z790 PG-ITX/TB4 LGA 1700 Mini-ITX. Außerdem sind 2x Thunderbolt 4 Ports vorhanden !

www.ebay.de

Topcashback gibt nochmal 5%, also für knapp 240€ inkl Versand zu haben.

 

[epilef]

vei what about those scewin variables
MRC ULT Safe Config
Force ColdReset
Training Profile
MCH Full Check
Safe Mode Support
Memory Test on Warm Boot

maybe any benefit with tradeoff of booting time?

 

[Veii]

MRC ULT Safe Config

Unknown, no MRC

Force ColdReset

Always good to be on

Training Profile

ASUS Profile is unknown, no found benefit so far

MCH Full Check

In theory good, but rarely functional
There was a reason why, but i forgot. Becomes irrelevant if you manually mess with training
Also may or may not seek improvement, given nobody i know on G4 uses that.

Safe Mode Support

Hardware button

Memory Test on Warm Boot

Whats the options ?
Never had any need to mess with any of those

---

Good morning.

 

[zebra_hun]

Guten Morgen!

Ich lerne immer etwas
Diese PSU, stromlos, usw., ganze Zeit habe ich es vielleich falsch gemacht.
Eine schlechtes:
Windows kaputt gemacht. Kein Problem, neu install geht.
Wie habe ich es gemacht?
Ich wollte extreme low Voltage 8200 Profil machen. Es bootet, aber windows ist nicht gestartet. Damals habe ich es immer Safe boot zuerueck gemacht. Windows wenn Probleme hatte, konnte sich selbe repaieren.
Jetzt habe ich es paar mal mit dem Schalter hinten ausgeschaltet, Durch falsch boot.
Ergebnis:
Windows ist total kaputt gegangen, alles weiss, blinkt es wie ein Stroboscope
Gespeicherte 8000 Profile zuerueck, aber konnte es nicht helfen.
Ich habe alle Datei gescpeichert (immer mache ich es), np, neu windows install.
Wenn es fertig war, kein debloat nichts, nur activated, hab ich Y ausprobiert. Kein 5 Sek, error...
OK, normal aus den Rechner, PSU aus, startet, discharge, warten halb Minute.
Nach dem total neu Restart (gespeicherte 8000 war drinn), traniert es sich, alles gut.

Wenn ich einfach wechsele die Profile, es wird nie in Leben gut einladen. Einziege Fehler war jetzt, ich habe PSU ausgemacht wenn es schlect war. Muss immer auswarten bis windows richtig eingeladet, dann aus.
Aber wenn ich gespeichertete Profile wechsel, muss auch PSU discharge.
Deswegen war es mir immer schlecht.
10 Minute habe ich es gelassan, voher konnte kein 5 Sek.
Naja, vielleich mein Fehler hilft Jemand...
Schoene Wochenende an Alle

 

[tibcsi0407]

Good Morning @Veii

I decided to build a new foundation from scratch.
Some of the things (RON, CPU_VDDQ) are from the old one, but I tought it would be smarter to build it up from 8400C38 again to hunt for the good VDDQ deltas.
The new profile looks like this right now. (did the capture during the run for ZEB )
MVDD: 1.44V
MDDQ: 1.35V
CPU_VDDQ: 1.24V
SA: 1.12V
VDD2: 1.40V
CTL0 and the most important timings are set based on your sheninigans.

BIOS is the new 9902 version. VDDQ and RON training off.

Will run longer tests later, just wanted to see how the low memVDDQ works.
Vcore is jumping between1.19-1.234V during the test based on core frequency, since TVB kicks in even during Y load. Maybe I will set to 8x58 +2TVB later.
Also, thinking on to buy a KS, please convince me that I don't need it!

 

[zebra_hun]

Tibcsi
Ich werde es kopieren. Aber so niedrig Spannungen, keine Chance. Glaube ich.
Schön gemacht.
Ich denke, 2x24Gb braucht weniger Strom, als mein 2x16Gb.
Nicht jetzt, später. Am Nacht.

 

[tibcsi0407]

Tibcsi
Ich werde es kopieren. Aber so niedrig Spannungen, keine Chance. Glaube ich.
Schön gemacht.
Ich denke, 2x24Gb braucht weniger Strom, als mein 2x16Gb.
Nicht jetzt, später. Am Nacht.

