[Sammelthread] Samsung 8Gbit DDR4 C-Die K4A8G085WC - OC-Ergebnisse im Startbeitrag! [updated]

[Nightguest]

It seems, you have not given in to the C-Die madness and managed to preserve your sanity. At least your result for 3733c18 looks pretty decent

You say that now, however, I tried to put GDM off and bring TRFC down to 584.
Booted into windows and ran 5 cycles of MemTest64 1usmus (I know short) and had no problems.
But then AIDA failed in 10 minutes and windows refused to boot up after that.

I reverted the changes and windows BSOD'd on me before login, then when I went into the BIOS the BIOS just froze.

Had to clear CMOS and revert to stock XMP for now. I've gotta figure out wtf happened as for the last 2 days I've been using my posted 3733 without any problems, but today it decided to fail after trying minor changes and then reverting it!

 

[Kammi]

habe nur noch Aida zum testen genutzt. Das scheint eine bessere Fehlererkennung zu haben, auch wenn es deprimierend ist,
ein stabiles System ist wichtiger als ein paar Benchmark Ergebnisse. Anno 1800 ist ebenfalls ein guter Stabilitätstest.

 

[Myrkin]

Hello guys, sorry that I'm writing in English, but my german is WAAAAAAY worse.

I'm looking for help to tweak my memory.
My current setup is:

Gigabyte B550 Aorus Elite
Ryzen 9 3900X
2xSamsung M378A1K43CB2-CTD (C-die)

So I've managed to get 3200 MHz with 18-18-18-34-74-1T and I thinking, that I've done a pretty bad job.

This is my first AMD build, and I've read that ryzen 3900x sweet spots for memory are 3200 and 3600 with as low timings as possible.

I've tried to use ryzen dram calculator, chose a safe setting for Samsung OEM, and it didnt work at all.
Maybe some of you will share your presets for 3200/3600 MHz. And I will also hugely appreciate any advice on how to get the maximum from this memory.

Thanks in advance mates.

 

[Kammi]

dram calculator arbeit nicht mit Samsung c-die. Ich empfehle die soc-voltage auf 1.1V zu erhöhen. Die c-die laufen scheinbar mit niedrigen Spannungen am besten.
Poste doch mal die Timings, bin sicher das Du höhere Frequenzen gehen kannst.

Beitrag automatisch zusammengeführt: 03.09.2020

auf der ersten Seite findest Du ein Beispiel für erfolgreichen übertakten der sticks

 

[Nightguest]

@Myrkin
As Kammi wrote, there is no entry for c-die in the dram calculator, just throw that away.
If you can share the settings that you've used to get your 3200CL18 working, it may help others to identify what you can improve. Also, do go back a few posts and pages, there are other ENG people here that have posted their OCs along with some very useful tips for the timings. If in doubt with German, google translate is your friend (I personally just open the page in Chrome and use the full page translate option.)

If I understood correctly, your RAM is DDR4-2666 at CL19 and is about as generic as can get. If it is possible for you to get new RAM, you may want to consider that (I know it's not always the easiest to swap out components so that's just 'if you can').

Part of the issue that you're going to run into is that c-die is not a friend of tight timings. Normally, we can compensate for that by increasing the frequency, HOWEVER c-die is extremely temperature-sensitive and your modules do not have a heat spreader. My concern for you is that by aiming for something like 3200CL16 (which should be possible), you may come into instability due to heat. For that, I'd recommend that you get a fan and have it directly over your RAM blowing air on the RAM sticks, that may help.

With that in mind, I actually do not think that your 3200CL18 is in fact that bad. Though I've got no experience with your particular RAM so I'm just guessing.
What you could try is to take the stock settings of say, the Corsair Vengence LPX 3200CL16, and see if that will work with you. You can see the default settings here:

https://image.prntscr.com/image/hL7ArpX2SSeQY3w4UroNLg.png

(That's mine, I just set it to XMP and everything else auto).

There is no promise that it would work, it's a total wild swing, but no harm in trying.
RAM Voltage: 1.35v
SOC: Auto

Alternative Idea
If you cannot get 3200CL16 at least, do consider an often-overlooked alternative, slow RAM (3200 or below) with maxed out FCLK (1800+).

The idea is, keep your RAM frequency at stock frequency and just focus on tighten the timings , then push your FCLK to 1900 (or whatever your maximum is). By doing so, you will get a much higher performance jump than by reaching 3200 worse than CL16.

Normally, you hear people say keep FCLK and MCLK coupled for the best performance due to a latency penalty when decoupled. However, below 3200, if you can hit FCLK 1900 you make up for the latency and then some. With that in mind, if you stay at 2666 and just tighten the timings a bit, then go all out on FCLK overclocking, you will get better performance.

