System won't POST with four sticks of RAM

[philo-sofa]

Hi all, my new build has started refusing to work using four DIMMs. After working perfectly for ~2 weeks *boom* come Friday she simply wouldn’t POST. The Debug LED initially showed a CPU error, after clearing CMOS this changed to a memory error. Removing any one or two sticks of RAM allows the system to post, boot into Windows and pass an hour of MemTest64

I’ve also tried:

1. Downclocking the memory to base timings and frequency (133 C20-21-21-21-42 1T / 1.35v / 1067 flck / gear down enabled & disabled) to test; this has no effect
2. Increasing NB (or on-die NB) voltage by 80mv; this also has no effect
3. Swapping out the memory for four entirely new modules; this changed nothing

I suspect that it’s the motherboard at this point (or I suppose the CPU) but would welcome any other theories or suggested tests. Full specs are in my sig but here’s an expanded list of the most relevant hardware:


• 4x8GB Patriot Viper Steel DDR4-4400 C19 (two dual channel sets) @ 3800 C16-16-16-16-32 1T / 1.35v / 1900 flck / geardown enabled
(tested at 2133 C20-21-21-21-42 1T / 1.35v / 1067 flck / gear down enabled & disabled)

• 5800X with PBO auto
(PBO off for testing)

• MSI MAG X-570 Tomahawk Wi-Fi
(7C84v14 BIOS – haven’t updated to beta BIOS as it’s beta, plus everything used to work perfectly on this one)

 

[philo-sofa]

@philo-sofa

Run memtest on the sticks. Try with all 4 at first and if you get any errors then try them 1 at a time in slot A2 only. Run them at JEDEC speeds 2133mhz or 2666mhz basically defaults with xmp off. If none fail, run them one at a time at the XMP profile on the stcks. You may find one fails then.

Make a bootable USB of memtest and remember if you have CSM enabled (in the BIOS) and no UEFI windows install to set UEFI to enabled so the usb stick can boot as it is a UEFI boot only.

https://www.memtest86.com/

Thanks purgatory, but do you think I need to do this given I've swapped out my original memory for four completely new DIMMS and gotten the same problem?

Also is Memtest86 much more useful than 64 IYO?

 

[philo-sofa]

If you are trying to get 3800mhz on 4 sticks and having issues then it is either BIOS or simple the cpu and motherboard will not allow it.

Have you tried to run them at 3200mhz ? Just turn on the xmp profile and drop the speed only to 3200mhz and if the sytem works fine it means you need to slowly work up and find where the system becomes unstable. Memtest I use to check if the memory is stable at any set speed, not just to test if the memory is working.

Appreciate the help dude, really, but I did bullet point all this in my OP. To reiterate:

I’ve also tried:

1. Downclocking the memory to base timings and frequency to test (tested at 2133 C20-21-21-21-42 1T / 1.35v / 1067 flck / gear down enabled & disabled) - this has no effect
2. Increasing NB (or on-die NB) voltage by 80mv - this also has no effect
3. Swapping out the memory for four entirely different modules - this changed nothing

I would go threw setting the xmp off and running memtest with all 4 sticks first, to rule out any bad sticks. If it fails any part of memtest with 4 sticks at JEDEC speeds (XMP off) then I would start testing them 1 by one.

I can't even POST in with four sticks man

This year I have had a shocking amount of bad sticks from new, this is why I test them minute I get them now, before never had these issues with DOA sticks or they produce errors even at JEDEC speeds

Rule out the sticks first or it will drive you crazy, work in order and test. If the sticks pass JEDEC all 4 in the slots then try XMP and see if they pass. If they fail on xmp could be that the speed is too high for 4 sticks for your setup (CPU or Motherboard don't like to run 4 sticks at xmp), so you will have to reduce the speed of the sticks in 4 way. Remember 4 sticks are harder to run so don't expect XMP to work. Also if you check your motherboards RAM QVL you will see for a lot of ram it will not allow certain speeds with 4 sticks.

Yep tried them at base speeds and cannot do four sticks. I feel I've ruled out the memory fairly conclusively by purchasing four new DIMMs and finding the exact same problem. Also relevant is that everything worked fine for two weeks and then failed.

But test the sticks and rule them out then start tuning the speed and timings for 4 sticks. No point doing any speed or timings unless you make sure the RAM is working 100%.

I did several hours in Memtest64 with three sticks.

 

[philo-sofa]

If you're getting no joy with new ram, check if there's a new bios for it, or if you can roll back to a non-beta one. Hope it's just a bios issue rather than a dead mobo

That's not a bad idea, could be a borky BIOS even if it did work previously. Might be the next step.

He clearly didn't read any of your post.

Take the bios battery out for half an hour and got the power switch a few times, completely clear it.

Double check RAM is FIRMLY seated.

Indeed...

Done and done. After 30 mins convincing the mobo to read an NVME drive I'm back. No dice but cheers for the good suggestion.

I did read it, that is why i kept repeating test each stick of ram on its own first, so to rule out the sticks then the slots. Troubleshooting in a logical way. Then we can move on to next stages.

