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

[Alastair_S1D]

@athosdewitt Timing rules do exist. Some of them are soft rules, that can be broken, but can/will show reduced performance. Others are hard rules that get automatically corrected by the motherboard during training for very practical reasons. In general those rules are not* IC specific, so I don't want to go into detail about them, plus there are other resources available that thoroughly cover. For non-german users this is a good starting place for general information: https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4 OC Guide.md#tightening-timings

@Alastair_S1D Since S8C can have a low voltage threshold and will show a rollover effect if pushed beyond, I would recommend starting with less than 1.35V (real) and normalized VTT. Going up in frequency will eventually require adjusting at least tRCD, tRP and tRFC, while the required voltage usually keeps hovering around a certain value. The last pages of this thread and first post have screenshots from other users with full timings, RTTs etc at DDR4-3600 for orientation. The memtesthelper guide I just linked also answers most of your questions regarding general timing interactions.

ps: welcome to Hardwareluxx

Yeah I have started my journey at stock XMP values. So 1.35v GET with VTT at half that. How does drive strength effect C die? Higher values or lower values to try aid in stability?

 

[emissary42]

The exact combination for a certain frequency can vary from system to system and kit to kit. Starting from defaults or straight 24s and then testing some of the "known good" combinations, is what most people do.

 

[Alastair_S1D]

The exact combination for a certain frequency can vary from system to system and kit to kit. Starting from defaults or straight 24s and then testing some of the "known good" combinations, is what most people do.

Disabling RttWR doesn't help. Makes stability worse. I can't post XMP with RttWR disabled. So I am leaving RttWR on auto which = RZQ/2

 

[Alastair_S1D]

The ram responds positively to dropping ProcODT. I went down to 36 ohms. But that was too low. It seems to be lasting longer in stress tests at 43.6ohm.

 

[Don Solo]

Part Number K4A8G085W[B/C]-BCPB ---

Die Density / Count 8 Gb / 1 die

16-18-18-36-56 --- 1T -- GDM ENABLED (disabled is unstable)

1.35V DOCPP profile to 3200.
https://www.gskill.com/product/165/184/1536139016/F4-3200C16Q-8GVKB-Overview is my RAM.

I try lower than 1.35V and it does not post, at all at 3200.

I'm so incredibly confused, I've set timings as high as 20-22-22-42 and it still won't post above 3400, have tried voltages as low as 1.17 -- 1.23, only 1.35 or more works for posting.

Here's TY screenshot.

Gyazo

gyazo.com

@emissary42 -- is there any advice you can give me? I've tried to change CADS, etc -- it just isn't working, I've spent 24+ hours overall trying to stablize my RAM after seeing you guys going 3600+ on this bullshit, I've been told it was Samsung C-Die and on the white sticker, it does indeed have a C as the last letter on the serial.

 

[emissary42]

If typical S8C settings from other users don't work at all, I'd test the pairs in the two primary dimm slots, since no current platform should have any issues with 2x8GB @ 3600. If both halves of the kit can manage stable 3600 with typical settings, you know that the memory itself is not the issue.

 

[Operation.Void]

Hi all! I'm back with a new motherboard and a new OC! I've finally managed to hit 3800Mbps at 1.344V on my 2666Mbps G.Skill Value RAM with an MSI B450M Mortar Max. I've tested for probably an accumulated 6 hours since I ran TM5 with every group tightening, so the 1 hour test in the image is no cause for concern, it was just my last test. I've noticed that moving from a 2 slot board to a 4 slot board tFAW does not go anywhere near as low as it used to, bottoming out at 36 on the new board, but other than that memory overclocking ability has been relatively similar other than the CPU being able to handle 1900 FCLK.

Voltages are as follows:

VDIMM: 1.32V set 1.344V read
VSOC: 1020mV set 1006mV read
VDDP: 950mV set
VDDG CCD: 850mV set
VDDG IOD: 1000mV set

And a side question: Should I enable the DRAM latency enhance setting on my MSI board? MSI don't make it clear as to what it actually does.

Update: And today the overclock is unstable I couldn't even extract archives without errors, and Anta777 failed in under a minute. Though I noticed in the BIOS that my VDDP was at 900mV, even though I had thought I put it at 950mV, so I am testing 950 now to see if that was the issue.