The only catch is, not all CPUs will hit 1900 FCLK, you'll need to test what your max is (for example, my 3700x maxes at 1866).

 

[emissary42]

This is my first AMD build, and I've read that ryzen 3900x sweet spots for memory are 3200 and 3600 with as low timings as possible.

C-Die are not exactly the ideal ICs for tight timings, for that good old B-Die are still king 👑

However I would actually disagree and say that the sweet spot for Zen2 systems usually is max 1:1 frequency. So in most cases 3733/1866 or 3800/1900, unless you have to make major sacrifices among important timings for the last frequency-step(s). Async configs can be a valid alternative if either your board or the memory just can't get there. However the C-Die typical 3600/1800 18-20-20 at around 1.3V shouldn't be that hard to achieve with a decent motherboard even for memory overclocking beginners.

 

[Nightguest]

From 05/09
Another Saturday and a bit more improvement to my RAM OC.
Still with 3733 (no luck with FCLK 1900, pretty much going to give up on that).

Achievement of the day was getting the latency to 67.9ns, previously I was stuck at 68.xns
(AIDA and Timings here)

Compared to the original some timings were actually loosened, but in return, others were tightened. In the end, it was a game of cat and mouse between finding something stable. However, I still wasn't able to get stable with GDM off.

Some new settings:
ProctODT: 36.9
RAM Voltage: 1.35v
VDDP: 1.0v
VDDG: 0.95v
SOC: 1.1 (LLC3)
CAD: 24-24-24-24

Stress tests so far:
AIDA: 1h
TM5 1usmus: 50 Cycles

Pretty sure that I can get some of the tertiary timings to be tighter, but I need to use my PC for now so I'll get back to this next week!
----
Edit: New (12/09)

I got GDM disabled, but that needed quite some changes to my timings:

I had to increase tRP and tRAS as I could not find a stable setting with them on the original (last week's) setting. However, increasing them did give me a bit more room to drop tRCDWR from 20 to 18, and a lot of the weird tertiary timings are now 1 (previous I couldn't do that).

Some future goals:
tRDWR - 16 is quite high, so I think I've got some space there.
tRFC - I feel that there is still a little bit of room here, but it's getting tight.
I think CAD can be brought a bit lower. What I have should be safe, but I don't really want it on 60-30.

I do have a personal target of trying to break into the 66.x n/s for latency. BUT I do think that that target may be a bit hard as at this point I think I'm nearing the limits of what I can do with my RAM, but I'm happy to keep poking at it every Saturday!

Tests for stability:
30 cycles of TestMem5 1usmus
1h with OCCT new memory test
1h with AIDA

 

[gelbwurstOtter]

Hallo zusammen,

ich bin etwas irritiert durch die Aussage in der "ultimativen 8Gb B-Die-Liste":

So sind Meldungen und Berichte von B-Die auf Corsair oder G.SKILL Modellen mit Spezifikationen von DDR4-3000 15-17-17 oder DDR4-3200 16-18-18 nicht erst ab 2019 äußerst unwahrscheinlich,

sondern in den meisten Fällen auch schlicht falsch. Man kann dies bei Corsair Modellen an Hand der Version des Kits verifizieren, welche für B-Die Ver4.31 betragen sollte. Bei solchen Kits handelt es sich aber statt dessen meist um Ver4.32,

Ich selbst besitze ein 2x8GB Speicherkit von Corsair (3000MHz 15-17-17-17-35) in der Version 4.31, sind das jetzt nun C-Dies oder nicht?
Ich hätte das o.g. jetzt so verstanden, dass die v.4.31 eigentlich v.4.32 sein müsste und es deshalb "nur" C-Dies sein sollten.

Kann mich bitte jemand aufklären?

Viele Grüße
g.O.

 

[emissary42]

Ver4.31 = B-Die
Ver4.32 = C-Die

Wenn dein Kit noch Ver4.31 ist, dann handelt es sich bestimmt um eines aus alter Produktion. Das Produktionsdatum kann man aus der Seriennummer ableiten (JJKW).

 

[Nightguest]

Hey folks, I've got a question that has been nagging me for a while.

It's been written in many places (including here) that c-die should not go over 1.35v. I pretty much went with that and have even advised others with c-die not to go over that. However, I realized that I've not really read any reasons as to why 1.35v is the limit other than some people reporting performance regression over 1.35 (or even over 1.3v).

So my question is, does anyone here know if over 1.35v can damage the RAM, or is the advice all based around potential regression?

