Question RAM module crashes on specific apps, the other one of the same model works fine.

[pabloasus]

Hi, I've been having irregular problems with crashes, slowdowns, and occasional BSODs and CTDs, for over 3 years since I built my current pc. And because they didn't seem to have any pattern, I always suspected software issues rather than hardware, and it took me a long time to discover that they only happen with one of the two ram sticks. (should've tested that earlier, I know...)

SPECS:

(All default settings, memory at XMP)
Gigabyte B450 Aorus M
ADATA XPG SPECTRIX D41 8GB DDR4 3000MHz x 2 (individual sticks that came separately)
AMD Ryzen 5 3600
Gigabyte NVIDIA 1660Ti OC
Samsung SSD 970 EVO Plus 500GB
Samsung SSD 850 EVO 500GB
Western Digital Blue 7200RPM 1TB
Thermaltake TP GRAND RGB 650W Gold 80+

The story is that I've ran countless tests, updated stuff, and more over the years, to try to figure out why certain games would crash a lot, but many others (the majority in fact) would be fine. Now I've forgotten lot's of what I did ofc, but every diagnostic I did came out fine. Problem was always in games, photoshop, 3D benchmarks, and maybe 1 or 2 other apps I don't remember.

What happens is that the apps will start and sometimes even run for 3 or 4 hours, but then crash, and only a full pc restart would "refresh" that a bit until it crashes again. However, many times it did not work and things would crash a lot after restart anyway, so if it really made a difference it was erratic. The crashes create reports of varied driver crashes, often video ones, and errors coming up from game engines often mention "device removed", "memory exceptions."

Eventually, I discovered that they only happen with one stick (tried it on default and XMP only), and nearly every game or benchmark that crashes uses OpenGL or DirectX12. AFAIK no DirectX11 or older game has crashed on me enough to be the same issue, but some DX12 or OGL apps have ran well without issues so its not a 100% trigger.

Some examples: Control crashes in DX12, not in DX11, Cyberpunk, Warzone 2100 crashes on OGL not on DX or Vulkan, Doom and Doom Eternal, Photoshop CC2019
Unigine Heaven, Superposition, Furmark, all crash on OGL. GFXBench however, runs all tests fine on all APIs except for the OGL offscreen tests, which all crash and could never finish.

So, what I want to know is what does everyone think my case might be about: faulty stick, or different stability threshold than for the other module? or maybe it is actually a software problem, or another part is the issue?
If there could be a simple fix like a tweak in the bios that'd be great, since going through the warranty process is more work and I'm scared they might not be able to reproduce the issue or won't care because of how specific it seems to be.

Thanks a lot, sorry for all the text.

 

[mikeymikec]

Does it matter? If it fails to perform to specs then you ought to send it back, I would have thought?

The thing is, you spent extra $ on getting XMP RAM. If you're OK with running it at stock rather than XMP and if it's OK with that, I guess you've got a solution that doesn't involve customer services.

However, if you spend longer at it than that, then you have to start weighing up how much time you've wasted vs potentially getting a replacement sent out more or less straight away.

 

[igor_kavinski]

I don't wanna bore you with details but from my own experience, Gigabyte mobos prefer Kingston RAM.

To rule out your current RAM sticks as the problem, run the built-in Windows 10/11 memory diagnostics FULL test, once with each stick and then with both sticks. If all three tests pass, the issue is something else (maybe a bad driver?). You can try to rule that one out by giving Linux a go for a few days, on a spare SSD.

Finally, you can try a different graphics card if you can borrow from somewhere, or get something cheap, like a Geforce 1030 or even 730.

 

[mikeymikec]

I don't wanna bore you with details but from my own experience, Gigabyte mobos prefer Kingston RAM.

To rule out your current RAM sticks as the problem, run the built-in Windows 10/11 memory diagnostics FULL test, once with each stick and then with both sticks. If all three tests pass, the issue is something else (maybe a bad driver?). You can try to rule that one out by giving Linux a go for a few days, on a spare SSD.

