What did you have to do to get 128GB ram working on Zen3 CPU?

[Deleted member 185158]

I'm at a point where I am not getting errors anymore (well the errors seem to occur only right before the reboot). Just reboots are occurring after leaving the machine idle for a few hours. I can do 2hr OCCT runs, 3hr zip jobs, 3hr handbrake jobs no problem. Even after thermal cycling the ram from 30c (now cold winter room) to 55c between runs. I just realized I left the new ram kit in the machine. I'll try swapping it out with the previous kit, maybe the reboot issue is simply with the new kit.

Error 41, sudden loss of power in event viewer?

Seems to be an SK Hynix and AMD cpu problem.

Buy running 128gb at 3200mhz (the memory divider itself or IF : Memory divider)

Can't put a finger on it, buy generally more ram = less speed for redundancy (stability) because that's a lot of memory for the IMC to try and handle at higher frequency.

2933mhz memory and IF would be just fine. After all you really only desire quantity and not performance, at least that's what I gather from needing such a large amount of system memory.

 

[A Computer Guy]

Error 41, sudden loss of power in event viewer?

Correct Error 41, The system has rebooted without cleanly shutting down first. This error could be caused if the system stopped responding, crashed, or lost power unexpectedly.
I am running off of battery backup. Just tested it to make sure the battery is functioning properly and it is.

Seems to be an SK Hynix and AMD cpu problem.

These are micron IC's. At least they are supposed to be. They do have the micron logo on the chips.

Buy running 128gb at 3200mhz (the memory divider itself or IF : Memory divider)

Can't put a finger on it, buy generally more ram = less speed for redundancy (stability) because that's a lot of memory for the IMC to try and handle at higher frequency.

Generally from what I know that is the typical expectation. You always gambling with the silicon lottery running out of spec. I was kind of surprised I could even get 128GB of 4xDR 3200 to work as I had low expectations going into it. The docs say on this board in 4xDR should run at 2666 but even at that speed the reboot problem still occurs even when trying to stabilize the system with increased voltages.

Other interesting problems that occurred were
- The Clear CMOS button stopped working. When I was trying to test various ProcODT configurations I couldn't get the UEFI to reset unless I pulled out the 2nd channel of ram.
- It won't even post if I don't preset the voltages (SOC in particular)

Neither of the two additional problems above happened when I was running a different config of 64GB DRx4.
There is some fundamental compatibility problem with this configuration of 128GB of 4xDR Nemix ram and my board I think.

2933mhz memory and IF would be just fine. After all you really only desire quantity and not performance, at least that's what I gather from needing such a large amount of system memory.

Yea, at 128GB I was prepared to accept the typical tradeoff in speed vs. capacity but even dropping it to 2666 the reboot problem wouldn't go away.
Interestingly it only happens after the windows lock screen in engaged and I leave it for at least a few hours. Really annoying that it works fine otherwise.

 

[Deleted member 185158]

Correct Error 41, The system has rebooted without cleanly shutting down first. This error could be caused if the system stopped responding, crashed, or lost power unexpectedly.
I am running off of battery backup. Just tested it to make sure the battery is functioning properly and it is.


These are micron IC's. At least they are supposed to be. They do have the micron logo on the chips.


Generally from what I know that is the typical expectation. You always gambling with the silicon lottery running out of spec. I was kind of surprised I could even get 128GB of 4xDR 3200 to work as I had low expectations going into it. The docs say on this board in 4xDR should run at 2666 but even at that speed the reboot problem still occurs even when trying to stabilize the system with increased voltages.

Other interesting problems that occurred were
- The Clear CMOS button stopped working. When I was trying to test various ProcODT configurations I couldn't get the UEFI to reset unless I pulled out the 2nd channel of ram.
- It won't even post if I don't preset the voltages (SOC in particular)

Neither of the two additional problems above happened when I was running a different config of 64GB DRx4.
There is some fundamental compatibility problem with this configuration of 128GB of 4xDR Nemix ram and my board I think.


Yea, at 128GB I was prepared to accept the typical tradeoff in speed vs. capacity but even dropping it to 2666 the reboot problem wouldn't go away.
Interestingly it only happens after the windows lock screen in engaged and I leave it for at least a few hours. Really annoying that it works fine otherwise.

It's just luck of the draw. This system seems picky with memory given all the variables.