Update 2: VDDP did not help. It turns out that a few subtimings were a little off, so I had to up tRRD_S and L to 6 and tFAW to 40. Now I'm getting WHEA Error ID 20 logged every 1-2 days, so I will be looking into whether it's VCore or VSOC related.

Update 3: I've been getting random BSODs and errors at 3800C18 all of a sudden so I've stepped back down to 3733 unfortunately. Oh well, I got more than I bargained for anyway.

 

[Alastair_S1D]

@emissary42

So it looks like I got 3636mhz at stock XMP (101Mhz X 18 x DDR). I pushed up Rtt_Park to /2 and Rtt_WR to /3. I am running 1.34V DRAM set = 1.37V GET. 2.62v DRAM VDDP.

I cant get the primaries to drop lower than the XMP 18-20-20-20 but I did manage to get tRAS to 40. I cant get tRC to = tRAS+tRCD. I can get 62 which is tRAS+tRCD+2.
When it comes to the DrvStr is a lower Ohm value a stronger strength or weaker. I imagine stronger?

Which secondaries make the most difference to stability? tRFC?

 

[emissary42]

Apart from your primaries and tWRRD, most secondary/tertiary timings are pretty loose, even for S8C. Our tRFC table (by Reous) is linked in the first post. 350ns are very safe. So if you have stability issues, those timings are probably not the culprit. Going up in frequency an increase of tRCDRD to 21 can/will probably be required and tWRRD 4 might be help with stability in general.

7/3/2 and 20/20/20/20 is not exactly a common configuration, at least to me. Since I am very busy with DDR5 nowadays and never ran a similar configuration to yours, I can't really help you with specific settings. So if someone else want's to step in, that would be appreciated.

 

[Alastair_S1D]

Apart from your primaries and tWRRD, most secondary / tertiary timings are pretty loose, even for S8C. Our tRFC table (by Reous) is linked in the first post. 350ns are very safe. So if you have stability issues, those timings are probably not the culprit. Going up in frequency an increase of tRCDRD to 21 can / will probably be required and tWRRD 4 might be help with stability in general.

7/3/2 and 20/20/20/20 is not exactly a common configuration, at least to me. Since I am very busy with DDR5 nowadays and never ran a similar configuration to yours, I can't really help you with specific settings. So if someone else want's to step in, that would be appreciated.

Yt I still don't understand what the settings actually DO! RTT_Park and RTT_WR do? What does Drv strength do? What settings are better? I'm just literally throwing stuff at the wall at this point and seeing what sticks with little understanding to what I am doing. I understand timings. That's a no bainer. But drive strengths and this termination stuff. No clue. I don't know what's good or what's bad or what might actually harm my components?

I can't get my tRFC below 575t. That seems kinda bad. What can I do to get better tRFC. And do the settings mentioned above help with that?

 

[emissary42]

If you want to dive deeper into understanding what all those settings actually do, beyond the paywalled JEDEC DDR4 spec there are documents readily available from all the big DRAM manufacturers. From among my DDR4 bookmarks: Samsung DDR4 DRAM Specification - Device Operation & Timing Diagram (PDF). Some of the Micron datasheets also are surprisingly detailed and worth a second look: Micron 8Gb DDR4 SDRAM Datasheet PDF). These documents cover most of the non-platform-specific stuff.

What settings are better?

For many of these settings there is no universal "best". Even two systems built from identical hardware can require different settings for a specific overclock. There might be general tendencies towards a certain range of settings, however. And that is where overclocking results from other users with similar configurations can come in handy. Ultimately you sill have to find the working configuration yourself. It can be helpful to have a starting point that is closer to the finish line, though.

I can't get my tRFC below 575t. That seems kinda bad. What can I do to get better tRFC. And do the settings mentioned above help with that?

Like many other ICs the tRFC does not scale with voltage for S8C. While optimal settings can help reduce the minimal value a tiny bit, don't expect any sort of miracle. That is the reason why the tRFC table exists in the first place.

