XMP Profile stable in Memtest86+ and Prime95, but freezes when exiting BIOS

[Mastakill]

Hi everyone,

I've just replaced my 32GB RAM with 64GB RAM, but I'm having some trouble I was hoping you guys can help me with...

Below my hardware:
AMD Ryzen 9 3900XT Boxed
MSI MEG X570 UNIFY (latest non-beta BIOS 7C35vAA)
ASRock Radeon RX 5700 Challenger D 8G OC
Fractal Design Define R6 USB-C
Arctic Liquid Freezer II 280
Kingston HyperX Predator HX436C17PB3K4/64 (replaces G.Skill Trident Z F4-3600C17D-32GTZ)
Fractal Design ION+ 760W Platinum
Corsair Force MP510 960GB
Crucial M550 2,5" 1TB

My 64GB RAM is listed on the motherboards compatibility list for functioning at the XMP profile at 3600MT/s with 4 DIMMs and that is also exactly how I run it.

It is Memtest86+ stable (4 passes ~10h)
It is Prime95 stable (~6h blend)

But it is having problems in the BIOS... When I reboot from the BIOS (after changing or not changing an option), the computer hangs when rebooting (about 80% of time - sometimes it also doesn't hang).
What happens is: I press F10, confirm, the computer reboots, black screen (monitor powers off) and then normally it should start POST, but the screen remains black and it hangs. Sometimes it works if press the reset button, sometimes it doesn't. Often I have to press the power button for 5s and then power it on again. It does store the BIOS changes (if I made them).

I didn't notice issues with the cold boot yet.

Also I just remember that the first time I ran Memtest86+, it "crashed" into black screen. I actually switched my monitor to my laptop while running Memtest86+, but when I switched back to my desktop about an hour later, then screen remained powered off. Later runs of Memtest86+ never repeated this issue though and it even passed a full 4 runs (10h).

Anyone know the cause or what I could try to fix this?

Thanks!!

 

[pendragon1]

try bumping the ram voltage to 1.4v and vsoc/vddrsoc(whatever your bios calls it) to 1.1/1.115v
(yes its safe, i would advise it otherwise)

 

[pitingres]

If pendragon1's advice doesn't cure it, I'd try backing off the memory transfer rate in the BIOS a bit. Try one or two notches below 3600. You can take it down to 3200 or even a bit less, without really impacting performance for most situations.

I'm assuming that you actually need the 64 GB, in which case getting stability would be more important than maintaining the top speed.

 

[Mastakill]

try bumping the ram voltage to 1.4v and vsoc/vddrsoc(whatever your bios calls it) to 1.1/1.115v
(yes its safe, i would advise it otherwise)

Thanks, but isn't there a way to "fix" the issues by only changing (like increasing voltage) something during POST? Because when in the OS, everything seems rocksolid stable, no?
I agree slightly bumping voltages probably isn't a disaster, but it will slightly decrease lifespan and I'd like to avoid it if possible. Also the RAM / Mobo officially supports these settings without increasing the voltage, so it should also work without, no?

If pendragon1's advice doesn't cure it, I'd try backing off the memory transfer rate in the BIOS a bit. Try one or two notches below 3600. You can take it down to 3200 or even a bit less, without really impacting performance for most situations.

I'm assuming that you actually need the 64 GB, in which case getting stability would be more important than maintaining the top speed.

I didn't buy expensive memory that is on the memory compatibility list for these exact settings, to then run it at a lower speed

 

[pendragon1]

Thanks, but isn't there a way to "fix" the issues by only changing (like increasing voltage) something during POST? Because when in the OS, everything seems rocksolid stable, no?
I agree slightly bumping voltages probably isn't a disaster, but it will slightly decrease lifespan and I'd like to avoid it if possible. Also the RAM / Mobo officially supports these settings without increasing the voltage, so it should also work without, no?

maybe, dig in the bios.
lol, youll have moved on from it loooong before thats ever an issue. nope. XMP is technically intels spec, so no it doesn always "just work".
i love when people ask for advice and then question it all....
gl.

edited speeling.

 

[toast0]

Thanks, but isn't there a way to "fix" the issues by only changing (like increasing voltage) something during POST? Because when in the OS, everything seems rocksolid stable, no?

Stop messing around in the BIOS then

 

[Mastakill]

maybe, dig in the bios.
lol, youll have moved on from it loooong before thats ever an issue. nope. XMP is technically intels spec, do no it doesn alwasy "just work".
i love when people ask for advice and then question it all....
gl.