Micron has low bin dies too, not always great for performance....

But I'd be curious the stability at Jedec 2667mhz. It's not a great decrease in bandwidth, but you might be able to cut latency with tighter timings.

Disable power down mode so the memory doesn't throttle Vdimm. Might help at higher frequencies.

 

[A Computer Guy]

It's just luck of the draw. This system seems picky with memory given all the variables.

Micron has low bin dies too, not always great for performance....

But I'd be curious the stability at Jedec 2667mhz. It's not a great decrease in bandwidth, but you might be able to cut latency with tighter timings.

Disable power down mode so the memory doesn't throttle Vdimm. Might help at higher frequencies.

Power down was disabled in all cases. I put a ticket in with ASRock so I'm going to wait for some feedback from them and see if I can get lucky twice getting a UEFI/BIOS update that fixes ECC ram compatibility issues. In the mean time I think I've finally pushed the system too far and the last reboot damaged my windows installation. I haven't had to repair a windows installation since the XP days so I have some research to do.

 

[tabascosauz]

@MIRTAZAPINE runs 4x32GB 3200CL16 on 3900X

Power down was disabled in all cases. I put a ticket in with ASRock so I'm going to wait for some feedback from them and see if I can get lucky twice getting a UEFI/BIOS update that fixes ECC ram compatibility issues. In the mean time I think I've finally pushed the system too far and the last reboot damaged my windows installation. I haven't had to repair a windows installation since the XP days so I have some research to do.

Corruption generally doesn't come from unstable UMC or Fabric, it comes from unstable RAM. If Windows is too damaged to boot, it's time to do some memtesting. Which Micron ICs are these?

The fact that it's 128GB doesn't matter as much as the fact that it's quad rank per channel, like 4x16GB B-die or DJR. Expect to have to run higher VSOC and maybe VDDGs to support it.

 

[Deleted member 185158]

Blue screen during post type corruption.

 

[jesdals]

I would sell your setup and go for a AM5 setup it would be a lot easier

 

[Deleted member 185158]

Wipe the OS and start over

 

[A Computer Guy]

@MIRTAZAPINE runs 4x32GB 3200CL16 on 3900X



Corruption generally doesn't come from unstable UMC or Fabric, it comes from unstable RAM. If Windows is too damaged to boot, it's time to do some memtesting. Which Micron ICs are these?

The fact that it's 128GB doesn't matter as much as the fact that it's quad rank per channel, like 4x16GB B-die or DJR. Expect to have to run higher VSOC and maybe VDDGs to support it.

I can boot just fine. The damage so far that I know of so far is Windows Search is crashing when I try to use it. (returned to my previous 64GB stable config) I didn't make a link to Event Viewer so I typically click the windows icon and type it in to find it and that's when I discovered it was broken. Did a disk check and it repaired something.

Nemix 64GB ECC kit ( 32GB modules, DDR4-3200 CL22 1.2v )
Chip: Micron OTE75-D9ZFV (all modules same chip)
Lookup https://www.micron.com/support/tools-and-utilities/fbga?fbga=D9ZFV#pnlFBGA
part no: MT40A2G8JC-062E:E

Below are the voltages I was using. Note at 2T, DRAM 1.35v, and the voltages below there were no errors after many hours of testing. MemTest86, OCCT sse and avx2, and very large 7zip operations, all cores trying to stress the system. It's only after leaving the computer idle for a period of time after the lock screen somehow it gets unstable spews out some WHEA errors right before it dies and reboots.
I also tried SOC up to 1.2v and that didn't help.

Wipe the OS and start over

Pretty sure I'm at that point. On an old 3-4yr install guess it's time for a good old refresh anyway.

I would sell your setup and go for a AM5 setup it would be a lot easier

I'm gonna skip the early adaptor route this time. I've already punished myself last time around but yet here I am punishing myself again with 128GB ram.

Blue screen during post type corruption.

If it was a blue screen I was asleep or behind a lock screen at the time it happened. Could never get it to happen while I was at the computer.

 

[tabascosauz]

@A Computer Guy if Powerdown is already disabled, you can give disabling Cstates a shot to see if it helps.

Also try bumping 1.8V PLL, up to a max of 2.0V. One of the tools to support Fabric stability.

