Overclock.net banner

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

4M views 25K replies 1K participants last post by  The_King 
#1 · (Edited)
AM4 Ryzen Memory Stability Thread


Please try to remember the following



Clarify what platform and CPU you are speaking about when asking a particular question or speaking about your experience.

Quote the user you are replying to when replying.

When posting stability results, be sure to include the CPU as described in the posting results instructions.

NOTE: SOC voltage should ideally not exceed 1.2v Anything above 1.25v has the potential to cause irreversible damage.

Recommended settings ca be found here in the guide provided by ASUS: http://www.mediafire.com/file/mciue95x0a2xfq7/C6H_XOC_Guide_v05.pdf




Happy posting!


Overview
This thread is dedicated to showing the various memory configurations of users with DDR4 on AMD Ryzen platforms.
There is no strict criteria here, all things AM4 memory overclocking welcome. However, to enter the stability chart certain criteria is to be met as this is generally speaking dedicated to showing what is obtainable on both platforms at an operational level.


If using ASUS you can also post here for help:



For stability results, using the recommendations found below and in the overview seem the most requisite on recent platforms:

Quote:Google stressapp test via W10 Bash Terminal(or another compatible Linux disti) is the best memory
stress test available. Google use this stress test to evaluate memory stability of their servers
– nothing more needs to be said about how valid that makes this as a stress test tool.


Windows 10 Installation
  • Install Bash Terminal: https://msdn.microsoft.com/en-gb/commandline/wsl/install_guide
  • Install the Google Stress App test by typing: sudo apt-get install stressapptest
  • Once installed open “Terminal” and type the following: stressapptest -W -s 3600
  • You can add argument "-M" and add the the amount of memory you wish to assign to the test (90% of available memory)
  • This will run the stressapp for one hour. The test will log any errors as it runs.

Mint Installation


Quote:HCI Memtest for Windows

HCI Memtest can be run via DOS or Windows. http://hcidesign.com/memtest/

An instance needs to be opened for each individual thread, covering a total of 90-95% of memory giving the OS a little breathing room.

As an example Ryzen 7 1700 - 16GB RAM

16 instances with 850MB per instance.





NOTE: Version 5.0 notes state that it's 30% faster than previous versions. For testing densities beyond 16GB - it's recommended you use 5.0 Pro.

http://hcidesign.com/memtest/





Stability Results

Please submit results with the following format.

GSAT Results
For sake of simplicity submitted results will only record primary timing sets, but feel free to show subsequent secondary and terts within screenshot.
Linux Mint's Stressapp test needs to be run for a minimum of 1 hour by typing stressapptest -W -s 3600 in the Terminal.
To take a screenshot in Terminal type: gnome-screenshot

HCI
HCI consider 1000% to be the 'golden standard' however for larger densities this can be time consuming. A minimal coverage of two laps (200%) is required to be added to the table for HCI for density over 16GB. 16GB or less requires a minimum of 4 laps (400%)


Example:

Silent Scone--7000@3.7---3200Mhz-C14-13-13-36-2T---1.38v---SOC 1.2v---BIOS 0902---Stressapptest---1 Hour--F4-3200C14-8GTZ
Or
Silent Scone--7000@3.7---3200Mhz-C14-13-13-36-2T---1.38v---SOC 1.2v---BIOS 0902---HCI---400%---F4-3200C14-8GTZ


NOTE: This is not a leaderboard, as it is not a benchmark. This thread's main purpose is to both discuss information and various results and to gauge what is possible between different configurations, DIMM capabilities, and CPU samples. Results are welcome all the way up the frequency spectrum. If it's obtainable, it should be posted!
.

Should go without saying that general system and CPU stability should be gauged via the suggested means before attempting an outright memory stability test.






Have fun!​
 
See less See more
3
#19,001 ·
Personally, then I'm very curious what effect cooling the back-side of the GPU chip has - as the active silicon is on the bottom of the whole "chip" piece, not the protective cover top.

