Random hard reboots with Corsair DDR4 and MSI MPG Z690 Edge WiFi DDR4

[PTJim]

I'm trying to re-use 64GB of CMK32GX4M2D3600C18 (4x16GB), moving from an i9-9900K on a Gigabyte mobo to a new i9-12900K on the MSI Z690 Edge WiFi DDR4, running Win10 Pro.  Memory has been working perfectly for a couple years, and the new MSI mobo is on Corsair's compatibility list for this DRAM part number.

Problem is that it's not stable; I'm getting random and instantaneous black screen and powered-off PC for a few seconds, followed by an auto-reboot.  Looks like a memory timing problem, but my knowledge is a bit limited, although I'm pretty good at trying suggestions.  BIOS is current Oct '22 1.91 and nothing is overclocked.  Tried manually entering the 18-22-22-42 timings and even just switched to Gear2 hoping I could run slower but stable until I get it fixed.  None of this, or earlier BIOS versions, have helped.  Have run MemTest86 for two hours with zero errors reported.

Have also tried rolling back the BIOS to 3 or 4 earlier versions; no joy.  I make my living with my PC, so this is starting to cost me money when the system randomly dies without warning in the middle of the day, over and over.  

I've run out of things I can think of to try next - suggestions would be most welcome!  I'll try attaching a few screenshots of my BIOS setup from earlier today.

THANK YOU for any help or suggestions!

 

Edited November 2, 2022 by PTJim
Added OS version to text

[Corsair Notepad]

24 minutes ago, PTJim said:

I'm trying to re-use 64GB of CMK32GX4M2D3600C18 (4x16GB)...

That is most likely your issue at this point. The part number you have provided is a 32GB(2x16GB) memory kit, and you have 64GB of system memory installed meaning you are mixing memory kits. Mixing memory kits such as you are doing, can lead to compatibility issues such as preventing XMP from working correctly.

 

[PTJim]

22 minutes ago, Corsair Notepad said:

That is most likely your issue at this point. The part number you have provided is a 32GB(2x16GB) memory kit, and you have 64GB of system memory installed meaning you are mixing memory kits. Mixing memory kits such as you are doing, can lead to compatibility issues such as preventing XMP from working correctly.

 

That's true; I bought two of those kits at the same time from Newegg (2x16GB per box), so unless manufacturing specifications changed between the two boxes, I'm not really "mixing" memory, am I?  I was very careful when doing my previous build to use identical memory sticks - I can only imagine the possible problems when mixing DRAM types. 

They've been running together fine at 3600 for quite some time with the 9900K yet won't run stably at all with the new mobo/CPU.  Sometimes they'll go for 8-10 hours, such as last night's regular backup (I closed all other apps first, just in case) but other times I can't get it to stay alive for 30 minutes without a reboot, when trying to run my usual email/browser/trading apps.   In fact, my system crashed just after I got a "Topic successfully posted" from this forum a little while ago.  It seems related to sudden changes in desktop activities, which changes memory demands, leading me to think it's got to be a timing issue.  Hoping it's not a motherboard controller IC problem; don't need more headaches with this upgrade build.

Surely there must be something I can tweak in the BIOS to get these four identical sticks to be happy, right?

And I hate to be pedantic, but that part number IS on Corsair's compatibility list for this mobo, with nothing mentioned about "sorry, you can only use two sticks" 😉.  I should have been clear about all four sticks being the same part number, purchased at the same time for the previous build.  It was a while ago, but I don't think a 64GB 4-stick box was available at that time (I would have bought it), although I see them now.