 

[junglist724]

5950X only officially supports 2666 with 2R 2DPC. I wasted a ton of time trying to get it stable at 3200 with 128GB of ECC RAM but my NAS would always end up randomly rebooting eventually. I just run at 2666 now with a couple hundred days of uptime so far.

 

[A Computer Guy]

5950X only officially supports 2666 with 2R 2DPC. I wasted a ton of time trying to get it stable at 3200 and but my NAS would always end up randomly rebooting eventually. I just run at 2666 now with a couple hundred days of uptime.

Tried that. Reboot still occurs at 2666.

 

[MIRTAZAPINE]

@MIRTAZAPINE runs 4x32GB 3200CL16 on 3900X



Corruption generally doesn't come from unstable UMC or Fabric, it comes from unstable RAM. If Windows is too damaged to boot, it's time to do some memtesting. Which Micron ICs are these?

The fact that it's 128GB doesn't matter as much as the fact that it's quad rank per channel, like 4x16GB B-die or DJR. Expect to have to run higher VSOC and maybe VDDGs to support it.

Not sure how well I can advise on this. Not quite verse in ram tweaking. All the things I did for mine is enable Docp/xmp on my ram in bios and turning off cstates and power saving functions to get it stable. 128GB all slot can be rather tricky. My previous gigabyte b550m pro-p is pretty much plug and play no turning off of cstate but on this asus it need abit of tweaking. Out of box with my current board before that is very infrequent blue screen. Also found that something as simple as not seating ram well enough could blue-screen too.

I corsair is 2 kits of 2x32GB of the same make with both being an unknown Samsung die..

Tried that. Reboot still occurs at 2666.

Have you tried other type of ram to see if the behavior happen?

 

[A Computer Guy]

Not sure how well I can advise on this. Not quite verse in ram tweaking. All the things I did for mine is enable Docp/xmp on my ram in bios and turning off cstates and power saving functions to get it stable. 128GB all slot can be rather tricky. My previous gigabyte b550m pro-p is pretty much plug and play no turning off of cstate but on this asus it need abit of tweaking. Out of box with my current board before that is very infrequent blue screen. Also found that something as simple as not seating ram well enough could blue-screen too.

I corsair is 2 kits of 2x32GB of the same make with both being an unknown Samsung die..

I'll try turning off cstates at the next opportunity since this is the 2nd suggestion hinting at this. The fact the problem only happens when the computer goes idle for awhile kinda of points me into a direction of power management, perhaps some power saving feature is my guess. It's seems I can't edit my poll anymore otherwise I would add the "Had to adjust cstates" option.

Have you tried other type of ram to see if the behavior happen?

Yes, I have another set of dual-rank ram 64GB (4 dimms) that worked without issue.

Also try bumping 1.8V PLL, up to a max of 2.0V. One of the tools to support Fabric stability.

I hadn't considered this voltage before. I'll put it on my list of things to try.

 

[Wirko]

I'll try turning off cstates at the next opportunity since this is the 2nd suggestion hinting at this. The fact the problem only happens when the computer goes idle for awhile kinda of points me into a direction of power management, perhaps some power saving feature is my guess. It's seems I can't edit my poll anymore otherwise I would add the "Had to adjust cstates" option.

Try to determine if the reboot occurs after a predictable amount of time, which would hint at Windows messing with power states. Indeed you're left with nothing logged just before the reboot. But when you log back in, you can check the system uptime and calculate when it happened.

 

[Mussels]

It's only after leaving the computer idle for a period of time after the lock screen somehow it gets unstable spews out some WHEA errors right before it dies and reboots.

This just screams typical ryzen "needs more voltage"
Both too aggressive curve undervolts and too-low SoC voltage give that behaviour

 

[A Computer Guy]

This just screams typical ryzen "needs more voltage"
Both too aggressive curve undervolts and too-low SoC voltage give that behaviour

I'm not doing CO at the moment (just running stock) and SOC up to 1.2v didn't seem to make a difference.

Wipe the OS and start over

I am just about ready to start retesting again with a good baseline. I finally got my OS reinstalled, validated no errors or reboots with original 64GB kit, and imaged the OS this time on an External USB 3.0 SSD for fast and easy recovery if something goes wrong again.