As for RTT talk, then I can only say while I haven't blown up any DIMMs - that misbehaving (not "good") RTT's at high voltage seemed to cause spontaneous hard CPU crashes (blackscreen), which is not a good sign.
Since then I've always tried to massage RTTs as a priority, so I haven't seen any other MISHAPS in this regard. 👋🤡

oh and yes, RTT 5/3/2 (nom/wr/park) on 2x8 b-die seems a little odd
if 6/3/6 or 6/3/7 runs fine, probably just keep it there?
VDDP seems a little high for 2x8 3866, but maybe that's just me stuck in the past of 2x16 4000 running at 0.83v :- )
rdrd/wrwr SCL 2 may just cause problems and no perceivable gain over 4 (is the "usefulness" of this timing still unknown?)

Curiously you also didn't mention trying tCKE 9, which is a rather important thing for no setup delay 1T GDM off - and may very well be very explosive if not set.

I think that's all?

oh right there's also trying raising CPU (VDD) 1.8v 1P8 (Standby Voltage in aSUS speak??) to 1.85-1.93v and bottoming out procODT (28.2) to see if WTRL 8 (and RRDL 4, if it has any benefits??) runs stable - because that's a dirty hack that has worked on several b-die setups I've tried
incorrect 1.8v voltages are either "nothing" or cause occasional errors in TM5 1usmus_v3 #1, (#2 ?maybe?), #10 on settings that were previously stable (i.e. increase 1.8v voltage and bottom out procODT first, decrease WTR/RRD second - otherwise they cause similar errors 🤡)
Hi there
Forgot to mention when I play 1T off will lower all my subtimings included SCL to 4 or even 5
Image from veii
I ve set my tcke to 10. This helped me to avoid windows startup bsod, I will take your advice to low my VDDP to 0.83v and drop ProcODT to 28.2 and Raise my CPU1.8V to 1.85-1.93 (Hwinfo show default 1.824v)

And RTT yea Ikr, 5/3/2 its weird. but this is the only value I can pass 100cycles 1usmus TM5 and aida64. lol
But I'm pretty sure it won't work at 1T off

And vdimm, I have no idea 1t off . Maybe 1.6v? Or I just need to remain current one?
 
#19,003 · (Edited)
And vdimm, I have no idea 1t off . Maybe 1.6v? Or I just need to remain current one?
Whoah buddy, slow down - command rate has no input on voltage requirement. : )
That's almost exclusively primary timing speed (DDR4-4200 easily runs around 1.3v with slow enough primaries) which determines how much voltage you need to bonk it with.
(and higher frequency = faster primary timing for the same value)

Slower command rates are simply more lenient towards suboptimal timing alignments (tRAS or tRC cycles they call them??) - the fastest one (1T vs. 2T vs. 2.5T for GDM on) just requires you to run more proper (lower is not always better) RDWR/WRRD, RRD/WTR and technically RDRD/WRWR but I'm not sure about those on AM4.

which really just means "try nudging them around, one by one, maybe more than one-by-one, and see which one performs best"
y-cruncher 2.5 Billion Pi Digit times with a static CPU frequency are good for this
 
#19,004 ·
Looking for an opinion, I'm running 4 sticks of GSkill TridentZ non-RGB, if I run active cooling with fans, it's better to remove the heatsink of the sticks and run naked DIMM's?
What should be the best approach?
Thanks guys!
No, the heatsinks are crucial to absorbing the heat. Active cooling with a fan is about getting the heat off the heatsinks as efficiently as possible.
 
#19,005 ·
The non-RGB TridentZ heatsinks are surprisingly OK, just remove the plastic insert at the top for slightly better air cooling.
 
#19,007 ·
Is there is anyone on here with x570 aorus master with 5900x or 5950x b0 stepping and fully stable flck over 1900mhz? I'm on F35b with a 5950x that is fully stable at 1900mhz. However, while i can boot and run 1933/1966/2000 (and possibly higher) without any system crashes i keep getting whea errors no matter what Vsoc, VDDP, VDDG IOD and CCD voltages i try.
Tell me your bios version and to try it out.
 
#19,008 ·
Hey everyone!

Does anyone here has some knowledge on 4x8GB Hynix DJR kits on Zen 3?