Geben Sie den meisten Spannungen eine Chance. Mvdd etwas höher, wobei die anderen Spannungen beibehalten werden

 

[Veii]

Also, thinking on to buy a KS, please convince me that I don't need it!

Just watched Roman's Video and i couldnt agree more.
The only reason for you to maybe buy one, is to have no ODT flaw.
Lets call it a flaw not a bug for now.

Outside of that, and who knows what they did to internal clock.
I dont think you! , can get a better deal out of it.
Zebra_hun might but then , if all of those KS are high leakage, ehh i dont know~

The potential approach for high SA while being interesting, will not be much of a benefit to our work
We run strong ODT and low voltages ~ which speaking of:

MVDD: 1.44V
MDDQ: 1.35V
CPU_VDDQ: 1.24V
SA: 1.12V
VDD2: 1.40V

110mV delta ??
Wasnt the plan supposed to be more vs less
Or is the initial plan to run as little as possible VDD2 & MVDD ?

What does mean "CTL0 and other Veii shenanigans"

---

At the end 8600 is a skill issue, its not a chip or bios issue
8800 is a bios + skill issue but maybe not "your chip" issue.
We talk about stable cache & ring stable too

The only potential reason i can see with KS is Gear4
But, given it looks like a fused curve redo and maybe slightly more stability on higher thermals
I would say, this CPU is only for people who were unlucky to get 400A dying 14th gen.

 

[zebra_hun]

Ich war zu "Naiv", 8400C38 gehts ueberhaupt nicht. Tested nur Y 2.5b, viele Variant ausprobiert. Low und High Voltages, TX und MC Voltages, immer Y error.
Einziege ist die mir wichtig, ich kann ohne Problem mein 8000 zuerueckladen und es bleibt stabil.
8400 ende, gehts nicht, 8000 nochmal 20 Min lang getested nach dem PSU, CMOS Reset.
Bild

KS werde ich nicht kaufen, ich mag es schon

 

[tibcsi0407]

110mV delta ??

That's why it's new foundation. But probably I can go lower with the CPU_VDDQ.
What is funny that higher MVDDQ wasn't stable.

For 8600 maybe I have to raise MVDDQ with 30mV, then I will have bigger delta if I can keep the TX. Need to experiment with that.

Zebra_hun might but then , if all of those KS are high leakage, ehh i dont know~

Yeah, I saw some pretty bad one.

What does mean "CTL0 and other Veii shenanigans"

TXSR and PD timings. :P The CTL0's are 178/96 from your shenanigans spreadsheet.

At the end 8600 is a skill issue, its not a chip or bios issue

Dunno, I played a lot with that and it was impossible with the old foundation. We will see how it works with this one.

I will ofc try the BIOS editor stuff you suggested.

The only potential reason i can see with KS is Gear4
But, given it looks like a fused curve redo and maybe slightly more stability on higher thermals
I would say, this CPU is only for people who were unlucky to get 400A dying 14th gen.

I am really afraid that I would get a worse IMC. For me the mem OC is more fun than the CPU OC. That's easier.

 

[Veii]

That's why it's new foundation. But probably I can go lower with the CPU_VDDQ.
What is funny that higher MVDDQ wasn't stable.

For 8600 maybe I have to raise MVDDQ with 30mV, then I will have bigger delta if I can keep the TX. Need to experiment with that.

Less is not bad.
Rule to keep MVDDQ sub MVDD is not soo important. They are split. *
// * another "it depends" but they mostly are split. Only micron i've seen to use SWA + SWB together. SPD&PMIC programmable;

Ive seen instances where less CVDDQ with same delta is less stable
vs more VDDQ on both @ same delta. For some reason.

Too many connections to CVDDQ 😣 confusing.

The CTL0's are 178/96 from your shenanigans spreadsheet.

Your next step is figuring out CTL0 ODT.
@zebra_hun can have fun with CTL1's.

Later we can scale up and down, i'll help.
But you guys need to get the values out so we can adapt to ODT and sample.
I can't use old data, of old rude people with own intentions, where not all belongs to me~~
Better to remake it and start clean~. We already make great progress.
Would need some Gear4 testing too, but lets see. Maybe if fortune shines and sky drops me some money, i'll tune 10k+ clock myself.