 

[zagreus1]

Hi everyone, I'd like to post my results with 4x16GB C-die. After days and nights of reading and testing, I finally stabilised the damn thing at 3200 and then tweaked the timings, maybe not to their limits, but to an acceptable level for C-die. The main thing that helped was kowing that it needs low voltage (1.32 VDIMM) and that the Memory Fast Boot option had to be disabl;ed at the BIOS and not only on Windows. Here are my BIOS settings:

• DRAM Frequency 3200
• FCLK Frequency 1600
• UCLK=MEMCLK
• Memory Fast Boot Disabled
• VCore Offset -0.0875V
• NB/SoC 1.05
• CPU LLC 3, NB/SoC LLC 3
• VDDP 0.9V
• VDDG 0.95V
• DRAM Voltage 1.32V

And here are my timings

I found it impossible to get anything higher than 3200.
Thank you for all the help!

 

[Krümel]

I found it impossible to get anything higher than 3200.

You have 4 Ranks per Channel, so I think that is normal. Higher frequencies could be reached with 2 or 1 DIMM-Rank per channel.

 

[zagreus1]

You have 4 Ranks per Channel, so I think that is normal. Higher frequencies could be reached with 2 or 1 DIMM-Rank per channel.

True. Also the kit I have is this CMW32GX4M2E3200C16 which is specced at 16-20-20-38, meaning it's worst than the majority of C-die kits in this thread that are specced at 16-18-18-38, so that made things more difficult. Anyway, it was a great learning experience! 😁 Hope this helps someone.

 

[Ryciafaustes]

I created an account on this site to add on to this post since there's so little documentation available.

https://pcpartpicker.com/product/Dw...8-gb-ddr4-3600-cl18-memory-f4-3600c18d-16gtzn
This model of ram comes with the C-die.

I got stuck with a pair of two of these about a month ago, and had issues as to where it wouldn't run on manufacturer's default XMP.
G-skill Trident-Z Neo 3600 CAS 18 should be added to this list.
Here is a reddit post regarding the problems I've had with it.

Reddit - Dive into anything

www.reddit.com

 

[emissary42]

I have added them to the first post. However with your system you shouldn't have any issues running the XMP. Just make sure the modules are placed in the two primary memory slots and that the motherboard does not overvolt Vdimm too much. If you still have trouble with stablity go down one step in frequency to DDR4-3533 or try full manual overclocking instead. The timings shouldn't really be an issue since your spec is easy even for S8C, but you might have to sort out the the ODT, CADs, SoC- and IO-voltages.

C-Die WWW Reviews
1) Corsair Vengeance RGB RS CMG32GX4M2E3200C16 @ PauseHardware.com
2) Corsair Vengeance RGB RT CMN32GX4M2Z3600C18 @ TweakPC
3) AITC Kingsman Gaming RGB AID416GC16RGB @ HKEPC

 

[Ryciafaustes]

I have added them to the first post. However with your system you shouldn't have any issues running the XMP. Just make sure the modules are placed in the two primary memory slots and that the motherboard does not overvolt Vdimm too much. If you still have trouble with stablity go down one step in frequency to DDR4-3533 or try full manual overclocking instead. The timings shouldn't really be an issue since your spec is easy even for S8C, but you might have to sort out the the ODT, CADs, SoC- and IO-voltages.

C-Die WWW Reviews
1) Corsair Vengeance RGB RS CMG32GX4M2E3200C16 @ PauseHardware.com
2) Corsair Vengeance RGB RT CMN32GX4M2Z3600C18 @ TweakPC
3) AITC Kingsman Gaming RGB AID416GC16RGB @ HKEPC

I went full manual and that worked fine. My board wanted to clock it to 1.55V for god knows what reason otherwise.
Slots are fine and correct, 2,4.
I shouldn't have to do much more than tick on XMP, but this die seems to say otherwise.

 

[emissary42]

Oh boy, 1.55V is a very fast way to kill your modules. Glad you were fast enough to notice it before anything bad happend.

 

[Ryciafaustes]

Oh boy, 1.55V is a very fast way to kill your modules. Glad you were fast enough to notice it before anything bad happend.

Oh yeah.

 

[gusss]

So I have this C-die RAM and I've been trying to overclock it for a while.