I have recently acquired the kit KF436C16RBK4/32, which were one of the best bins I could find for my money. Other than that I could only find 4000Mhz+ ram which was just too expensive.
Turns out the kit is Hynix DJR, or so Taiphoon says:
Font Material property Screenshot Number Document


But I am unable to run at stock XMP, I get Memtest errors after a few hours, not many though. I also have some weird USB issues, where the controller seems to disconnect/reconnect everything. I get no WHEA errors, which seems to be memory related at this point. Those also causes stuttering and my G603 seems to lose connection sometimes.

All this doesn't seem to happen at stock clocks on the memory, but then the latency spikes as well as the motherboard just sets the ram to run at 2T.

Previously, I have tried upping the SOC voltage to 1.1V and VDDR to 1.36V, since those are fairly close to the stock XMP ones, to no avail, errors would still happen as well as the weird USB issues.

Is someone familiar with those kits on Ryzen and could lend some time to help?
 
#19,009 ·
I have notice another issue that should not be allowed.
Proof of RAM testing/stability should not be provided has a separate pic. It should all be in one single desktop screenshot.

Separate Zentiming and AIDA64 screenshot should not be allowed everything should be in one screenshot including the stability tests.
Everything is ultimately on the honor system.

If someone wants to cheat with their stability results--disadvantaging the community, and themselves, by obscuring real data--there honestly isn't much that can be done to stop them. Most of us have no reason to lie, but if a claim is extreme enough to require special proof, it's probably extreme enough to ignore entirely; there is no prize at stake here.
 
#19,010 ·
@Kuroihane
Start with providing a screen from Zentimings for your current settings when you boot XMP.
Might be your board just setts something wrong with AUTO settings that could be easily seen if we see what it does.
 
  • Rep+
Reactions: Kuroihane
#19,011 ·
Sorry that question, I'm confused..

Is

tWR = tRTP*2 (SR) OR tRTP*4 (DR)

OR

tWR = tCL(+N)

right, or both false / not optimal? For next bench/stability test I setup tRTP=8 and tWR=16, so I follow tWR=tCL on a DR setup.
 
#19,012 ·
I managed to nab a 3080/3090 active backplate for £55 on eBay. 2nd hand but hasn't been used. Never do auctions for tech stuff folks unless you have something with a high ceiling. Buy it now or risk losing out. An open box active backplate should have got nearer £90+ as long as unused.

So, I still get to watercool something else after the failed NVMe attempt lmao

View attachment 2560302

It's a sickness I tell you. Once you start watercooling your PC you end up wanting to watercool everything.

Time to redo some bends as well and get rid of right angle pieces.

Even although everyone says no point in watercooling the back of a 3080 the active backplate looks nice and my previous nickel backplate with thermal pads still got warm to the touch. So if passive cooling on the back helps a little, water will help a little more.

The addiction isn't just about performance, it's about driving those temp readings as low as possible!
So I can confirm active backplate changes on a 3080 are minimal, but not 'nothing'. Memory junction runs cooler and core tends to be between 1~3 degrees lower depending on wattage load.

Restriction wise, marginal changes. There isn't a dense fin-stack or anything on the backplate block. But that's now me with 4 rads, 4 blocks and 1 D5 lol. People on Reddit severely underestimate the power of a single D5 pump.

Unless you are obsessed with running it down at like 40~60% or something. I find 80% a good point for me now in terms of noise and flow.

All in all I'd say the biggest benefit for watercooling outwith GPU and CPU has been on memory, especially if you have B-die. I'd prioritise getting a water block for your memory over an active backplate (unless you have memory chips to cool) and NVMe (really not needed). My memory basically tracks water temp now to a few degrees max. Which means even on a 600w power supply load and all rads intake, my memory will be at like 33/34 degrees if water temp is 30~31 degrees.
 
#19,013 · (Edited)
@Kuroihane
Start with providing a screen from Zentimings for your current settings when you boot XMP.
Might be your board just setts something wrong with AUTO settings that could be easily seen if we see what it does.
Rectangle Font Screenshot Technology Parallel


There you go, after looking hard at the timings and XMP profile, I noticed that tRC and tFAW seems a bit different, tRC is 85 and TFAW 36 in XMP. tRFC is also 631 instead of 630.

Other than that, I don't think there were more timings related to it. I was able to somewhat stabilize the 3600 XMP settings pushing 1.37V on the ram. Sadly, this motherboard does not seem to have a VDDR readout...