Yup, I go Crucial or Kingston all the way too. In my line of work I just want to be able to stock RAM and for it to just work as I see a fair variety of boards.

 

[Shmee]

Sounds like one of the DIMMs isn't stable at rated XMP. You could run memtest86 on it to be sure, and will probably get lots of errors. If so, return or RMA it.

 

[igor_kavinski]

If there could be a simple fix like a tweak in the bios that'd be great, since going through the warranty process is more work and I'm scared they might not be able to reproduce the issue or won't care because of how specific it seems to be.

You said that you have been experiencing these issues for three years. Does the RAM have lifetime warranty?

Simplest thing you can do is with only the bad RAM stick installed, start dropping the RAM speed and/or CAS latency and try to crash your PC in the way you know it crashes with whatever games/applications. Hopefully, at some lower setting, you will achieve perfect stability.

@Shmee @mikeymikec

Is it possible for the memory controller to run one stick at a lower speed and the other at a higher speed in different set of slots, both in single channel mode?

If the above is possible, OP can benchmark to see if doing that gives better performance than running both sticks at a lower stable speed in dual channel mode.

 

[mikeymikec]

@Shmee @mikeymikec

Is it possible for the memory controller to run one stick at a lower speed and the other at a higher speed in different set of slots, both in single channel mode?

If the above is possible, OP can benchmark to see if doing that gives better performance than running both sticks at a lower stable speed in dual channel mode.

Not in my experience. Everything gets run at the lower speed that the memory controller thinks all the modules can be run at.

---

If one of the modules fails during memtest86 it should be an open-and-shut case with the manufacturer. I think I've only RMA'd memory in one or two cases but there's never been any argument.

 

[pabloasus]

Does it matter? If it fails to perform to specs then you ought to send it back, I would have thought?

The thing is, you spent extra $ on getting XMP RAM. If you're OK with running it at stock rather than XMP and if it's OK with that, I guess you've got a solution that doesn't involve customer services.

However, if you spend longer at it than that, then you have to start weighing up how much time you've wasted vs potentially getting a replacement sent out more or less straight away.

Hi.
Basically It matters to me because I'd be willing to accept an easy fix even if it runs a bit worse, depending on how the RMA service goes. Naturally, I'm gonna try to get it replaced (and have already been talking to ADATA), but I'm just keeping my options open in case the replacement process is too much of a hassle.

I don't wanna bore you with details but from my own experience, Gigabyte mobos prefer Kingston RAM.

To rule out your current RAM sticks as the problem, run the built-in Windows 10/11 memory diagnostics FULL test, once with each stick and then with both sticks. If all three tests pass, the issue is something else (maybe a bad driver?). You can try to rule that one out by giving Linux a go for a few days, on a spare SSD.

Finally, you can try a different graphics card if you can borrow from somewhere, or get something cheap, like a Geforce 1030 or even 730.

Okay thanks for the tip.

I've ran the windows10 full test 2 times with both sticks, and 1 time with the bad one, they all passed. In fact every single RAM test I've ever tried on any configuration has passed always, that's what delayed my suspicion about RAM for so long. iirc I did Windows test, memtest86, passmark, occt, memtest. Although not all of them I ran on the stick just by itself, that'd be only windows, memtest86, passmark.

I've tried my old GTX970, problem persisted.

Next week however, I'm going to test both sticks in a friend's pc.

Sounds like one of the DIMMs isn't stable at rated XMP. You could run memtest86 on it to be sure, and will probably get lots of errors. If so, return or RMA it.

Hi, I ran it already along with other tests and it passed all of them, some with both sticks installed, some with only the problematic one. Memtest86 came out fine in both configurations.

You said that you have been experiencing these issues for three years. Does the RAM have lifetime warranty?

Simplest thing you can do is with only the bad RAM stick installed, start dropping the RAM speed and/or CAS latency and try to crash your PC in the way you know it crashes with whatever games/applications. Hopefully, at some lower setting, you will achieve perfect stability.