I certainly appreciate the advice and don't question it. I was just wondering if there were perhaps less "invasive" alternatives than increasing the voltage "in all situations, even when it isn't really required".

 

[Mastakill]

I just tried increasing the RAM voltage from 1.35v to 1.38v and rebooted succesfully from BIOS about 5 times. So it certainly seems to help... Next I'll try to play with this setting a bit more, to see which is the minimum that is stable and then set about 2 0.01v steps higher than that, to make sure.
Edit: 1.35v failed to boot on the third attempt, 1.36v didn't fail in 16 attempts, so 1.38v was set

I could not find vSOC or vDDRSOC though. Following settings looked most like it:
CPU NB/SOC Voltage
Chipset SOC Voltage
Both have a default around 1v

Which one did you mean (although on first sight, it may no longer be required to change that voltage setting as well).

 

[owcraftsman]

This is why I test memory from within Windows. I use HCI Design Memtest which is free to use but I use the paid version because it's less hassel setting up to utilize 100% of the memory during testing. If you can't boot its your 1st indicator something is wrong.

 

[Mastakill]

That's the weird thing... I could boot to Windows and run Prime95 blend (which also does a pretty good job testing memory I thought?) for many hours without any signs of instability. It was only when booting after a BIOS F10 that it would often hang...

 

[owcraftsman]

P95 is good but honestly I've run both (HCI/P95) at the same time. Even blend doesn't do much to stress ram the RAM it self the memory controller that resides on the CPU die is another story. Hoenstly, looking at you specs again I'd say the memory controller is weak and has issue with the 64GB. It's kinda like the silicone lottery when it comes to that. Any time all 4 dimm slots are populate it is stressful on the CPU add to that the huge amount you have it's even more stressful. Try 2 sticks populated in the appropriate slots as recommended by your motherboard manual and see if problems still exist. If no issue there at the same setting then I'd say you have your answer.

 

[Mastakill]

It seems like it is stable at 1.38v, but I don't mind running some additional tests to make sure of it...

"I've run both (HCI/P95) at the same time"
How did you do this practically? Both Prime95 and HCI are designed to use ALL your RAM. Should I start HCI 24 times (because I have 24 threads) and then divide half of my RAM over these 24 instances and start testing this half of the RAM with HCI? And then start Prime95 blend to use the other half of the RAM?

Btw: I'm not interested in running sticks of RAM. That's not why I bought of them. Again... Everything seems ok at 1.38v... It is Memtest86+ stable (10h) and Prime95 (6h) and I didn't encounter any issues for a week now...

 

[owcraftsman]

It seems like it is stable at 1.38v, but I don't mind running some additional tests to make sure of it...

"I've run both (HCI/P95) at the same time"
How did you do this practically? Both Prime95 and HCI are designed to use ALL your RAM. Should I start HCI 24 times (because I have 24 threads) and then divide half of my RAM over these 24 instances and start testing this half of the RAM with HCI? And then start Prime95 blend to use the other half of the RAM?

Btw: I'm not interested in running sticks of RAM. That's not why I bought of them. Again... Everything seems ok at 1.38v... It is Memtest86+ stable (10h) and Prime95 (6h) and I didn't encounter any issues for a week now...

I understand you want all 4 sticks of RAM installed but we are troubleshooting to narrow down the cause of your issue. I only suggest it as a temporary test. If everything runs fine with only two sticks we can point to #1 your CPU and #2 Motherboard as the root cause. In other words its likely one or the other causing you problems. Your next step would be to replace one or the other and test again. Conversely if you still have issues with only two sticks then it's likely the BIOS or a bios setting or the RAM it self. If you think it's RAM then you need to test each stick individually until you find the culprit. In other words it's not likely all sticks are bad. You should be getting the idea by now. You need to be methodical trouble shooting.

BTW if 1.38vram solved your problems? Great however, if you had issue with the XMP Profile defaults set in bios then I would run the test and report the issue to GSkill and AsRock and seek there advice. The 1st thing they will ask is, is it on the QVL which you claim it is so they would want to know you are having the issues. This is how bugs in a BIOS version get fixed. As a general rule if you are having the problem then there is likely many more having the same issue but until enough people report the same issue it's not on their radar to fix. Contrary to popular belief they do not data mine Forums to fix their stuff.