[SOLVED] Crazy crash when using 2 RAM modules (32 GB) - lines across the screen and repeating sound

[rene]

The issue of this happening with 32G but not 16G denies basically any and all pedestrian causes such as corrupted files. Yes, it's very interesting to test with Windows. Which would not be to say that I expect it would show the same problem. I'd have found more reports if it did...

[SMG]

I moved the swapfile to a HDD not a long time ago, because I didn't like the system constantly swapping to my SSD and wearing it off.

Does your current inxi information indicate that drive is part of your Mint system?

[rene]

Oh, by the way, you seemingly didn't yet test the newer 5.11 kernel as I suggested above as well. So that's then what I suggest: Windows, dis_ucode_ldr, 5.11. And then I'll admit defeat...

[kubajz 22]

The last crash happened on 5.11 and it was in Blender. I too don't expect the Windows system to crash but I want to be sure.

[kubajz 22]

Probably not related but this is what the output of sudo dmidecode -t 17 looks like:

Code: Select all

# dmidecode 3.2
Getting SMBIOS data from sysfs.
SMBIOS 3.0.1 present.

Handle 0x000E, DMI type 17, 40 bytes
Memory Device
	Array Handle: 0x000D
	Error Information Handle: No Error
	Total Width: 64 bits
	Data Width: 64 bits
	Size: 16384 MB
	Form Factor: SODIMM
	Set: None
	Locator: ChannelA-DIMM0
	Bank Locator: BANK 0
	Type: DDR4
	Type Detail: Synchronous
	Speed: 2667 MT/s
	Manufacturer: Kingston
	Serial Number: 268A63DF
	Asset Tag: 9876543210
	Part Number: 9905663-030.A00G    
	Rank: 2
	Configured Memory Speed: 2667 MT/s
	Minimum Voltage: 1.35 V
	Maximum Voltage: 1.5 V
	Configured Voltage: 1.2 V

Handle 0x000F, DMI type 17, 40 bytes
Memory Device
	Array Handle: 0x000D
	Error Information Handle: No Error
	Total Width: 64 bits
	Data Width: 64 bits
	Size: 16384 MB
	Form Factor: SODIMM
	Set: None
	Locator: ChannelB-DIMM0
	Bank Locator: BANK 2
	Type: DDR4
	Type Detail: Synchronous
	Speed: 2667 MT/s
	Manufacturer: Kingston
	Serial Number: 1CBF5963
	Asset Tag: 9876543210
	Part Number: 9905663-030.A00G    
	Rank: 2
	Configured Memory Speed: 2667 MT/s
	Minimum Voltage: 1.25 V
	Maximum Voltage: 1.5 V
	Configured Voltage: 1.2 V

When you look at the minimum voltages for both modules, they are
1) not the same, even though the modules should be identical
2) above the recommended operating voltage of 1.2 V
I don't understand this.

[rene]

Looks odd indeed... but if the system's stable under Windows, apparently it's not relevant.

I believe the min/max voltages are read from the DIMMs themselves. They are definitely both part no. 9905663-030.A00G so apparently 1.25 vs. 1.35 minimum is a valid tolerance. But as to the configured voltage being below either...

For Windows I know of the utility CPU-Z to view this information. While I expect that it'll be the same I'd check what it says; generally and/or historically speaking I have little faith in data reported by dmidecode as such. Second I'd definitely run the system through a memory stress test under Windows (although I note you said you already tried with memtest). No recommendation for a specific Windows tool...

Tried that dis_ucode_ldr thing already as well?

[kubajz 22]

I did and it crashed. Kernel 5.8.

[rene]

Crap. Yah, well, seeing your DIMMs report being configured with a voltage lower than their advertised minimum, in one case significantly, does raise a brow. I'd as said check CPU-Z and a memory stress test under Windows.

[kubajz 22]

The thing is that 1.2 V is their recommended operating voltage. It's written on the box they came in and on every website with the specs for these modules.

[rene]

Yes, as to websites I noticed as much as well when I googled the part number. As said, not generally very trusting as to dmidecode information and CPU-Z is as such interesting, even if I don't in fact expect it to show different; if its read from the DIMMs (the SPD) itself then there seems little opportunity even for DMI data to be incorrect, but, well, now that you have a Windows install to test things on as well...

[kubajz 22]

CPU-Z doesn't tell me the minimum or maximum voltage, it only shows the one configured.