It was a bit of a pain and I ended up doing it twice. I forgot Kaspersky did this HTTP intercepted proxies thing for more "secure" browsing. I was able to turn off before (maybe back in 2017-ish) when I noticed they introduced that but couldn't seem to turn it off anymore this time around. I noticed it when the TPU website fingerprints weren't matching up in Google Chrome even after uninstalling Kaspersky. Interestingly enough Edge and Brave browsers did not seem affected. ( if anyone is interested to know more you can read about fingerprints here https://www.grc.com/fingerprints.htm ) Getting aggravated with trying to find a solution to fix my Chrome validating certs properly I just decided to wipe the machine again and do over without Kaspersky only to encounter yet another struggle session with Win10 forcibly replacing my installed AMD driver with a different one. I got lucky this time and there was a newer certified AMD graphics driver that Windows 10 seems to like and it stopped trying to override me.

On the bright side for the moment my PC is now a bit (356 pts) faster at stock than it was before now that all the junk has been cleaned out.

Now I'm stuck contemplating if I want to reinstall iCue because I discovered HWiNFO64 recognized Commander Pro's fan and temp sensors without iCue being installed.

I tried to see if I could skirt around the issue by installing it in a VM and reattaching the Commander Pro and Lighting Node Pro USB's to the VM but it was a failed experiment. On the first try Command Pro connected but the Lighting Node Pro wouldn't connect and be recognized. I ended up loosing the hardware preset for my fans RGB in the process but the lighting strips hardware profile survived. After that the Command Pro wouldn't reattach to the VM at all. iCue v3 was already a whopping 485MB of bloatware with v4 more than double that. I had a sweet Christmas color theme going but I also want to be able to control my fans so I guess I have no choice so it comes down to v3 w/ Equalizer APO to fix my headset mic boost or v4 and not sure what issues I might have.

------------------------------------------------------------------------------

Sorry took a bit of a detour, back to the memory topic...
Updated status...
- Waiting for an Asrock tech support reply (Nemix is not on the QVL vendor list at all, so I don't have any high expectations with this going anywhere but you never know.)
- Will try disable cstates
- Will try 1.8V PLL to 2.0v max

 

[Dr. Dro]

Not specifically a 128 GB kit, but I have a 4x 16 GB dual-rank, dual-sided kit that takes all available memory banks on the platform. My 5950X runs it happily, and with low vsoc and 1.375V memory voltage. Timings, however, are not exactly the tightest, I would say just about standard - but it's not terrible like 18-22-22-22 on 2T either. It's a weird B-die downbin Dominator kit.

I have used the sticks with these settings on both X370 (C6H) and this B550-E with both 3500X, 3900XT and 5950X CPUs, rock stable.

QVL is nonsense, don't limit yourself to these models, especially down the road as the platform ages, as QVL is not updated every time new memory kits release. Just make sure your AGESA is updated to get the best memory behavior. If anything, motherboard makers should just drop QVLs entirely IMHO.

 

[Sf!]

Hello
Sorry for my english, I'm french

I search to understand why my ownbuild server (5950x, 4*32GB patriot 3600c18, on a X570-A pro) not work when activating xmp.
Server reboot 3 times, and disable XMP. Ram runs at 2667Mhz so...

As I found, I maybe understand there is a system of ranks, that plays on the maximum frequency.

How I can use a better frequency on this config ?

Last time I touch to overclock, it was more simple... 3500+ and q6600 : only fsb and ratio 15 years ago.

Thank in advance if you can help or redirect me to an ultra assited tutorial :s

 

[Toothless]

Hello
Sorry for my english, I'm french

I search to understand why my ownbuild server (5950x, 4*32GB patriot 3600c18, on a X570-A pro) not work when activating xmp.
Server reboot 3 times, and disable XMP. Ram runs at 2667Mhz so...

As I found, I maybe understand there is a system of ranks, that plays on the maximum frequency.

How I can use a better frequency on this config ?

Last time I touch to overclock, it was more simple... 3500+ and q6600 : only fsb and ratio 15 years ago.

Thank in advance if you can help or redirect me to an ultra assited tutorial :s

You're never going to get 3600 with 128GB installed. Be happy with 2666 and leave it be.

 

[Sf!]

You're never going to get 3600 with 128GB installed. Be happy with 2666 and leave it be.

Thank for this fast answer

The problem is I have performance issues... cpu seems saturated by pics sometimes, resulting in rollback in game for players.
I think cpu is enough strong to supports, but think the ram is too long.
It's about 14 ark servers.