I will ofc try the BIOS editor stuff you suggested.

I wish we can dodge that part, but i dont like to depend on ppl.
Lets tune it ourself while HQ does other things. They have G4 fun to worry about and KS now.
ODT tuning is happening actively, soo we may assist a bit.

TXSR and PD timings

Next next (far future) is some check and continue work on CS_L/H
But thats for the future. I just inspected it a bit early, and there is some work to be done.
Tho, its not that important for now. Powerdown/Clockhalt stuff.
It surely can be better too, but if you look at everyone's else results "its fiine".

I am really afraid that I would get a worse IMC. For me the mem OC is more fun than the CPU OC. That's easier.

IMC hmmm
Nobody knows~~
y-cruncher 8600+ is too rare.
I can't depend on results. We have to work it out ourself, but till nobody shows low leakage V/F on KS.
I personally would stay away of it.
Your sample is better~

 

[tibcsi0407]

Your next step is figuring out CTL0 ODT.

From BIOS data?

Better to remake it and start clean~. We already make great progress.
Would need some Gear4 testing too, but lets see. Maybe if fortune shines and sky drops me some money, i'll tune 10k+ clock myself.

I can help you to research in ODT's and etc. Gear 4 can come too, when Gear 2 is finished.
Why don't you contact Asrock? They sent some samples from the Z790i PG, which is a small beast actually.

IMC hmmm
Nobody knows~~
y-cruncher 8600+ is too rare.
I can't depend on results. We have to work it out ourself, but till nobody shows low leakage V/F on KS.
I personally would stay away of it.
Your sample is better~

You are right.
I want to be in the rare group of 8600 Y cruncher guys. At least with a 90 minutes test.

 

[Veii]

From BIOS data?

Bruteforce.
SKU & CPU LkgFactor dependent.
Also SA dependent.

Why don't you contact Asrock? They sent some samples from the Z790i PG, which is a small beast actually.

I need to get global #1 for people & companies to respect me
Basically it's that.

G4 non LN2 has the highest chance, most silly and technical challenge where progress stops at people's knowledge.
// Because MemICs don't scale on LN2. There aren't that many shenanigans you can do. Its just data;
Rest is too expensive towards exotic cooling and requires >700$ CPUs.

 

[tibcsi0407]

I need to get global #1 for people & companies to respect me
Basically it's that.

I am not sure. They sent samples for less experienced users too.
Asrock is a nice company, I always liked them. Actually I would own an Asrock board instead of this ASUS if they would be able give us better DDR5. I think they will be more close with 15th gen.

I had an Asrocck X370 Taichi, loved that board!
Bought it about 7 years ago and it's still working.

 

[Veii]

I had an Asrocck X370 Taichi, loved that board!
Bought it about 7 years ago and it's still working.

My B550 PG-ITX still is a beast (daily)
Does 4600MT/s dual on A0 B-Die (Vipers), and 5000MT/s dual on A2 Micron Rev.E
4400MT/s on Hynix CJR OEM. Dailying it at 4200:2100 G1.
// all are stable examples.

B550 ProArt sits here and X370 (X470) Taichi sits here.
X670E Taichi (white) was a great experience on AM5 , same as the ROG GENE.

Its just that 1DPC > 2DPC. Always. Unfortunately absolutely always.

 

[tibcsi0407]

Bruteforce.
SKU & CPU LkgFactor dependent.
Also SA dependent.

That won't be easy. Do you have any idea where to start?
Need to set ODT for that too?

Beitrag automatisch zusammengeführt: 15.03.2024

My B550 PG-ITX still is a beast (daily)
Does 4600MT/s dual on B-Die Vipers, and 5000MT/s dual on Micron Rev.E
4400MT/s on Hynix CJR OEM. Dailying it at 4200:2100 G1.
// all are stable examples.

That's very nice on Ryzen!

 

[Veii]

That won't be easy. Do you have any idea where to start?
Need to set ODT for that too?