I managed to get it to 3733 18-21-21-42-68 @1.34v stable but without tightening the secondaries and tertiaries too much (I've done some tightening but I reckon I still got some work to do there) but it occurred to me that I could at least try my hands at 3800 first to see if was stable as well, basically just set everything to auto aside from the primaries, tRC and tCWL:

I ran a few cycles of TM5 with the 1usmusv3 cfg and I got a couple of #0 errors and I didn't get any other errors during it, which according to a ryzen calculator spreadsheet means that I probably should add more VDIMM. I bumped it to 1.35v and ran for a good 40 minutes without errors but there's a problem now: my second monitor turns off and my main monitor blinks as if it was turned off and on every few minutes.
I went to event viewer to see what was up and I get the gpu driver crash error "Display driver nvlddmkm stopped responding and has successfully recovered." at the exact moment my monitors start getting messed up, I even ran the test with my gpu at stock and it didn't seem to help.
The funny thing is that at 1.34v the overclock is unstable for TM5 but it's almost fine at 1.35v, with the exception of the gpu driver crash above. Has anyone ran into this situation before and knows a way to fix this or is 3800 just not doable at all for me?

 

[emissary42]

Since your timings are pretty loose, you will probably have to touch other settings (ProcODT, RTTs, CADs) to stabilize 3800 and/or have to dial down voltage a bit. As for memory voltage maybe try 1.345V instead of going straight to 1.35V, it might seem minuscule but can make a difference. In case of running into a brick wall, you can still optimize timings like tRRDs/l, tFAW, tWTRl, tWR, SCLs, etc at 3733, to improve bandwidth and latency. Performance-wise properly tweaked timings at 3733 should put you way beyond 3800 with auto-timings.

 

[gusss]

Tried playing around with ProcODT, RTTs and CADs (didn't manage to try 1.345V since my motherboard only works in 10mV steppings so not able to do it) but it was to no avail. Decided to go back to 3733, turned GDM on so I could get tighter timings and lower voltages like VSoC, VDDP and VDDG - I'll be trying GDM off once I'm happy with what I have - and really just wanted to come here post an update in case someone out there has similar hardware. I ran TM5 over night with the 1usmus cfg and didn't get any errors whatsoever.

One thing though that seems weird to me and that I wanted to add: it feels like I can't really lower my tRC that much more - might have to change IMC related voltages perhaps - but tRFC, which usually gives me trouble when trying to tighten it, seems to be working fine with lower values, I think I could maybe even lower it by 10ns more, hopefully it works.

 

[athosdewitt]

hallo , i'm stuck and seems i can't figure out on my own

1 should i keep my values or use tCWL 9 or 12 with TRDWR 11 TWRRD 0 respectively 8 and 4?
2 tRFC i can drop to around 470 but idk if i need and of course how to change values of trfc2 trfc4 or to get rid of spamming errors if i play with tRFC
3 it's there a thing over tight ? do y'all recommend to increase something
4 should i try to lower any timing? sadly for main timings really can't 14-16-16 throws errors instantly then bsod etc. trcdrd likes to be at 17 up to 3200mhz(worse results gdm on forced, without erros)
some quick & basic guidance would be fantastic!

 

[emissary42]

@athosdewitt I personally would not invest any more time in 2933 + CL14 and instead try to figure out the maximum frequency you are able to run with CL16/CL18 and GDM on. Then see if you can get GDM off working at your resulting frequency and start tightening the timings.

In general S8C is definitely not the IC for tight primaries, since running them tight requires voltage increases that reduce frequency headroom. In most cases trading frequency for timings ends up not being worth it. That is why 3333/3466 CL16 and 3600+ CL18 are what most people end up with.

2) The tRFC frequency table is attached to the first post. Unless your board is bugged, just ignore tRFC2/4. They won't be used under normal circumstances.

 

[athosdewitt]

at least on my end with these c-die and this b450 motherboards Trfc frequency table fallows the numbers at certain speed i just copy paste the numbers from 2666mhz and gives tRFC 467 318ns

above 3000 it's a dead horse throws too fast errors without GDM ON but 100% stable with it. it's not that i don't like gdm on but feels wrong to use it just to fix a bunch of errors. i did post at some point 3533 with 18-22-22 1,37v (refuse to post bellow this voltage at 3400 or above ) rest auto around 67ns then bsod , still was a nice sight

that table it's spot on as i recall around 512 was the last stable trfc with 3200 gdm on.

edit: there she blows.. tried 18-19 and 18-20 3400 2t ( i'm not confident if gdm on instead 2t helps) rest auto getting 30 errors with anta at around 4 min. 67.something ns was impressive for c-die with Ryzen+ combo tho.
i'll stick with 3000 for a while

 

[athosdewitt]

UPDATE: my silly c-die likes ''normal'' voltages and freeze at 1.34 and lower dafuq!? I do tend to be stubborn! cl15 3000 1.36v bottomed to the ground. any other lower number and refuse to post. *TRDWR can be lower if i increase TCWL. I think i did a decent job?
since gdm 3333/3400-3533 2t post but it's a no go without errors idk 3200 cl16-18-18 GDM with worse subtimings doesn't feel faster
edit:
<= 68ns unstable 3400 18-19-19-38.