Really? You didn't seem to pick up that I tried the memory at base speeds, or that I had four new modules. We've fairly much ruled out a bad DIMM (trying four new ones) or that there's a bad slot (any arrangement of 2-3 DIMMS works). Again though I do appreciate the help man.


All of which brings up the question (if a BIOS update doesn't fix the board) of is it likely the board or CPU?

 

[philo-sofa]

Update:

Due to a delivery snafu I ended up with a third set of basically-the-same memory (Patriot Viper 4GHz, as opposed to the original 4133 and the second batch which were 4400) and this set works on all four DIMMS?! It's a weird one; given the four original sticks used to work, I'm still a bit worried about long term viability and what might be causing those other two sets not to work. On the positive side they do run up to 3733 CL6 and 1866 fclk on all four slots without issue.

Any thoughts on this folks?

Clutching at straws here, are all the power connectors firmly in, 24pin and the CPU power cables?

Try reseat CPU.

It's worth just checking......

It'll be hard to diagnose between the two CPU/Motherboard without swapping each sorry. Unless someone else ok here can think of something else.

You know I'm gonna try reseating everything and reinstalling the original RAM over the weekend, that can't hurt.

What voltages were you running before it failed to post?

Didn't touch anything voltage wise. It was 1.357 on the DIMMs, 1.0s_something on the NB voltage.

I'd go with the beta bios as you can always swap back to the one your on now so you really have nothing to lose.

Agreed if I don't get anywhere with the other tests, I'll give this a whirl.


Thanks for your help so far @Purgatory @BUDFORCE @Hakanese @Doug2507 @Joxeon

 

[philo-sofa]

The longer I stay on the forum the more convinced that I know nothing...
I get the feeling that with the amd chips they're a Royal pita sometimes.
But last thing I would suggest is up the voltages and see if you can get it to work.
Treat it as an OC. Good luck, and at least you got an extra set of high speed ram out of it

Yeah me too mate. This is my first new build in half a decade and my first AMD build since Athlon 64.

I agree that it's evident that Zen 3 is super difficult and borky when it comes to memory. My main concern is that I've got something else going on, given I was running happily on the original four DIMMs and then suddenly they failed to POST at the slowest speeds and loosest timings. But some of that is me being over-protective of my rig

 

[philo-sofa]

Yeah I think there's something to that @Purgatory; my 'sense' of it is that the RAM itself is okay given any 2-3 sticks run at high speeds but there's likely that weakness in AMDs 5000 series memory management rearing its head, possibly combined with some weakness in two of the memory modules. But that's just a sense, I could be wrong.

Either way for now, I've got 32GB working and I don't have to replace my motherboard Plus AGESA 1.2 is incoming (thread) and I've got a 5900X OCUK say will be with me next month, so I'll get to narrow down where the bug lies a bit further.

 

[philo-sofa]

Hang on, you were trying to use 4400mhz RAM downclocked significantly, then 4133, then 4000, the 4000 worked, clocked down to 3733..

Originally I was using 4133, downclocked to 3800. I never used the XMP profiles so set it to 3600MHz and 1.35v at first, then 'overclocked' it to 3800. It all worked until it failed to POST one day, which is a key piece of evidence.

Then I tried a 4400MHz set and it wouldn't post, and finally four 4000MHz DIMMs.

Seems you are putting in RAM of a much higher frequency than the CPU can handle at stock speed, and/or that you intend to run the ram at, then the lowest frequency kit worked?

I am guessing here (stress that) but isnt that a factor here?

Maybe something in the bios was defaulting/clocking the ram at the original rated speed and it was soiling itself, or maybe the RAM itself doesnt like being run at such lower than rated speeds? Enough to work with 2 but not 4?

Just wierd how the RAM with the lowest stock fequency and that closely matches what you are actually running it at worked.

I'd guess not, but it's not inconceivable? It's 32GB of Samsung B-die spread over four DIMMs for each set. If you don't enable XMP it doesn't know anything about itself; it doesn't now it's 4400MHz memory running at 3800MHz. I also repeatedly tried running it all at 2133 and no dice whatsoever and surely that should work?

Zen 2/3 are notorious for being fussy with ram, although most people put lower stock ram and try and overclock it, you are doing the opposite, maybe it doesnt like it either way.

There's space in memory sub-timings and stuff we don't typically see that might cause that. But also there's the CPU factor you mention; so much going on there and all we can be sure about is that the result is a system that's fussy with RAM.

 

[philo-sofa]

Probably a coincidence @fastboy, but I've been using patriot vipers too.

They should still run at 2133 with everything on auto

Yeah 2133 should certainly allow us to POST.

 

[philo-sofa]

well damn,waste of money paying extra for fast ram then lol,should have just grabbed some cheap 3600.

tbh wish i had stuck to intel.

With the new firmware from AMD (AGESA 1.2) you’ll very likely be able to hit 4GHz. It’s comes out end of jan to start of fab. So our 4GHz memory is pretty much perfect.