Here, I saw working 3200mhz with 128gb :

What did you have to do to get 128GB ram working on Zen3 CPU?

I'm at a point where I am not getting errors anymore (well the errors seem to occur only right before the reboot). Just reboots are occurring after leaving the machine idle for a few hours. I can do 2hr OCCT runs, 3hr zip jobs, 3hr handbrake jobs no problem. Even after thermal cycling the ram...

www.techpowerup.com

Ok if I can't reach 3600, but 2900 or 3000 must be possible with stability no ?

 

[Dr. Dro]

Thank for this fast answer

The problem is I have performance issues... cpu seems saturated by pics sometimes, resulting in rollback in game for players.
I think cpu is enough strong to supports, but think the ram is too long.
It's about 14 ark servers.

Here, I saw working 3200mhz with 128gb :

What did you have to do to get 128GB ram working on Zen3 CPU?

I'm at a point where I am not getting errors anymore (well the errors seem to occur only right before the reboot). Just reboots are occurring after leaving the machine idle for a few hours. I can do 2hr OCCT runs, 3hr zip jobs, 3hr handbrake jobs no problem. Even after thermal cycling the ram...

www.techpowerup.com

Ok if I can't reach 3600, but 2900 or 3000 must be possible with stability no ?

This will depend on your motherboard and BIOS update level. Higher end motherboards with B550 (exclusively the premium segment B550s such as the ROG Strix-E, Strix-XE or AsRock Taichi) or X570 chipsets will have it the easiest. Low end X570s will also prove uncooperative, for that much memory you want a very clean, perfect signal to your CPU and a friendly topology (such as daisy-chain).

First ensure that you have the most current AGESA by updating your motherboard BIOS, and then try to tweak your memory. Start with 3000, and don't be afraid to give it some voltage - start with 1.2 VSoC, 1500 FCLK, 3000 DDR, 1.45V VDIMM, and then start raising the frequency on FCLK and DDR (with FCLK always at half DDR) until you can no longer boot. Once you hit that, back off by one value (if it didn't boot 3600, then try 3466, for example), lower your voltages to more acceptable, optimized values and start working on your timings, it's an iterative process that takes some time and patience to achieve, but it is well worth it.

Don't rely on things like the Ryzen DRAM calculator, they aren't going to be very helpful.

 

[ir_cow]

You're never going to get 3600 with 128GB installed. Be happy with 2666 and leave it be.

You can if the FLCK/Mem is Gear 2, but that kinda defeats the purpose and 2666 will be faster at that point.

 

[Sf!]

Cela dépendra de votre carte mère et du niveau de mise à jour du BIOS. Les cartes mères haut de gamme avec B550 (exclusivement les B550 du segment premium tels que les ROG Strix-E, Strix-XE ou AsRock Taichi) ou les chipsets X570 seront les plus faciles. Les X570 bas de gamme se présentent également peu coopératifs, car pour autant de mémoire, vous voulez un signal très propre et parfait vers votre CPU et une topologie conviviale (telle que la connexion en guirlande).

Vérifiez-vous d'abord que vous décrivez de l'AGESA le plus récent en mettant à jour le BIOS de votre carte mère, puis essayez de modifier votre mémoire. Commencer avec 3000, et n'ayez pas peur de lui donner une certaine tension - connaître avec 1,2 VSoC, 1500 FCLK, 3000 DDR, 1,45 V VDIMM, puis devrait augmenter la fréquence sur FCLK et DDR (avec FCLK toujours à moitié DDR) jusqu'à ce que vous ne conduisiez plus démarrer. Une fois que vous avez atteint cela, reculez d'une valeur (s'il n'a pas activé 3600, essayez 3466, par exemple), abaissez vos tensions à des valeurs plus acceptables et optimisées et évitez de travailler sur vos timings, c 'est un processus itératif qui prend un certain temps du temps et de la patience pour y parvenir, mais cela en vaut la peine.

Ne comptez pas sur des choses comme la calculatrice Ryzen DRAM, elles ne seront pas très utiles.

Thank for answer

Vsoc ?
It's a new name for Vcpu ? ^^

Why need change Vcpu if only adjust ram ?

If the ram is sell at 3600 cl18 in XMP, why must give more of the 1,,35 of the xmp?

 

[Dr. Dro]

Not CPU voltage, SOC voltage, each motherboard has its own name for this