The reason I ask is that I've been playing with my lovely (and cursed thanks to c-die) Corsair Vengence LPX 3200CL16 and I found that I can get it to 3200CL14 with 1.4v perfectly stable, I even used it for two days with no problem at all. So I'm curious about what I could do (if anything - I don't know if it would do anything) with my 3733OC by increasing the voltage. However, I'm not in a position to buy new RAM anytime soon so I'm rather on edge about using more.

 

[emissary42]

There might be some kind of misunderstandung or a translation issue going on here, at least about the info in the starting post. In general the voltage tolerance of any certain C-die kit can vary. Some might at higher frequencies already show stability problems at or below 1.35V, while others do scale well beyond 1.35 and even up to 1.4V. We even had examples for that even here in the thread. From my own experience I would consider 1.4V to be on the higher side in terms of voltage tolerance for C-Die, which at least in theory makes it one of the better kits. So if your kit does still scale with 1.4V and that allows you to reach CL14 @ 3200 there is nothing wrong with that.

However in general 3200 is a pretty low frequency for any current platform and is usually not worth it even at CL14, if your setup can handle DDR4-3600+ without too much trouble. At least C-Die should not have any issues with that.

It is recommended to stay well below 1.5V, since I have read of several cases where kits died at that voltage within a couple of days in the Corsair user forums.

 

[gelbwurstOtter]

Ver4.31 = B-Die
Ver4.32 = C-Die

Wenn dein Kit noch Ver4.31 ist, dann handelt es sich bestimmt um eines aus alter Produktion. Das Produktionsdatum kann man aus der Seriennummer ableiten (JJKW).

Hi emi,

das klingt ja vorzüglich... dann habe ich tatsächlich mal Glück gehabt. Ich habe die v4.31, ich würde zwar gerne upgraden kann aber leider nirgends ein zweites Kit mit derselben Version finden.

Beste Grüße
gO

 

[Nightguest]

Another weekend and another Saturday spent on my RAM with some nice results!

AIDA tends to vary between 67.5-67.3 ns for latency, however, I have got 67.3 more than once (so it wasn't a one-off-best-case situation, hence I'm using it).
The above settings are still with the same voltages/VDDG-P/CAD/RTT settings as before.

I'm looking to see if I can get tRP to 21, however, get some odd errors after an hour of 1usmus on TM5. I actually get error 0 - which tends to mean not enough voltage which is a pain.

Moving to 1.36v did help pass 1h of tm5 1usmus and 1h of OCCT so I left it at that for half a day to see how my system works. However, re-testing in tm5 in the evening gave me an error at around 50m in, so I ditched it for now as stress testing RAM OCs just takes so long. However, I do think that it should be possible.
tRDWR and tWRRD is driving me mad, trying to get tRDWR lower will result in a failed boot. I can boot (and be relatively stable) if I increase tWRRD and decrease tRDWR, however that results in much worse latency (67.7+) which I do not understand.

If anyone knows of the rule for tRDWR and tWRRD it would be amazing if you could explain it.
Also, another question that I have is, how safe is running CAD strength at 120-20-20-20 (or 30 instead of the 20's).
I can't find anything online about the safety or the impact on the performance of the CAD strength, other than it can help stability.

Edit: 27/09
Not gonna double post so editing this one.

At this point, the changes are so focused/small that it's not worth it other than personal obsession with finding the best.

Changes this week have been:
tRP from 22 to 21
tRDWR from 16 to 10
tRDRD (SD and DD) from 1 to 5
tWRWR (SD and DD) from 1 to 7

I pretty much learned for those SD/DD timings, lower is not always better. Going up to 5 and 7 here let me bring down tRDWR from 16 to 10 which helped stabilize the latency.
By stabilise, based on last week's settings I could run AIDA and the latency would vary from 67.3 up to 67.7 (I ran multiple tests on that and it was very chaotic).
With the current settings, I'm getting 67.3 to 67.4 very consistently, with a rare 67.5.

Though DRAM Calculate is useless for c-die, the Membench is a nice way to see improvements a bit more visually. Saved results 2 is my XMP, Saved results 1 was my results from last week, and the 'current' is today's settings. So the improvement is from 129.48 to 127.49 - very small, but I like improvements.

Oh, and it wouldn't be c-die without playing with voltage. I had to go from 1.35v to 1.34v to pass all stress tests. All other settings (RTT/CAD) are the same as before.

Tests have been:
20 Cycles of TM5 1usmus
2h on OCCT

It is a bit short, but there were no errors and I've been using the PC with no problems so I'm calling it stable!