Edited November 2, 2022 by PTJim
Added photo of memory stick

[Corsair Notepad]

If you do some searching here on the forums and on our support site you will see that mixing memory kits such as you are doing is something that we have always advised against, and stated is not guaranteed by the warranty. I have even provided a link below to our public facing help article on our support site.

https://help.corsair.com/hc/en-us/articles/8670306985613-Can-I-mix-CORSAIR-memory-kits-

[PTJim]

If that article is correct, then how in the world would anybody EVER be able to upgrade from 32GB to 64GB by populating the two empty memory slots with more of the same product?  Are you asserting that unless the sticks are IN THE SAME BOX TOGETHER, they won't necessarily work together, even if they came off the production line in sequence and were bought together?  Who designs things that way?  I know something about designing for mass manufacturing and tolerances designed specifically to allow things to work together, and that doesn't sound like anything I've ever worked with.  Imagine a car company installing identically-spec'd pistons from two different boxes shipped in by  the piston supplier, into an engine without being sure they'll all work at the same time?

That doesn't seem completely correct, so I'll stand by to see if other users might chime in to either confirm that or offer a workaround.

Thanks for getting back to me promptly, by the way; that's very much appreciated.

[Technobeard]

The two modules in one kit have been tested together and therefore can be guaranteed to work together. Two separate kits have not been tested with each other and therefore cannot be guaranteed to work together. It's not that they won't work together, it's just that no one knows because it hasn't been tested. Might work, might now.

A car's piston is different from a memory IC. CPU and memory binning come into play when it comes to mixing kits, not to mention electrical engineering and the billions of transistors measured in nanometers. I'm not going to pretend I understand any of that, other than to say mixing memory has never been a simple thing even from before the SDRAM days (20+ years).

 

Since you're using a different CPU and motherboard, there's other new factors to consider outside of your memory. The memory controller may be more sensitive to the differences between the kits than the old one. The slots could have an issue.

 

Going back to basics for a second:

• What's the memory version of each kit?
• Have you tried one stick at a time in the same slot (or at least one kit at a time) at their rated speeds?

[c-attack]

The three interacting components here are the individual RAM modules, the motherboard, and the CPU internal memory controller. You changed two of the three factors, so a different outcome was possible. I can see why you would be caught off guard and we tend to think of later products as an improvement on the prior, but there have been some unexpected behaviors reported on the DDR4 versions of the Z690s. It’s not clear if MB makers are moving over Z590 style implementation or if they are forced to create a unique set up to work with the Z690 chipset meant for DDR5. 
 

 

Instant black screen and reset makes me wonder if you have a VCCIO or SA (system agent) voltage issue.  If the timings were bad, the board would be struggling to work through them on the boot sequence  unless you put in impossible values and that’s not the case here.  Those values are set by the board and not included in XMP presets.  With larger RAM volumes or mixed kits this can be a tricky spot and MB auto values may not suffice.  Particularly with VCCIO more voltage does not mean “more stable but hotter”.  Too much can cause a black screen or freeze.

 

I would try the suggestion above and work each kit individually.  That would also make the IO and SA requirements lower and if successful provide a starting point for tuning those values. 

 

[PTJim]

Thanks for the feedback, info and ideas guys.  Before I came back here and read them, I was thinking about the need to match them up together, and it occurred to me that maybe the two kits need to be installed as matched pairs into the memory-bank slots.  Since I knew they were spec'd identically, but wasn't aware they're a matched set in the box, when I moved the memory over from my previous install I didn't pay attention to which stick was in which slot.  So I just finished cracking it open again - had to remove a fan from the Noctua D15 to get to the slots.

Sure enough, they weren't slotted with one kit in slots 1/3 and the other in slots 2/4; they were jumbled up.  So I took before/after photos of them showing serial numbers and paired them up.  This is being written just after the reboot with this configuration, so I have my fingers crossed that maybe that slight mismatch might be the root of this problem.  

I'll continue running / testing the system and see if the reboots will cease now - and I'll report back to this thread with results.

Thanks again, guys!

P.S.  I previously adjusted SA voltage from 0.91 to 1.00 to no effect, so had set it back.  Haven't messed with VCCIO  I'll keep those adjustments in mind as well, as needed.  And hey - it stayed  up long enough to finish typing this reply ..... so far, so good!