Hard set voltages
And start with value ~98 on DQ ODT UP.
Probably till value 132. Ones within +/- 1 step can boot and be stable (randomness of 3 values)

Unfortunately , requirements are 6h y-cruncher
Soo thats like 12 CMOS resets if you do 3 value jumps (34 values that may or may not work. only one of them can be stable @ target SA)

EDIT:
You probably target near 120-130 value for 8400MT/s.
But given very low SA and new ODT tuning, it could be in the 110-120 range. Who knows
98 to 132 , GL

CTL1's remain Groups+RTT
Soo that can be zerba's work, given the RTT struggle.
Later i can scale things , once foundation of 4 values (whole CTL1 or whole CTL0) is found.

 

[tibcsi0407]

Hard set voltages
And start with value ~98 on DQ ODT UP.
Probably till value 132. Ones within +/- 1 step can boot and be stable (randomness of 3 values)

Unfortunately , requirements are 6h y-cruncher
Soo thats like 12 CMOS resets if you do 3 value jumps (34 values that may or may not work. only one of them can be stable @ target SA)

Uhh man, that's a lot of time.
Is it clock dependent too?
If not maybe I could get a NASA stable 8000-8200 and use that for experimenting.
The current profile still needs long stability tests.
I like the low Voltages. Maybe for 8600 I can use a similar profile with loosenedd timings.

 

[Veii]

Uhh man, that's a lot of time.

It fails either fast or doesnt boot.
CMOS reset, tho poweroff CMOS reset.

2-3 options may run 1h , some may run 15min, some may not boot at all.
Changes are drastic, and out of 34 examples 2-3 values only will run "a tiny bit".

If not maybe I could get a NASA stable 8000-8200 and use that for experimenting.

Having a 6h foundation stable, is a requirement. Minimum 4h. Any slopes testing needs long long time to fail.

Is it clock dependent too?

CTL0 yes (you already know)
CTL1 , no. tiny bit but no. Its rather strain factor.

Like delta, once found it stays that one, till one changes ODT foundation.
we have 12 slopes that all need to sit perfect. At least CTL1+2 are so so ok. Its a lot of time.
But CTL0 needs to be fine, to have no boot-randomness.

 

[tibcsi0407]

ut CTL0 needs to be fine, to have no boot-randomness.

Yeah, but that's based on SA. So for 8600 1.12V SA won't be enough in my opinion.
So need a profile with 1.18-1.20V SA and experiment on that one.

 

[Veii]

Yeah, but that's based on SA. So for 8600 1.12V SA won't be enough in my opinion.
So need a profile with 1.18-1.20V SA and experiment on that one.

You don't change ring much. And 2150 QCLK is nothing.
Need more SA on Gear1.
On G4, SA barely to doesn't change at all.

CVDD2 is what scales between QCLK.
CVDDQ & SA barely to non.

 

[tibcsi0407]

You don't change ring much. And 2150 QCLK is nothing.
Need more SA on Gear1.
On G4, SA barely to doesn't change at all.

CVDD2 is what scales between QCLK.
CVDDQ & SA barely to non.

So you do believe that SA 1.12V could be enough for 8600 too?

 

[Veii]

So you do believe that SA 1.12V could be enough for 8600 too?

Slightly afraid that CVDD2 will not scale much further.
But i don't think SA will need a change, especially as slopes shape VREF.
And again that QCLK is nothing, compared to usual Gear1 results

At the end when too strong ODT becomes a problem is high heat and noise. This approach is sensitive to bad quality PCBs.
Board PCB is great. Maaybe low SA has more headroom on G4, but you should try~

EDIT:
As long as you can boot it, its enough.
No real package-throttle on SA side , on intel.
Slightly yes, but you will notice it with y-cruncher.
As long as it makes it to windows, its enough voltage. +/- 5 mV

EDIT2:
On AMD we usually go stronger ODT with higher UCLK (QCLK), not weaker.
You are just there where it belongs. Don't think it needs much of a change.
Its more sample dependent where sweetspot is. And well fortunately? will change by BIOStuning.