 

[KillAllTeammates]

Thank you very much for the tutorials. I'm still adding voltage foolishly without this!

Below is my current configuration.

This is the result without optimized tertiary timings and the sweet spot of voltage hasn't been tested yet.
The reading speed seems too low comparing to the writing. The copying and latency are not good either. What could be the cause of this?
I think there may be two reasons. The first is the performance of 9400F and its IMC is not good enough for DDR4 3866. The other is the tCL and tRCD are too long and make the reduction of other secondary timings meaningless. The tRAS is only 34 now and there may even be room for further reduction. I think this is abnormal. But it's nearly impossible to continue reducting the tCL and tRCD.

 

[emissary42]

@KillAllTeammates Welcome to Hardwareluxx. To figure out any potential performance hindrances on Intel platforms I recommend ASRock Timing Configurator, since it shows most of the relevant timings. However if you look at my old Kaby Lake screenshots in the first post and considering the lower CPU frequency, your result is pretty much in line with my result for 3866 18-21-21-38 1T (screenshot). Overall lateny is on the high side, but can also be impacted by background processes and your Windows installation. The gains from minimizing tRAS are very minimal though, in comparison to fully tweaked tertiary timings.

 

[busiderbaer]

Ich hab heute mich nochmal mit OC von Ram und CPU beschäftigt, jetzt, wo CTR von 1usmus veröffentlicht ist.
Mein 3600CL18 ist in CB20 sogar ein paar punkte schlechter als die 3200CL16. Dafür konnte ich die CPU auf 4350 MHz Allcore @ 1,275V bringen, das find ich schon ganz in Ordnung.

 

[Nightguest]

I'm back with more c-die fun, this time I spent most of my efforts on optimising the power settings as I felt (and verified) that that was what was holding me back.
From the power setting changes I managed to bring tRAS down from 42 to 39, not much but every last bit helps!

The biggest changes this time round come from: CAD_BUS, ProcODT, and DRAM voltage.

CAD_BUS: 24-20-24-24 (previously it was 60-30-30-30).
ProctODT: 36.9 (previously it was 40)
RAM Voltage: 1.33v (previously it was 1.34v)
VDDP: 0.9 (from 1v to 900mv)
VDDG: 0.950 (the same)
SOC: 1.1 (the same)

The timing changes:
tRAS: 42 to 39

---

About ProcODT and CAD_BUS (and why I changed it/how I found the stable levels when it was so much different to my original one)

1. A lower ProcODT lets you have a lower SOC voltage (inversely more SOC = more ProcOTD)
2. A higher CAD_BUS helps you be more stable on lower DRAM Voltage.
This is hugely important for us c-die users as most c-die sticks have issues at 1.35v or higher, and it seems that the sweet spot would be between 1.3 to 1.35. It also explains why alxns needed 120-20... to get 1.33v to work on his 3733 OC).

My personal guide is:
ProcODT: 36.9 for SOC 1.1v or below.
ProcODT: 40+ for SOC 1.1v or above.
And as AMD say, don't go over 80 (though that was for Ryzen 1000, I think they said over 40 for Ryzen 3000, though that would probably be due to the SOC relation).

Ultimately it is a balancing act as lower ProcODT lets you work with a lower SOC voltage - lowering ProcODT makes the current more sensetive to electrical noise and can cause stability issues.

I personally go with the following process when I need to find stability:

1. Increasing VDDG IOD (to the safe limit) before increasing ProcODT
2. Then a higher CAD BUS before increasing ProcODT
3. ProcODT increases should be the last thing you increase, the reason for this is that ultimately increasing the ProcODT puts more stress on the IMC, which is pretty much one of the most sensetive parts of your system. I'm pretty sure that it would 'always' be safe, but I'm not going to test that theory.

Unverified:
I heard that a lower ProcODT helps with FCLK - that's something I might try as I've never been able to boot into 1900 FCLK (even when I'm on 3200CL16).
However, by the logic of how ProcODT works, lowering it would increase the current (Current = Voltage/Impedance) and a higher current is much more sensetive to the electrical noise (which is why we have ProcODT). So there will have to be a balancing act.

Hope this helps someone (and I hope it's all correct information. I wrote it up with confidence, however I welcome anyone that can correct me as I'm always willing to learn!)

 

[ensar492]

Hi everyone, before I start, I'll be using English, as my German is quite bad, please excuse my ignorance. Now with that out of the way, I recently got two 8 GB sticks (SKU: AX4U320038G16A-ST50) and here is what Thaipoon Burner shows me when I read them:



Apparently they're both Samsung C-die. And considering they refuse to do 3600 CL16 even with CL16-22-22-40, I think its fair to assume Thaipoon Burner is reading them correctly. If I am mistaken, please tell. DRAM Calculator is no use as some of you have already said, so I had to let most of the secondary timings on auto since I'm basically a beginner at memory overclocking, and I don't want tighter timings that can cause instability, which is what I had in my first attempts.