P.P.S. I've been messing with PC hard/software since paper-tape and punchcard days and have been building/fixing since the Osborne days, including buying individual memory ICs in tubes and manually shoving them into sockets for CP/M and XT-clone builds a hundred years ago.  Yeah, I'm old.  So I understand what y'all are saying; just never ran across this particular little compatibility landmine.  Now all I need is to remember what I've just learned; grateful for your chiming in.

Edited November 3, 2022 by PTJim

[PTJim]

Sigh, no joy.  System rebooted about 10 minutes after posting the above, with the 4 sticks paired up into the memory banks.

Yesterday, just in case, I ordered this kit of 2x32GB sticks from Amazon, which is on MSI's compatibility list, and will swap them into slots 2 and 4 (counting from the CPU; don't remember the official designations) when they arrive tomorrow:  CMK64GX4M2D3600C18

Hoping just two "approved" sticks will work reliably.  If not, I either need to tweak deeper or consider parts replacement, I guess.

Again, thanks, and I'll report back to this thread with updates in case somebody finds it useful.

 

Edited November 3, 2022 by PTJim

[Technobeard]

Before you try a new kit, try just ONE existing matching kit. See if that works.

[PTJim]

1 hour ago, Technobeard said:

Before you try a new kit, try just ONE existing matching kit. See if that works.

Ok, sure; will do - have to wait for delivery of new sticks anyway.  Thanks for the prodding!  Off to yank one pair out of slots 1 and 3.

Would you please outline what my next steps might be, assuming one pair of 2x16 works (which I assume it will)?  Would that indicate that I can get all four sticks working with appropriate BIOS tweaking?  That would be the best outcome, and I have some time to return the new sticks if I don't need them, but time is literally money in this case, so I'd want to have a final resolution one way or the other by Sunday night if possible. 

Have messed with this all week so far and the lost opportunity cost is already several times the price of the newly-ordered sticks...... now and then throwing money at a problem is actually the best solution, but on the other hand the (mechanical) engineer in me would love to get this figured out.

Cheers.

Edited November 3, 2022 by PTJim

[PTJim]

For those who might be wondering, no a total 32GB setup isn't a good idea for me - I regularly see 22-28GB of memory usage in my work flow and I don't like to push that close to the limit, so if one pair works but two pair just won't, then I'll have to upgrade the one-pair specs to 2x32GB.

[Technobeard]

Test one kit in 2/4 from the CPU like you're doing. If it's stable at the rated settings, swap with the other kit. If they're stable, you know that the 2/4 slots are solid as is each individual kit. From that point you know you should be good with the new 64 GB kit you have coming in. That's barring any odd issues with the increased memory density, but your motherboard and CPU support that. The business side of you should now be happy 😃

 

For the engineer side of you - What is the memory version of each 32 GB memory kit? It's on the bottom right of the label (see your pic above). One kit is v5.49. What's the version of the other kit?

[PTJim]

All four sticks are v5.49, see attached pic in the order most-recently installed.   S/N's xxx330 and xxx331 are in slots 2/4 now.  So far, so good; not sure how long it might take to evaluate if they're working stably because I haven't been able to figure out a way to make them fail - it's just a random thing taking a few minutes to several hours (and random number of apps/windows open, too) to trigger. 

There's no thermal issues at all BTW; CPU running in mid-20s mostly and my GPUs fans don't even come on at all unless I have a ton of charts or a VR game running - which isn't happening given the circumstances.  The Noctua blower in the Torrent case is an amazing beast for those of us who don't like water looping around electronics but still appreciate a quiet system.  More engineer-brain always worrying about the worst possible outcome.....

Since I've got a bit of time, I may try both pairs each one at a time in both pairs of slots, just to prove to myself the hardware is more-or-less ok.  My ME brain fails me again, though; I know slots 2/4 are the primary for using 2 sticks, but is it a problem just running in slots 1/3 or is that a no-go with current hardware designs?