 

[ph00p]

Motherboard: Gigabyte B550M AORUS PRO-P

CPU: AMD Ryzen 5700G

RAM: Corsair CMK16GX4M2B3200C16 v4.32

Has anyone found stable settings for this kind of configuration at 3200Mhz? Or is 3600MHz stable possible?

I'm completely new to OC'ing, sorry. But if anyone has settings I'd greatly appreciate it.

I could post back about my findings with any timings I'm given, but right now I have nothing to offer sorry.

I'm worried I should return this ram with all the issues I've read with people not being able to get to even 3200Mhz with it, people have made it sound like this is the worst ram on the planet.

 

[emissary42]

Welcome

In general any S8C based kit should be able to hit 3600(+) on a 3rd gen MB. Check the first post for OC results with settings.

 

[Dubstec]

Ich habe aktuell 3800 CL 18-22-22-42 bei 1.35V oder 4000 CL 18-22-22-42 bei 1.4V.
Habe den G.Skill Trident Z Neo F4-3600C18D-32GTZN als Basis der scheinbar laut Thaiphoon Burner ebenfalls auf Samsung C-Die setzt.
läuft in 1:1 mit meinem 5800X stabil bisher kein Problem gehabt.
Bei 3600 CL 16-19-19-39 stieg das System mit 1.35V schon aus, aber noch nicht so das ein CMOS Reset von nöten war, sondern er 3x booten wollte und vermutlich gerade so zu wenig Spannung hatte, denke 1.375V wären dann stabil.
Sind die Werte für C-Die gut, schlecht, durchschnitt, geht da mehr? Ich hab in dem Forza Horizon 5 Bench jedenfalls bei 4000 CL18 ca. 10 FPS mehr im Max und 5FPS im Low zu den 3800 CL18 Setting.
Denke mal 3600 CL18 Stock wäre nochmal schwächer im Bench, hab ich aber nicht getestet, interessant wären halt noch mal zu sehen wie sich da 3600 CL16 verhält - alles immer in 1:1 IF Takt, meiner läuft mit 2Ghz stabil ohne Anpassungen, nur -0.05V Offset bei der CPU VDD Spannung bei 1.225V max SoC Voltage bei 1.1V max laut HWiNFO.

Hab null mit RAM OC am Hut, das Letzte mal vor Jahren mit DDR3 Speicher den ich ohne Änderungen von 1866 auf 2133 packen konnte und nie was anderes dran gemacht hab - der lief so Jahre stabil und gut war.
Jetzt dachte ich gerade mit Ryzen wollt ich schon das Meiste an SPIELE Leistung rausholen, mich interessieren Synthetische Benchmarks nicht, wichtig ist halt die reine Spieleleistung.
Wollte die paar Werte nur mal da lassen.

Jetzt lese ich aber paar Seiten zuvor C-Die "maximal 1.35V" - sollte ich also den auf 3800er laufen lassen oder kann man den auch auf 1.4V 4000er laufen lassen? Temp technisch habe ich dank Sensor keinen Unterschied bis auf maximal 1 Grad erkennen können, irgendwo um die 38-40 Grad.
Weiss man da inzwischen was da jetzt wirklich auf Dauer "gut" für den RAM ist und machbar ist? über 1.4 wollt ich nicht gehen, aber dachte, wenn die CPU schon so hoch geht stabil wäre es natürlich cool wenn der RAM entsprechend mit rennt.

 

[emissary42]

@Dubstec Schau mal auf den Aufkleber der Module, ob der Lot Code wirklich auf -8810C endet oder vielleicht nicht doch eher auf -8810D. Für S8C wäre schon recht ungewöhnlich, dass 4000 @1.4V so problemlos laufen, für S8D nicht so sehr. Außerdem Glückwunsch, wenn du 4000 1:1 tatsächlich stabil und ohne WHEA nutzen kannst