So here are my observations:

* DRAM voltage seems to be a little higher than what I set in the BIOS. Setting it 1.35v results in 1.37v, setting it 1.37v results in 1.39v etc... Concerning.

* I was able to boot to Windows at 3800MHz (IF clock 1900) with 18-22-22-38 If I remember correctly, however it was unstable and Ryzen Master crashed before five minutes passed.

* I tried 3600MHz CL16-22-22-38 and failed miserably. Tried 3600MHz-CL18-20-20-38, and passed 6+ hours of memtest86, and used my system on this speed for a month. At this point I realized the motherboard has been using really high SoC voltages so I manually tuned them down to 1.1 SoC voltage, 0.97 VDDP, 0.95 VDDG. DRAM voltage is set to 1.35v however it shows 1.37v, worth investigating if different DRAM, VDDP or VDDG voltages helps stability, however I'm already very stable at 3600MHz.

* I then tried 3733MHz CL18-21-21-38 however did not stress test it properly because I was honestly lazy. It ran stable for almost a week, however just today I had instability and data corruption. Might have to do with heat however since the sticks have no heat sensor I can't tell.

* I didn't give up after this failure, tried 18-21-21-39 this time, started by setting my fan curves higher for better cooling, I also manually set my ProcODT to 43 ohms. (Motherboard normally defaults to 60 Ohms at 3733MHz) It passed memtest86 without any issues however I'm not quite sure if I'll have instability again. I have no idea if the extra air or ProcODT helped with anything, only time will tell. In that time please feel free to help me out with tips to help stabilize my system.

Here is the final state:

Even though my timings are very loose compared to the default XMP profile I've gained a lot of performance in CPU bound games. I know buying better memory in the first place would be the smarter move here however I was tight on budget and I'd like to have an agressive, but safe overclock. Hope these voltages are okay for daily use.

 

[emissary42]

Hi everyone, before I start, I'll be using English, as my German is quite bad, please excuse my ignorance.

Don't worry about it

* DRAM voltage seems to be a little higher than what I set in the BIOS. Setting it 1.35v results in 1.37v, setting it 1.37v results in 1.39v etc... Concerning.

That is very normal for many motherboards, even higher end ones. Apparently it is somewhat intended to help with general memory compatibility.

Tried 3600MHz-CL18-20-20-38, and passed 6+ hours of memtest86, and used my system on this speed for a month.

It is a very typical result for C-Die and usually also requires less work than 3733/3800. So if you don't want to invest the time required to fully stabilize a higher frequency setting just stay at 3600c18 instead.

Hope these voltages are okay for daily use.

They look pretty safe to me.

 

[fromanothercountry]

About G.SKILL Ripjaws V F4-3600C19D-16GVRB.
I bought 2 of them yesterday and sticker on it says 2020 July.
I got hynix CJR

reason I'm posting this is
1. want to give different alternative situation from

Confirmed Models From The B-Die List & Other Sources:

spoiler
Corsair Dominator RGB CMT16GX4M2C3200C16 ( Review @ The PC Enthusiast )
Corsair Dominator RGB CMT32GX4M4C3200C16 ( confirmed Corsair Forum )
Corsair Dominator RGB White CMT64GX4M4C3200C16W ( Review @ PC Perspective )
Corsair Vengeance LPX CMK16GX4M2B3000C15 ( quota )
Corsair Vengeance LPX CMK16GX4M2B3200C16 (confirmed on several occasions, Including here or in the Corsair Forum )
Corsair Vengeance LPX CMK32GX4M2B3200C16 ( confirmed by user )
Corsair Vengeance RGB Pro CMW32GX4M2Z2933C16 (QVL of the ASUS ROG Crosshair VIII Impact)
Corsair Vengeance RGB Pro CMW16GX4M2C3000C15W ( Photo @ B-Die Liste )
Corsair Vengeance RGB Pro CMW16GX4M2C3200C16 ( confirmed in the Corsair forum )
Corsair Vengeance RGB Pro CMW32GX4M2C3200C16 ( Photo @ GEPC )
G.SKILL Aegis
F4-3 Aegis F4-3200C16 Photo , CW17 2020)
G.SKILL Aegis F4-3000C16D-32GISB ( Photo , CW08 2020)
G.SKILL Flare X F4-3200C16D-16GFX ( Review @ The Overclocking Page )
G.SKILL Ripjaws V F4-3200C16D- 32GVK ( Photo )
G.SKILL Ripjaws V F4-3600C19D-16GVRB ( Photo )
G.SKILL Trident Z F4-3200C16D-16GTZB ( Photo )
G.SKILL Value F4-2666C19S-8GNT ( Photo , [ Q ])
Transcend TS2666HLB-16G ( Review @ coolaler by johnuahuang)