@Shmee @mikeymikec

Is it possible for the memory controller to run one stick at a lower speed and the other at a higher speed in different set of slots, both in single channel mode?

If the above is possible, OP can benchmark to see if doing that gives better performance than running both sticks at a lower stable speed in dual channel mode.

Hi thanks for responding.

Yes the memory has lifetime warranty and I'm talking to the manufacturer, but the better I understand the issue the betterI can prove it to them, or help them reproduce it, or in the worst case maybe I can work around it if I can't get a replacement.

So far the problem has behaved apparently the same both in the default and XMP configs. So I assume you mean I should go even lower than default?

Could you tell me more specifically what settings should I tweak and by how much to test it? and if I find a stable config will the other ram stick need to run at the same settings to match it?

Oh and about your hypothesis of the memory controller, when I test my RAM individually, I put them always in the 2nd slot, A2 or slot number 1 of 0-1-2-3. So both sticks have been tested in the same slot and the results are still different.

 

[igor_kavinski]

From the specs: https://www.xpg.com/us/xpg/552?tab=spec

In BIOS, disable XMP, set speed to 3000 MT/s, set the RAM voltage to 1.2V and CAS latency to 19. Try crashing the PC with the bad stick with this setting. If it passes, start lowering CAS latency bit by bit, 18, 17 and finally 16 and see if it still passes. If it passes, insert the good stick in the other slot and repeat to test stability.

If crashes happen despite lowering the settings, something else is the matter or simply, ADATA RAM doesn't like your mobo.

 

[pabloasus]

From the specs: https://www.xpg.com/us/xpg/552?tab=spec

In BIOS, disable XMP, set speed to 3000 MT/s, set the RAM voltage to 1.2V and CAS latency to 19. Try crashing the PC with the bad stick with this setting. If it passes, start lowering CAS latency bit by bit, 18, 17 and finally 16 and see if it still passes. If it passes, insert the good stick in the other slot and repeat to test stability.

If crashes happen despite lowering the settings, something else is the matter or simply, ADATA RAM doesn't like your mobo.

With XMP disabled, the default settings were 2666, 1,2V, and CAS 20, and the crashes are still happening, with no noticeable improvement at all.

 

[igor_kavinski]

With XMP disabled, the default settings were 2666, 1,2V, and CAS 20, and the crashes are still happening, with no noticeable improvement at all.

That confirms it can't be the RAM. I mean, going lower than 2666 MT/s would be a disgrace to that CPU. Can you borrow RAM from your friend?

 

[altaftp]

Sounds to be like one of the RAM module is corrupted. Remove one RAM module at a time and check if your PC works fine. Repeat the process with the second RAM after removing the First module.

Or you can run a Windows Memory diagnostics to find the problems with your RAM modules.

 

[DrMrLordX]

Gigabyte mobos prefer Kingston RAM.

Not really. It has more to do with the ICs than the PCBs. I wouldn't touch Kingston these days for AM4. Micron e-die all the way. Crucial. Unless you just have to get b-die for some reason.

@pabloasus

I think you're giving yourself too many headaches worrying about the RAM. Take the time to learn about Ryzen Memory Calculator and see what you can do with your DIMMs. Using XMP on AM4 is not always the right way to go; in fact, it's frequently problematic in some fashion. Many people never notice problems over the life of their system.

Unless you're budget-constrained or otherwise dedicated to not spending any more money on that particular system, I highly recommend looking at some cheap Micron e-die kits from Crucial. 2x8GB from them will be stupidly cheap, and 2x16GB is still affordable. You can get very good RAM that will play along well with your R5 3600 and your motherboard.

 

[igor_kavinski]

Using XMP on AM4 is not always the right way to go; in fact, it's frequently problematic in some fashion.

Ah. Makes perfect sense since XMP was developed by Intel. I naively thought that maybe BIOS support on AMD mobos allows Ryzen CPUs to take advantage of XMP but even if that's the case, it seems that it is not the same level of compatibility as on Intel platform.