I ran MemtestPro overnight and it got to around 1100% without errors. I also noticed less stuttering and the USB issues didn't happen after a restart yet.

Is there any timings that might be wrong since it's needing more voltage than stock XMP? Should I change the timings above to the XMP sheet?
 
#19,014 ·
Sorry that question, I'm confused..

Is

tWR = tRTP*2 (SR) OR tRTP*4 (DR)

OR

tWR = tCL(+N)

right, or both false / not optimal? For next bench/stability test I setup tRTP=8 and tWR=16, so I follow tWR=tCL on a DR setup.
It is tWR = 2x tRTP, which is a relationship for the setting in MR0 of the ddr for read or write with auto-precharge commands since the timings share the same address bits.

But you can set them to a different relationship if you want.

View attachment 2560387

There you go, after looking hard at the timings and XMP profile, I noticed that tRC and tFAW seems a bit different, tRC is 85 and TFAW 36 in XMP. tRFC is also 631 instead of 630.

Other than that, I don't think there were more timings related to it. I was able to somewhat stabilize the 3600 XMP settings pushing 1.37V on the ram. Sadly, this motherboard does not seem to have a VDDR readout...

I ran MemtestPro overnight and it got to around 1100% without errors. I also noticed less stuttering and the USB issues didn't happen after a restart yet.

Is there any timings that might be wrong since it's needing more voltage than stock XMP? Should I change the timings above to the XMP sheet?
XMP for those is "3602" MT/s, so that is part of why tRFC gets calculated to 631 instead of 630.

tRC through tFAW usually just get calculated from base speed nanoseconds on AM4 rather than loaded from the XMP values.

My G.Skill 3600 kits do the same thing since they are "3602" MT/s, and tRC is 80 something instead of 52 for the CL16 or 50 for CL15.

Were these a kit of 4x8, or individual? You would probably be ok through at least 1.45V if needed if these are really DJR, if somehow CJR might be better to work at 1.4. tRCD and tRP probably won't want to move much.
 
  • Rep+
Reactions: mvdev0
#19,015 ·
XMP for those is "3602" MT/s, so that is part of why tRFC gets calculated to 631 instead of 630.

tRC through tFAW usually just get calculated from base speed nanoseconds on AM4 rather than loaded from the XMP values.

My G.Skill 3600 kits do the same thing since they are "3602" MT/s, and tRC is 80 something instead of 52 for the CL16 or 50 for CL15.

Were these a kit of 4x8, or individual? You would probably be ok through at least 1.45V if needed if these are really DJR, if somehow CJR might be better to work at 1.4. tRCD and tRP probably won't want to move much.
Thank you for looking out!

Those are a kit, 4x8. I noticed that the box they came in are like 2 boxes of 2x8, so I am not too sure if there are differences on that.

Either way, I did an attempt at clocking it higher. I really wanted it to get to at least 3800 Mhz with a decent latency, but even up to 1.45V and loosening CL to 18 I could barely boot to windows before getting a BSOD, even had a random bricked BIOS message from the motherboard come up.

Maybe it is a very low bin DJR or just too many DIMMs for this Motherboard. I used to have B-Dies XMP at 3200 CL15. I was able to get to 3800 CL16 just using the Ryzen Ramcalc with that G.Skill kit, but it was 2x8 and recently 16GB has not been enough for me, hence why I tried to upgrade.

I did keep the other primary timings fairly "standard", the last setting I tried that booted to windows, but then had a BSOD was 18-20-20-40, rest on Auto. SOC had LLC at Level 2 and 1.112V, reporting dead on 1.1V on softwares, I could about load HWInfo before the crash happened.
 
#19,016 ·
Thank you for looking out!

Those are a kit, 4x8. I noticed that the box they came in are like 2 boxes of 2x8, so I am not too sure if there are differences on that.

Either way, I did an attempt at clocking it higher. I really wanted it to get to at least 3800 Mhz with a decent latency, but even up to 1.45V and loosening CL to 18 I could barely boot to windows before getting a BSOD, even had a random bricked BIOS message from the motherboard come up.