and 2. asking why all programs showing different information? I'm downgrading it to 3200mhz cl16 because I'm using ryzen 1600 + asrock from350m pro4 mobo at the moment.
even dram calculator 1.7.3 says cl is 18 so confusing. ofc in bios, it shows as 16 as I set it.

 

[emissary42]

About G.SKILL Ripjaws V F4-3600C19D-16GVRB. I bought 2 of them yesterday and sticker on it says 2020 July. I got hynix CJR.

That is not unexpected, both IC types are pretty decent fit for them.

reason I'm posting this is
1. want to give different alternative situation from
and 2. asking why all programs showing different information?

1) Regarding your quote, which you also could have spoilered: The models listed are not exclusively built from Samsung 8C and were just examples. Pretty much any current IC type can clear typical C-Die speficiations, so you can get a wide variety of ICs from Micron 8Gbit B/D/E/J to Hynix 8Gbit C/D/J among others.
2) Timing Checker and CPU-Z show the same timings. Not sure about ZenTimings, I think of it more as a Zen2 app with some support for older gens.

 

[Nightguest]

A big milestone for me - 3733 CL17 has been unlocked and is stable!

Passed 4 cycles of Memtest86 (I booted in via USB as I wanted to catch the 'worst' instability before going into windows). I'm pretty sure that this step saved me a bit as I needed to quit/change timings 3 times before I cleared Memtest.
After that TM5 1usmus - 30 Cycles cleared.
Very quick HCL to 500% (I know, I should let this go longer, but I got impatient. I'm going to do a longer run overnight just to be sure)
Also been using my PC for the last 4 hours and it's been totally fine - Discord chat, Minecraft (local multiplayer with 1 other person), watching youtube, Photoshop (basic image editing), Shotcut (a quick-ish video edit).

Power related changes:
Voltage: 1.36 - I tried 1.33, 1.34, 1.35, and none were stable. 1.36 was the magic number.
CAD_BUS: 30-20-24-24 (30 was needed to stabilize the setup).

Timing changes:
tRP: 20 -> 21 (I didn't post this one last week, but on my old settings tRP 20 worked fine).
tRAS: 39 -> 42
tRC: 66 -> 68
tWTRL: 9 -> 8 (honestly, I think this was a mistake as I thought I put 10 here, but hey it worked so great!)
tRDWR: 10 -> 8 (thanks to the relation with tRDWR and tWRRD, when I had to bring RD up, I could bring WR down)
tWRRD: 1 -> 4
tRFC: 584 -> 586 (I actually think that I misunderstood the situation and didn't 'need' to increase this, but I've not tried lowering it back yet, that said the actual n/s change is minor from 313 to 313.956, so not much change there.)

How did I get to this point?
I found this amazing spreadsheet: https://docs.google.com/spreadsheets/d/1cJmhO62WHPLNKGBtsJV3BjdL-dcfJJeyhdSAoJmuzJE/edit?usp=sharing
The 'calculator' part is mostly useless as it does not support c-die at all, however, it has what I felt was a 'gift from God' - it has a section that explains what the errors on the TM5 tests relate to. So on previous versions, I got an error first on 7 (that's why I changed CAD_BUS), then on 1 (which is why I re-balanced tRFC and tWRRD - which meant that I had to change tRDWR as well.) All the other changes were made simply to boot and then clear memtest86.

Before this, I did try to get tRCDRD to 20, but there was nothing that I could do that would stabilize that, most of the time it would simply fail to POST. One time it got to the post screen but then it would lock up before begin able to load windows, so unless I find some magical settings with CAD_BUS and Voltage, I don't think my tRCDRD will go any lower than 21.

Next Steps
I want to get to 66.x n/s latency (greedy I know, but that's why I'm overclocking!)

That should be possible as I had to change quite a few of my sub-timings simply to boot, and so far I've not really done anything to optimize them.
I'm also tempted to see if tCL 16 is possible at all, but I have my doubts.