So I'll let these two sticks run at least overnight or until they ........ glitch 😬.  Fingers crossed once more.  If things are ok tomorrow noonish, I may as well swap them for the other pair to see how they fare.  As mentioned way above, though, all four sticks have been 100% stable since installing in my i9-9900K platform 2.5 years ago, so I assume all are fine.  If there's an unfixable problem running the four of them on this mobo, I'm assuming the new 2x32 sticks can handle that.  I've probably got enough spare parts to just use the four older sticks to rebuild the 9900K to functioning status as a complete spare PC to satisfy my belt-and-suspenders backup approach to staying functional.  Might have to buy a boot MVNE or SSD at most, I think.

Thanks again.

Edited November 3, 2022 by PTJim

[PTJim]

It ran fine last night for a few hours, put it into sleep mode.  Ran about 45 mins this morning, then failed again with another shutdown/reboot with just those two sticks in place.

Since the new 2x32 sticks haven't arrived, I've swapped in the other pair of 2x16 into slots 2/4 to see if the problem might be a specific memory stick or if these just won't work in this mobo.  Really hoping the mobo itself isn't the problem - that swapout is a much bigger deal and several days delay.

Will continue to update here.

[PTJim]

The second pair of sticks just failed/rebooted after about 40 minutes.    So I'm left with trying the two new 2x32 sticks when they arrive later on today, I suppose.

Anybody know if it's worth inserting these older ones into slots 1/3 instead of 2/4, or does that not work for a 2-stick configuration?  Thinking about trying to rule in/out a mobo problem, perhaps.  

I'm getting pretty adept and fast at pulling the GPU and one Noctua fan to swap these things...... 

Edited November 4, 2022 by PTJim

[Technobeard]

Glancing at the manual yesterday I seem to recall 2/4 is the way to go. Try one stick at a time in the same slot (check the motherboard manual but from memory I think it should be in slot 2). May want to double check the BIOS settings, and then run memtest one stick at a time in that slot. Rare for one stick let alone 4 to have issues though, so it'll either be a slot issue, setting issue, or memory controller issue.

 

I'd suggest creating a support ticket at https://help.corsair.com (and feel free to post a link to this forum post).

[PTJim]

1 hour ago, Technobeard said:

Glancing at the manual yesterday I seem to recall 2/4 is the way to go. Try one stick at a time in the same slot (check the motherboard manual but from memory I think it should be in slot 2). May want to double check the BIOS settings, and then run memtest one stick at a time in that slot. Rare for one stick let alone 4 to have issues though, so it'll either be a slot issue, setting issue, or memory controller issue.

 

I'd suggest creating a support ticket at https://help.corsair.com (and feel free to post a link to this forum post).

Yes, you're correct the manual specifies 2/4 for one pair and slot 2 for a single.  This specific memory model number is not on the MSI compatibility list, so I think either the list isn't comprehensive or this mobo just doesn't like this particular configuration.  The 2x32 sticks on order are on the list (with a v5.49 note to the part number), so I hope I get that version and will give them a try before spending time with single sticks and alternate slots, for now.  If the new sticks don't work (even though "on the list") I may take that approach to try to diagnose if it's a board failure.  The fact that it can run normally for several hours tells me it's probably a timing glitch with this specific DRAM model number, but that's just a hunch.

And I did open a ticket a few hours before starting this thread earlier in the week - haven't heard anything back yet so this has been far more beneficial.  Will continue to update and continue to convey my gratitude for the assist.

[PTJim]

Ok, new 2x32GB sticks arrived and are installed.  BIOS was happy with them upon boot and everything started up normally, so I'm going to leave it running with my usual batch of apps/windows open and will see how stable it is - will probably let it sit overnight and run the bootable-USB MemTest86 tomorrow to give it a bit of a workout.  If that goes well, then other stress/benchmark tests and maybe push it a bit in the BIOS to make sure it's stable.

As always, fingers crossed................

[PTJim]

Crap, crashed again, with the newest DRAM, after about 25 minutes.  All my usual apps/windows seemed just fine, then I tried to add TestMem5 to the mix - it crashed/rebooted right after double-clicking, even before that app loaded.

