prolly under advanced dram, maybe command rate (1t/2t/gear down)…maybe?
try setting 1t or 2t explicitly
1 Like
Awesome, found it! Thanks!
2t is slower, 1t might be safe though. if 1t doesnt work try 2t
Don’t want to jinx anything, but I’ve installed kernel 5.10.3 for the Fedora testing day and it’s running smooth and without any resets so far for 3 days now. Same settings as before, SMT on, no voltage tweaks, memory with XMP profile at 3600 mHz… just saying, no conclusion just yet.
1 Like
Hello,
I have the same problem.
CPU - 3700x, MB - MSI B550 Tomahawk.
I tried 3 memory kits: 16 gb GSkill Flarex 3200 cl14: 32 Gb GSkill TridentZ 3200 cl14: 32 Gb Crucial 3200 cl16.
I had the same mce error with Fedora 32 and kernel 5.10.3 .
Now I try @wendell advice - memory with 1T explicity and
@Riotvan advice - ddg and vddp voltages with 950/900.
I use memory only with XMP settings.
I have this mce error from september 2020. I am sad to say that but if I cannot find a solution I will jump in Intel boat (again).
Just crashed on Linux with 5.10.2 kernel and memory set to 1T. I’ll try 2T and see if that makes a difference. It seemed to last longer with 1T enabled, but that may have just been a coincidence.
1 Like
Your memory (C16D-32GVK) is not 4-DIMM compatible. I’ve went over the QVL list (Last updated 2020-12-09) compatbility. With 3000 series, ONLY in two dimm operational mode is supported. Sorry, I would stick with dual dimms (16GB) or try and get 4 dimm compatible memory if you can. Best of luck.
@thorn
Hey, if you can provide some specific module info to help troubleshoot. (I’ve had my fair share of RAM issue on both platforms regardless). Both my Ryzen systems (Zen+ and Zen3) run memory settings WITHOUT XMP enabled and manual timings per XMP specific to the modules. Welcome to the forums 
Seems like those are dual-ranked modules (16x 8Gb chips each).
Ryzen memory controller supports 4 ranks per channel.
So it should work with 2 DR sticks in one channel and 2 in the other at least with JEDEC standard.
Hello,
Thanks @Whipple_Jaeger.
Now I use F4-3200C14D-32GTZ. FlareX was F4-3200C14D-16GFX.
I would like to run memory settings without XMP enabled but I don’t know how to proceed. I really appreciate some help. I only want to run it with 3200cl14.
I swaped my RX 550 with a GT 1030. Maybe is GPU to blame.
I only have two dimms installed, 16gb each. What is the QVL list and where can I find it?
Heh, should and will/does are miles apart. Whilst the SOC can do the memory ranking… the topology is motherboard specific and MSI has mounds of dimms modules to certify and saw fit to qualify as THEY read NOT AMD is capable of supporting. Stick to their warranted guidelines is my recommendation.
@thorn
Your 14D-16GFX is not supported on the list of qualified modules for an 3000 series cpu.
The 32GTZ is however. Manual is available at https://download.msi.com/archive/mnu_exe/mb/E7C91v2.1.pdf Go to Page 53 to start the advanced Memory timings.
One point of reference is this:
https://www.overclock.net/threads/ryzen-memory-support-duel-rank-2-x-f4-3200c14d-32gtz-kit-64gb.1635742/
Their timings are aggressive but the main cl14 is within range.
One more link:
https://rog.asus.com/forum/showthread.php?95000-Need-Help-G-Skill-Trident-Z-F4-3200C14D-32GTZR/page2
They are the GTZR version, while there is the slightest difference you can use the Manual timings as a reference point to input and Work from. This will take trial and error.
GPU (AMD/nvidia) would make no difference.
@Gandhi
At this page. https://us.msi.com/Motherboard/support/X470-GAMING-PRO-CARBON/#support-mem-19
Pick: Matisse CPU option and skim down to your modules.
err… checking again I cannot find your modules… 
I may have entered the wrong modules initially. Sorry about that. Nevertheless I can use a second opinion on supported modules. Combover it again.
@Whipple_Jaeger So, I went through the list and while I couldn’t find my ram I did find F4-3200C16D-32GVR which appears to just be the red variant of my current ram. I’m not entirely sure if it matters or not.
@wendell I just crashed again on the 2T setting.
1 Like
I’m trying to understand something here, do I understand correctly, that in your case system crashes occasionally?
I’m asking because my system previously only crashed once, unless powered off and never more than once per power cycle, granted, we have different series, but still.
P.S. I’m running stable now since 30th with no restarts (currently on 5.10.4 kernel). I’ve checked my settings, I’m running with PSU Idle set to Standard.
P.P.S btw, before it became stable, I was playing around with memory settings and on one of the failed attempts system stopped booting, so I removed battery from the motherboard and since then it’s working fine, it happened to coincide with 5.10 kernel update, so I’m not sure what did the trick tbh. Talking to AMD about RMA at this point as suggested by support, but also not sure if it makes sense to go through with it, since it’s working perfectly right now.
Any chance you have a 2019 or 2020 manufactured psu you can borrow or use temporarily? Wondering if you got two different ones with the same design flaw…
What is your Soc voltage running at?
QVL means it is tested working, not that anything on there does not work.
Very true, the ax4u300038g16-dbg used on a recent Asrock b450m pro4 build are not “qualified” in Asrock’s qvl listing. Matter of fact, 4 dimm kit is differing production dates plus differing IC (hynix mfr & spectek [micron] e-die). IC’s also go back and forth on internal subtimings. These modules on a i5_7600k + z270 is PRIMO, the XMP setting applied = totally normal behavior, 24/7 stability and burn in without a hiccup. In the Zen+ system (swapped only motherboard and cpu) xmp is a crawl in windows 10… and eventual crashes minutes in. (Linux is a no go, USB 2.0 distro drive I’ve used countless other builds does not even make it through initial setup). Zen+ at 2933 and same subtimings and this mother is as stable as the kaby-lake.
All I’m saying is: get KNOWN working components for your build or else be mindful of concessions to stability and or speed. You’re trying to turn yourselves into pretzels and doing mental gymnastics nearly 80 posts in this matter… swap out the ram or find other points of failure.
Thank you @Whipple_Jaeger.
I updated my BIOS to Agesa 1.1.9.0 (beta). The boot messages “No irq handler for vector” are gone now.
Now I try with this BIOS and XMP settings.
I will try memory settings with manual timings if the system crashes again.
@agurenko, I had a stable system for 13 days (Fedora 33 and kernel 5.10.x). After 13 days, on December 31, I had another mce errror. Now I will wait one month before I said that I have a stable sysstem.
I just got the “no irq handler for vector” boot message recently with fedora 31. Happened after updating an Asus x570 to uefi 3001. I couldn’t pin it down initially, next day ASUS took down the bios, relisted it as beta, THEN updated the 3001 on their internal servers. I now have to wait for a progressive uefi… and CANNOT roll back into previous version because… reasons. Kind of a paradox for me now, this board has over 25 uefi revisions since release… ohh brother 
No irq handler for vector messages are due to IOMMU, disabling it, removes the message, but it’s good to know that new bios fixes that. I’ll try mine tomorrow.
As for the mce errors, you made me so das, I was hoping it’s over. I’ll proceed with rma then, I guess