 

[tibcsi0407]

Slightly afraid that CVDD2 will not scale much further.
But i don't think SA will need a change, especially as slopes shape VREF.
And again that QCLK is nothing, compared to usual Gear1 results

At the end when too strong ODT becomes a problem is high heat and noise. This approach is sensitive to bad quality PCBs.
Board PCB is great. Maaybe low SA has more headroom on G4, but you should try~

EDIT:
As long as you can boot it, its enough.
No real package-throttle on SA side , on intel.
Slightly yes, but you will notice it with y-cruncher.
As long as it makes it to windows, its enough voltage. +/- 5 mV

So I need to focus on VDD2 instead. And MVDD, rest could be the same. Since I got some good results on 8533 with 1.12V SA too, it could be possible.
Heat is not an issue here.
If I would be @zebra_hun I would use DD. He has very cool water with a basement cooling. That could give a lot of profit, especially on his CPU.

 

[Veii]

So I need to focus on VDD2 instead. And MVDD, rest could be the same. Since I got some good results on 8533 with 1.12V SA too, it could be possible.
Heat is not an issue here.

Technically we can skew SA a bit by PLL. But i don't want you to use such helping methods for just 8600.
That's 8800+ fun.
5-10mV difference on SA wont make or break ODT. But you'll notice when its an ODT issue.

 

[tibcsi0407]

Technically we can skew SA a bit by PLL. But i don't want you to use such helping methods for just 8600.

Yeah, PLL's are still on auto. I used ring PLL in the past, but it didn't help in every case.

 

[zebra_hun]

Sicher was sicher, nach so viele aenderungen weiter getestet.

TM5 jetzt 50 Cycle

Ich bin dabei in CTL 1 u. 2 Job auch. Ich weiss es noch nicht, soll ich es mit 8200 versuchen? 8000 denk ich fertig. 8200 habe ich 10 Cycle TM5 geschafft.
Aber es gut zum wiessen, dass einen Profil einladen, F10, restart nicht gut. Muss total neu traniert werden.
Es war komisch mir damals. Getestet, gespeicherte Profile einlade, test, passt es nicht. Spannungen gespielt, und doch war die original, gespeicherte Wetren die besten.

 

[epilef]

Unknown, no MRC

Always good to be on

ASUS Profile is unknown, no benefit found so far

Good in theory, but rarely functional
There was a reason why, but I forgot. Becomes irrelevant if you manually mess with training
Also may or may not seek improvement, given nobody i know on G4 uses that.

Hardware button

What's the options?
Never had any need to mess with any of those

---

Good Morning.

Setup Question = Memory Test on Warm Boot
Help String = Enable Or Disable Base Memory Test Run on Warm Boot
Token =642 // Do NOT change this line
Offset =1A3
Width =01
BIOS Default =[01]Enabled
Options =[00]Disabled // Move "*" to the desired Option
*[01]Enabled

Here is the full settings of encore


on dark i activated the SAFE mrc settings, but couldnt notice any diff

 

[zebra_hun]

Oder, du kannst beginnen CTL1 CmdVref UP für deine ODT+RTTs herauszufinden.
Zwischen 125-145 sollte es sein.
Wert 1,2,3 (innerhalb 3 Werte) wird booten. Sind einiges an restarts 🤭
Stabilität muss über 3h sein (slopes eigentlich ja 6h), aber wird auch in 10min failen bzw garnicht booten.

Achte auf die "raw compute" Werte, wenn du slopes änderst.

Ich moechte garnichst versprechen. Ich werde mit denen anfangen. Paar Frage hab ich ueber das.
Ist es besser, wenn ich es mit einen stabil Profil mache, oder kann (darf) ich mit einen instabieles Profil anfangen?
Damals die ODT/RTT Werten so habe ich audgefunden, dass ich die Zeit gemessen habe. Error wann bekomme ich.
A Version 3 Min, B version 10 Min, C version kein boot, usw.
Weil jetzt 8200 ist noch instabil, die Spannungen 100%, dass sie nicht passen. Oder soll ich mit meinen 8000 ausfinden, aber ich kann nur schlecht machen, weil es Gott sein Dank, gehts.
Bis jetzt, ich nutze kein "Hilfe", PLLs, DFE keine Ordnung was sind die so wie so. Nur RTT/ODT's, weil ohne das, geht es nicht.
Hilfe von dir habe ich damals gefunden. Korregieren sollte ich, aber konnte ich anfangen.
So schaut jetzt aus:

(Am Abend/Nacht moechte ich anfangen, hoffentlich habe ich Zeit, und schaffe ich etwas. Das ist mein 8000 Settings)