 

[Linepe]

Hallo zusammen,

aktuell versuche ich mich an den HyperX Fury HX432C16FB3AK2/32. Aktuell bekomme ich diesen Speicher auf 3200 nur unter Auto Timings. D.h. XMP deaktiviert und manuell 3200 eingestellt.
Einzig Gear Down und Power Down Mode sind manuell deaktiviert. Nun zeigen sich sporadisch Bluescreens unter Windows. (KMODE... IRQL...). Gleichen Verhalten zeigt sich wenn ich den Speicher auf 2933Mhz mit CL14 einstelle. Dabei läuft Memtest ohne Fehler durch.

Kann ich an den Timings & Subtiming etwas drehen um den Speicher stabiler zu bekommen oder bin ich an 2933Mhz gebunden?
Noch bin ich in der Lage den Speicher zurückzugeben. Obwohl diese Speicher auf der QVL von MSI steht.

Edit:

Memtest86: zeigt keine Fehler
Memtest5: Zeigt Fehler

D.h. der Speicher war in dieser Konfiguration nicht stabil.

Edit2:

Die einzig stabile Konfiguration scheint tatsächlich diese zu sein:

Falls noch Ideen kommen wäre ich dankbar. Andernfalls behalte ich diese Einstellungen.

 

[Nightguest]

In the end, my 3733CL17 was on the very edge of stability - I had 1 BSOD this week, so had to do some fixing.
A TM5 test left overnight gave me 2 errors, which pointed out what was wrong:

1. ProcODT/SOC was too low
2. tWTRS/L and tRTP had a slight hiccup.

Fixes:
ProcODT is now up the 40 (increasing SOC to 1.15v did not fix it).
tRTP 6 -> 8
tWTRL 8 -> 10

However, I was able to bring these down:
tWR 11 -> 10
tWRRD 4 -> 2
tRFC 586 -> 584 (in ns that was 313.96 -> 312.89)
My RAM seems to cap out at 312ns for tRFC, but that is totally on edge. 313 is much more stable. Though considering that most c-die kits go from 350ns to 320ns, I'm very happy with mine!)

Full timings:

DRAM Voltage: 1.36

Stress tests:
Memtest86: 4 Cycles (did this first outside of windows)
TM5 1usmus: 30 cycles
OCCT: 1h

And then played some games to chill out and all seems to be fine.

I'm not sure what more I can do with this now as tRAS and tRC do not want to go any lower (I'm assuming due to the tRC relation to tRFC ns).
tRP 'may' be able to go down to 20, but I think that will involve a lot more tweaking to the power.
Maybe increasing CAD Bus and upping voltage to 1.365 'might' do something, but I need a lot more time to see if it would. (1.37 is completely unstable).

Just for fun, I might try to get 3600CL16 working
I do have a stable setting for 3200CL14, but that needs 1.4v to be stable and it's just not as good as 3733CL18, let alone CL17. Higher FCLK all the way!

There is a new BIOS for my motherboard, but I'm kinda scared to update right now as ASUS tends to break things with each version. I might give it a go later on to see what happens there.

@Linepe
Sorry for replying in ENG (my written DE is horrible).
A few points for you to consider.

1. Ryzen 1000 and 2000 are much harder to take above 3200 (though with the 2700X, it should be possible to go up to 3600, but it's a huge silicon lottery play).
2. For your 3200 setting, use GDM one and set it to CMD2T to 1T (you really really do not want to be on 2T).
3. If possible, use ZenTimings are it displays your timings better: https://zentimings.protonrom.com/
4. If you really have your CAD Bus at 120 across the board, drop it to 24-20-24-24 (or 30-20-24-24).
   Reason: By using 120, you're pushing a lot more current through your system which stresses the IMC a lot more, on top of that you're generating a lot more heat. Both are bad as the IMC on Ryzen 2000 isn't the best (better than 1000, but still not as good as 3000), and c-die in general is very heat sensetive. On that point, what's your airflow like? A lot of c-die related instability is due to heat (I've got a fan directly above my RAM, without it I BSOD in about 5 minutes of using my PC, with it, I'm stable).
5. What is your ProcODT, and how much volts are you feeding your RAM?
6. I think that if you were to enable RttPark at RZQ/5, you'd get more stability. However, I have no clue how this works, just that 'it's needed' (on my setting at least), so I can't explain this one.

 

[KillAllTeammates]

Thank you very much for the tutorials. I'm still adding voltage foolishly without this!

Below is my current configuration.
Anhang anzeigen 534430Anhang anzeigen 534431Anhang anzeigen 534432
This is the result without optimized tertiary timings and the sweet spot of voltage hasn't been tested yet.
The reading speed seems too low comparing to the writing. The copying and latency are not good either. What could be the cause of this?
I think there may be two reasons. The first is the performance of 9400F and its IMC is not good enough for DDR4 3866. The other is the tCL and tRCD are too long and make the reduction of other secondary timings meaningless. The tRAS is only 34 now and there may even be room for further reduction. I think this is abnormal. But it's nearly impossible to continue reducting the tCL and tRCD.