[rene]

Yes, so sorry again for providing "non-optimal" information. Or, in fact, maybe it's a good thing, as it almost certainly means that those values are not in fact read from the DIMM itself and would as such be subject to the general issue of DMI-data, BIOS-set/maintained data, being unreliable. This seems to jell with e.g. https://github.com/lpereira/hardinfo/issues/380 where the minimum and maximum voltage are listed under the motherboard header and only the supported voltage(s) under the SPD (an on-DIMM chip) header. Which then is to say that I feel that chances of this being able to be ignored just now increased --- although even for DMI data it seems weird that they're not at least the same.

If Windows is definitively stable it might be the case that people with a chance of knowing are to be found mainly on the linux-kernel malinglist. Should you decide to post there be sure to post the journal from 5.11.16 (a kernel relatively new still also for the LKML) while noting 5.4 and 5.8 to also have the problem and without the binary nvidia module loaded since otherwise no one will even look at things. Link to that video (again, if nvidia-less) will be useful, and mention things being stable on Windows. The list tends to be very technical and it tends to be expected that you can test e.g. kernel patches but otherwise it's a simple matter of sending email to linux-kernel@vger.kernel.org. Archive at e.g. https://lore.kernel.org/lkml/. It's not generally expected that you subscribe before posting.

[kubajz 22]

I will continue testing Windows and if it doesn't fail, then I'll boot an unrelated linux distro (I'm thinking Arch). If the computer crashes there, it's obviously a linux thing and I will proceed to sending that e-mail. I have a lot of work I would like to do on the computer, so I'm constantly switching between testing and working on things..

[rene]

Yes, good idea. Arch tends to take a bit of setup; if you in the context of time needed want an Arch-based one that's more ready OOTB try Manjaro. Fedora would also work.

[kubajz 22]

So, I tried Manjaro and it kept crashing no matter what I did. I also tried booting Mint with parameters pci=noaer and pci=nomsi, but it didn't help. I remember trying nomodeset a while ago, but I couldn't boot with this parameter while using the proprietary nvidia driver, so I abandoned this idea. I later found out that if I switch to nouveau, I can get to the desktop. This actually worked and stopped the computer from crashing. On the other side, it disabled HW acceleration, so it's useless. I'm running with 24 GB now and it's fine, so the issue is definitely related to 32 GB..

[rene]

Note that, no, "nouveau" doesn't disable hardware acceleration. Also said above somewhere in the context of it being possibly enough for you. The Nvidia binary driver may be better (and as far as I'm aware it certainly is in the context of e.g. multiple displays and/or switching integrated/discrete graphics on the fly for example) but in a basic sense the difference between "nvidia" and "nouveau" is one being closed-source and the other open-source. The less basic sense in which former is/can be better is it being developed directly by the people who know the hardware "secrets" whereas latter needs to be reconstructed by observing e.g. former.

That "nomodeset" thing you say is the one that disallows any and all hardware acceleration and will indeed as such not be useful to you but without it "nouveau" certainly is hardware accelerated, so if something is/was up there it should be investigated as not being expected.

Windows is stable?

[kubajz 22]

Didn't read the post after myself.. I meant to say that nouveau + nomodeset gets me to the desktop and the system is stable. This doesn't work with proprietary + nomodeset and results in a black screen. Although I can switch to a different tty using ctrl+alt+Fx. Windows seems to be stable as well.

[rene]

OK, and I take it then that "nouveau" without nomodeset was not stable. Just now reviewed the thread and I don't believe that was already mentioned.

Well, with Windows stable I'd myself be rather keen on getting to the bottom of things but without the hardware and apparently up to now no good ideas either.... dunno

[kubajz 22]

Just noticed that in this thread viewtopic.php?f=47&t=348909 the author posted an output of inxi and it seems like they don't use modeseting (it's unloaded). Why can't I do that as well without losing acceleration? Is that because I have 2 GPUs?

[SMG]

Just noticed that in this thread viewtopic.php?f=47&t=348909 the author posted an output of inxi and it seems like they don't use modeseting (it's unloaded). Why can't I do that as well without losing acceleration? Is that because I have 2 GPUs?

Yes, it is because you have Intel graphics. The display driver for Intel is modesetting.

If your laptop has an option in BIOS/UEFI to use only the discrete (Nvidia) card, then modesetting will not load.