Maybe it is a very low bin DJR or just too many DIMMs for this Motherboard. I used to have B-Dies XMP at 3200 CL15. I was able to get to 3800 CL16 just using the Ryzen Ramcalc with that G.Skill kit, but it was 2x8 and recently 16GB has not been enough for me, hence why I tried to upgrade.

I did keep the other primary timings fairly "standard", the last setting I tried that booted to windows, but then had a BSOD was 18-20-20-40, rest on Auto. SOC had LLC at Level 2 and 1.112V, reporting dead on 1.1V on softwares, I could about load HWInfo before the crash happened.
Try a looser tRCDRD... 21...22... or even 23 Might be low bin kits.
 
#19,018 ·
Try a looser tRCDRD... 21...22... or even 23 Might be low bin kits.
Thanks again for helping! :)

I just tried that now, 21 didn't even get past booting, 22 and 23 got past booting and crashed into windows, just same as before, using 1.45V.

I decided to try setting things slowly to auto to see what the MB would train for, and the results were quite absurd. It does make me wonder how this runs at CL 16 3600.

At those timings, I honestly didn't bother trying to boot, but it was a slightly better situation than not even getting past the memory checks and into the Bios as it did before.

Font Line Screenshot Software Electronic device
Font Line Screenshot Technology Software
Font Line Software Screenshot Technology
Font Line Screenshot Technology Software
Font Line Screenshot Technology Software


I understand the motherboard isn't too happy training 4 dimms like that, but does the timings indicate this could be some other type of IC? I wonder which though. I tried throwing the exported Thaiphoon file into Dramcalc and it claims this is a "good overclocking memory" about the timings (don't think this makes that huge of a difference though).

At this point, is there anything else I should try for?
 
#19,019 ·
#19,020 ·
Thanks again for helping! :)

I just tried that now, 21 didn't even get past booting, 22 and 23 got past booting and crashed into windows, just same as before, using 1.45V.

I decided to try setting things slowly to auto to see what the MB would train for, and the results were quite absurd. It does make me wonder how this runs at CL 16 3600.

At those timings, I honestly didn't bother trying to boot, but it was a slightly better situation than not even getting past the memory checks and into the Bios as it did before.

View attachment 2560419 View attachment 2560420 View attachment 2560421 View attachment 2560422 View attachment 2560423

I understand the motherboard isn't too happy training 4 dimms like that, but does the timings indicate this could be some other type of IC? I wonder which though. I tried throwing the exported Thaiphoon file into Dramcalc and it claims this is a "good overclocking memory" about the timings (don't think this makes that huge of a difference though).

At this point, is there anything else I should try for?
Maybe the default ProcODT or RTT aren't very good for four dimms in your case. You could try ProcODT up or down a step, I would think Hynix could like 40. I don't know if B550-F might like something different for RTT.

If tRCD 22t doesn't work for 3733 or 3800, it is probably not the problem. Make sure you are also using tRP 22 - 23. You can try setting WRWRDD to 9t if that isn't what the motherboard tries to train anyway. Most other stuff is pretty high.

Might also check SOC up to 1.125. I think VDDG and VDDP could be adjusted, but I don't think they are necessarily your problem.

For four dimms it probably saves some headache if you can take the time to test each one at a time and figure out what they can overclock to (and secondarily make sure one isn't weak or faulty if not tested already).

This may not matter but try the VRM settings back on Auto. I didn't have very good luck with phase control "Optimized" on my Asus X370, but also 200 for switching frequency seems like it is lower than what I would expect for default (if it is in kHz).

Why does this show Samsung D Die for F4-3600C14Q-32GTZNA G.SKILL Trident Z Neo Series 32GB (4 x 8GB) 288-Pin PC RAM DDR4 3600 (PC4 28800) Intel XMP 2.0 Desktop Memory Model F4-3600C14Q-32GTZNA - Newegg.com
Perhaps I'm misreading or plain not "in the know" about something?

View attachment 2560448
Because most dimms don't have the die stepping / revision programmed into the SPD. I have no idea why thaiphoon makes this particular guess though. Those are B-die.

For G.Skill, look at the code on the heatspreader label starting with 042. It will end 8810B for 8Gbit Samsung B-die.

edit for typos
 
Top