So now I'm in the soup, not sure where to go from here.  Bad mobo?  Bad CPU?  Both were the core of this upgrade last weekend; I'm just not sure how to debug this.   Since the change was from LGA1151 to LGA1700 I don't have a different mobo or CPU to swap in for testing like the YouTubers have shelves and shelves of spare parts for their fiddling around.

Is it possible my MSI mobo just doesn't like Corsair memory?  Should I return the new sticks and try a different brand, or do a mobo or CPU swap first?  One problem is that I got a good deal on both, the prices have moved back up about $200 total at the moment, and Amazon usually just does refunds, not replacements - that means I'd have to be absolutely sure I've got a dead part, because a return means I pay a significantly higher price on a replacement, which would be pretty annoying if the part turned out not to be the problem.  Trying a different DRAM brand would be simpler but may not be the problem at all.

Now I'm rambling; time to break away from this for a while to enjoy a Friday evening...... cheers, all.

Edited November 4, 2022 by PTJim

[c-attack]

I would start on the motherboard end.  See if you can find an ownership or discussion group at one of the various independent sites.  That is usually the best place to find comparative data specific to your board, trouble with brand X, success with brand Y, etc.  A bad CPU is fairly rare.  You are having issues with multiple RAM kits.  The board seems like the first suspect and also where we tend to see more variability.  

[PTJim]

Last night (at the very end of my 30-day return period) I arranged with Amazon to return my mobo and replace it with same model, which arrives Monday.  After I shut this down, I plan to rebuild my i9-9900K older system in this case and rebuild with the replacement mobo early next week.  With all the swapping/testing I've done with two sets of DDR4, it should have worked by now in some configuration.

I'll come back to update if anybody's curious how it's going.

[PTJim]

Well, I'm back.  Replacement mobo arrived and I've been on the hunt for compatible DDR4 that will fit under my Noctua cooler - man, this isn't easy!  Most everything is NOT on the MSI QVL, but I finally realized I should try PC Parts Picker to help narrow things down.  Also, low-profile sticks seem to be getting more rare as everybody moves to unicorn-barf RGB sticks that stick up much higher than the Corsair Vengeance, which I've loved for years in my 9900K build.

So I have on order a pair of 2x32 3600 C18 from TeamGroup, having found nothing from Corsair that makes the MSI list  (unless the eBay vendor who promised to check the version number has the v5.49 available) to have as a backup:  TLZGD464G3600HC18JDC01 - which is on the QVL and at a very low price on Amazon at the moment.

I plan to use the Corsair sticks because I like the brand, but so far have been out of luck trying to find v5.49, but maybe I'll get lucky.  If not, supposedly the TeamGroup sticks have passed the compatibility test.

Whew.  For the sake of completeness for anybody still sticking with this boring thread, I'll update again, hopefully for the final/solved time.

Thanks for the assist, gang.

Edited November 10, 2022 by PTJim

[PTJim]

Grrrrrr!  Replacement mobo (same model) and new TeamGroup DDR4 (on MSI's compatibility list) and it's STILL rebooting!  Was ok overnight but this morning just dumped out in the middle of email.  The only thing I haven's swapped out is the 12900K CPU, but all tests I've run, including stress tests, have passed fine - could it still be a suspect part?

Is there a fundamental problem with Z690 chipsets, MSI motherboards - getting desperate and very tired of rebuilding and reverting my system..... thanks.

[PTJim]

Just noticed something new: my MSI mobo sets the DRAM Voltage to 1.19V in both stock and OC modes and I discovered that TeamGroup specifies 1.35V - could that be the issue all along?  Have had three reboot/resets this morning so far.

I've set the mobo to Profile 1 overclock and manually changed DRAM Voltage to 1.35 - system is running (as you can see) and I'll see if it hangs in there, but was hoping for some thoughts on this voltage issue.  

Gracias.