 

[DAPUNISHER]

Swapping for the friend's ram is the way to go.

Note that 3&4 series boards frequently saw improved memory compatibility and stability via bios updates. If it doesn't have the latest version, it is a good idea.

If the other ram kit experiences instability, and a bios flash doesn't resolve it either, it is the board or CPU. I have read more reports by users, around the web, of faulty Ryzen 3600s than any other model. That could simply be because it has been wildly popular of course. Bigger sample size, more complaints.

As to Gigabyte AM4 boards and ram: I have 3 of their AM4 boards at the moment. An X570, B550, and B450 based. I have 4x8GB in the X570 and B450, The X570 has 2 completely different kits from different vendors with XMP set, Zero issues. The B450 is the super cheap $47 a kit Silicon Power 3200, also zero problems. The third has 2x16GB Patriot Viper Elite II 3600, and you guessed it, no problems.

I don't know what is supposed to be problematic about XMP and AM4 but I have yet to see it. This includes a half dozen combos I have sold on after playing with for a few months. And the Zen and Zen+ client systems I built and/or worked on when in the biz. Benchmarks are always right where they should be. Systems run rock solid.

EDIT: I will note that I have had a kit that wouldn't run the rated XMP in AM4 or Intel Z490. Geil Orion 3600MHz. Wouldn't do better than 3466 even with extra voltage and looser timings.

 

[DrMrLordX]

Ah. Makes perfect sense since XMP was developed by Intel.

More specifically, XMP normally provides timings that are intended for use with an Intel IMC. Using the same timings on an AMD CPU can be difficult and require silly amounts of voltage. Not just vDIMM either.

 

[pabloasus]

That confirms it can't be the RAM. I mean, going lower than 2666 MT/s would be a disgrace to that CPU. Can you borrow RAM from your friend?

Can't borrow to test in my system, but I what I'm gonna do tomorrow is put my ram into his pc and see if it functions weird.

But again, for a working setup with a different ram module, I got my own 2nd module right? that one works flawlessly in the same situations the other one doesn't. Sure its not the memory still? I don't know what to think of other than the memory, unless my system has some problem that manifests itself with one of the sticks only.

Swapping for the friend's ram is the way to go.

Note that 3&4 series boards frequently saw improved memory compatibility and stability via bios updates. If it doesn't have the latest version, it is a good idea.

If the other ram kit experiences instability, and a bios flash doesn't resolve it either, it is the board or CPU. I have read more reports by users, around the web, of faulty Ryzen 3600s than any other model. That could simply be because it has been wildly popular of course. Bigger sample size, more complaints.

As to Gigabyte AM4 boards and ram: I have 3 of their AM4 boards at the moment. An X570, B550, and B450 based. I have 4x8GB in the X570 and B450, The X570 has 2 completely different kits from different vendors with XMP set, Zero issues. The B450 is the super cheap $47 a kit Silicon Power 3200, also zero problems. The third has 2x16GB Patriot Viper Elite II 3600, and you guessed it, no problems.

I don't know what is supposed to be problematic about XMP and AM4 but I have yet to see it. This includes a half dozen combos I have sold on after playing with for a few months. And the Zen and Zen+ client systems I built and/or worked on when in the biz. Benchmarks are always right where they should be. Systems run rock solid.

EDIT: I will note that I have had a kit that wouldn't run the rated XMP in AM4 or Intel Z490. Geil Orion 3600MHz. Wouldn't do better than 3466 even with extra voltage and looser timings.

Hi thanks for all the info.

I will report what happens with my friend's pc tomorrow.
bios was updated yesterday to last version along with chipset drivers, no change or improvement, still crashes.

Just to be clear, I have only 2 8GB sticks, and one works perfectly fine as far as I can tell, no crashes or anything weird. But if the other stick is in my pc, either along with the other one, or by itself, that's when the problems appear.