Finally I found the problem after I' ve got a 9900KF
The two configurations below have exactly the same frequency and timing. The only difference is the CPU. This proves a bad CPU will indeed become the bottleneck of memory.

 

[Reous]

aktuell versuche ich mich an den HyperX Fury HX432C16FB3AK2/32. Aktuell bekomme ich diesen Speicher auf 3200 nur unter Auto Timings. D.h. XMP deaktiviert und manuell 3200 eingestellt.

Kannst dein Setting nochmal mit ZenTimings posten. Mit RTC werden leider nicht mehr alle Werte korrekt ausgelesen.

ZenTimings

ZenTimings is a simple and lightweight app for monitoring memory timings on Ryzen platform.

zentimings.protonrom.com

 

[Linepe]

Kannst dein Setting nochmal mit ZenTimings posten. Mit RTC werden leider nicht mehr alle Werte korrekt ausgelesen.

Die aktuelle Einstellungen. Bei der 3200er Einstellung ist der PC sehr instabil. Unter anderem fährt dieser nicht ordentlich runter und hoch. Automatisch befüttert das MB die Riegel mit 1,36V. Was ja schon ordentlich für C-Dies ist. Demnach habe ich mit Werten von bis zu 1,4V versucht. Die Ergebnisse waren stabiler aber nicht stabil genug für Windows.

Sorry for replying in ENG (my written DE is horrible).
A few points for you to consider.

1. Ryzen 1000 and 2000 are much harder to take above 3200 (though with the 2700X, it should be possible to go up to 3600, but it's a huge silicon lottery play).
2. For your 3200 setting, use GDM one and set it to CMD2T to 1T (you really really do not want to be on 2T).
3. If possible, use ZenTimings are it displays your timings better: https://zentimings.protonrom.com/
4. If you really have your CAD Bus at 120 across the board, drop it to 24-20-24-24 (or 30-20-24-24).
   Reason: By using 120, you're pushing a lot more current through your system which stresses the IMC a lot more, on top of that you're generating a lot more heat. Both are bad as the IMC on Ryzen 2000 isn't the best (better than 1000, but still not as good as 3000), and c-die in general is very heat sensetive. On that point, what's your airflow like? A lot of c-die related instability is due to heat (I've got a fan directly above my RAM, without it I BSOD in about 5 minutes of using my PC, with it, I'm stable).
5. What is your ProcODT, and how much volts are you feeding your RAM?
6. I think that if you were to enable RttPark at RZQ/5, you'd get more stability. However, I have no clue how this works, just that 'it's needed' (on my setting at least), so I can't explain this one.

No problem for replying that way. I'm glad for every helping hand. So as you see above there are the correct values for the stable and non-stable configuration. The CAD Bus values are corrected and I've tried to enable RttPark... But unfortunately I'm not consider myself an expert on that area, yet. So I don't know which value to use.

Regarding the Voltage, the Motherboard is automatically giving 1,36V to the RAM and I've tried to increase the number. This endet in a more stable System but not stable enough to run windows with 3200.

Thanks to both of you

 

[Reous]

Wenn ich das richtig sehe hast du die RAM Riegel in den Slots 1 und 3 von der CPU aus gesehen. Kann das sein? Wenn ja, dann bitte umstecken in Slot 2 und 4 und nochmal probieren.

 

[Linepe]

Wenn ich das richtig sehe hast du die RAM Riegel in den Slots 1 und 3 von der CPU aus gesehen. Kann das sein? Wenn ja, dann bitte umstecken in Slot 2 und 4 und nochmal probieren.

Ich glaube das zeigt Zen Timings tatsächlich nur falsch an. Die Riegel stecken in DIMMA2 und DIMMB2.

Ich hatte vor kurzer Zeit den selben Verdacht da nichts auf dem MB verzeichnet ist. (Ohne den Dark Rock Pro 4 abzuschrauben ) Aber das Handbuch zeigt es wie folgt:

Es sollte alles stimmen.

 

[Reous]

Was du mal probieren könntest wäre dann andere Rtt und ProcODT Werte zu nehmen. Ebenso kann es auch mal hilfreich sein die RAM Spannung zu erhöhen oder zu reduzieren. Zum Beispiel 1.375V oder 1.325V

RttNom 7
RttWr 3
RttPark 1
ProcODT 48