If the processor was faulty how could I find out testing it with my good ram module? I'd need to see that case, so I can be sure it wasn't the bad module causing the problems instead like I suspect now.

Also a semi-random question, does anyone know what "XMP high frequency support" option does in the BIOS for gigabyte? it's at Auto by default if I turn XMP on.

 

[igor_kavinski]

But again, for a working setup with a different ram module, I got my own 2nd module right? that one works flawlessly in the same situations the other one doesn't. Sure its not the memory still?

Could be that the "bad" RAM stick just has some different chip than your "good" RAM stick and that different chip doesn't like your mobo BIOS. If that's the case, the bad one's gotta go.

 

[DAPUNISHER]

XMP high frequency support is the overclock profiles I think. E.G. level 1 3200MHz level 2 3466MHz Level 3 3600MHz

I have to agree with Igor, it sounds like a bad stick.

 

[igor_kavinski]

Some companies use bait and switch tactics to make "easy" money. Release good product, build reputation then start skimping on materials. If OP got the 2nd module after some passage of time, that might mean a lower quality product was pushed into the market with the same model number.

 

[DrMrLordX]

Just to be clear, I have only 2 8GB sticks, and one works perfectly fine as far as I can tell, no crashes or anything weird. But if the other stick is in my pc, either along with the other one, or by itself, that's when the problems appear.

Yeah it does seem like you just have one stick that won't run XMP settings properly. Which is why you should be looking at something like Ryzen Memory Calculator if you absolutely must salvage that specific hardware. If you have the budget, you really should replace it with some Micron e-die.

 

[pabloasus]

Alright so I tried both of my sticks in a friend's pc, and both of them worked properly at least for the little time I had to try them. It has to be the interaction of that specific module with my system what triggers the issue.

One thing I noticed though, was that the memory had different timings by default. I suppose this is normal for a different board and cpu, but it makes me wonder if I can tweak something to make it stable on my computer. The main timings were all the same except cas, which was 19 instead of 20, and command rate at 2 instead of 1 as it is on my pc.

Would it make a difference to try matching it?

Yeah it does seem like you just have one stick that won't run XMP settings properly. Which is why you should be looking at something like Ryzen Memory Calculator if you absolutely must salvage that specific hardware. If you have the budget, you really should replace it with some Micron e-die.

I looked a bit into RMC, and it seems people use that with another one called Thaiphoon, and I'd rather not use it cause of all the virus discussion around it, any alternatives?

 

[igor_kavinski]

command rate at 2 instead of 1 as it is on my pc.

CR at 2 is more stable but slower. That could explain the instability if your BIOS is forcing the RAM to run at CR 1.

 

[DrMrLordX]

I looked a bit into RMC, and it seems people use that with another one called Thaiphoon, and I'd rather not use it cause of all the virus discussion around it, any alternatives?

Virus discussions? Thaiphoon Burner is a legit application. Or at least it was when I downloaded it. The alternative is learning to hand tune DDR4.

 

[igor_kavinski]

Virus discussions? Thaiphoon Burner is a legit application. Or at least it was when I downloaded it.

May be referring to this: malware - Can being hacked/cyberattacked once make you more vulnerable to being hacked again? - Information Security Stack Exchange

I had a pretty nasty Trojan attack trying to run some stupid mp4 file with non-standard codec when I downloaded some (maybe Korean, maybe KMplayer) media player through a standard Google search (it was in the top few). Bloody virus only got detected after I tried to uninstall the application (it was probably in the uninstaller EXE, encrypted to evade detection). Despite being detected and removed, it managed to hook deep into my work PC's Win7 installation and soon after, my CC got compromised. I immediately firewalled that PC and can't use internet freely on it (I've a backup work PC for that). That installation has too many important daily used applications on it, so I'm not going through the pain of a complete wipe and re-install. I'm still astonished that it was so stealthy and dangerous. I thought getting infected so easily was a thing of the past. Apparently not.