[SOLVED] Unstable RAM, but it's good RAM ?

[Veedrac]

I built a new computer. I'm technically competent but this is still my first desktop, and I mostly don't have spares for testing. I'm running Linux.

1. My computer is unstable with the RAM starting at ~3300 MT/s. Below 3200 MT/s it is stable. I am using DDR4 3600 memory.
2. prime95 on mixed mode almost instantly fails when using unstable settings, but does not fail on the Small/Smallest FFTs modes that fit in caches.
3. My power supply makes a quiet buzzing sound at low power use. This buzzing sound also appears like-for-like through my USB audio interface, and I can hear it very audibly in my speakers. (It's the sort of high pitched coil whine you might recognize from old computers when you moved the mouse.)

Observations 1+2 make it seem that I have an issue with memory. However, I got a replacement pair of DRAM sticks, and they have the same problem. Therefore the DRAM itself is probably fine.

Possibilities that occured to me:

1. The motherboard has a signal grounding issue of some sort.
2. The motherboard traces to the DRAM are damaged, and observation 3 is unrelated.
3. The CPU has a faulty memory controller, and observation 3 is unrelated.
4. My power supply is broken, and not supplying properly isolated/smooth current.
5. Another component is producing weird frequencies and this is messing with signal integrity.

Possibility 1 seems the most likely, since I feel there should be more noise isolation that wouldn't show up so loud in my audio interface.
Possibilities 2-4 all seem plausible.
Possibility 5 seems unlikely. I don't believe my AIO, SSD or HDD would cause possibility 5, and my GPU is a stopgap old Quadro 2000.

Any suggestions for what's the most likely problem? I'm happy to test things.

Spoiler: Hardware

CPU: Ryzen 9 5900X
DRAM: Kingston FURY Renegade, 2x16 GB, DDR4 3200 CL16
Motherboard: MSI B550-A PRO
Power Supply: Corsair RM1000x, 1000 Watts
AIO: Kraken X63
Case: Fractal Torrent
GPU: NVIDIA Quadro 2000 (2010 Fermi GPU, 62 W)
Audio: evo 4 + JBL LSR305

 

[Grobe]

In general, there is one thing most people tend to forget - there are also a voltage regulator on most motherboards, feeding the CPU, RAM and possibly other components directly or indirectly, such as sound chip (you cannot be sure unless having access to the full schematic/PCB documentation of the motherboard).
If the internal voltage regulator on motherboard gets unstable, this can cause stability issues for RAM and CPU. If that is the case, you'd replace the motherboard unless you happens to be able to try to replace all capacitors (the upside "bucket" -like components that tend to burst inside).

I've also observed failures that I was convinced was RAM being bad, but later on turned out being the main PSU actually being the bad component. But this is many years ago, like before DDR2 - and this make sense, because having two power supplies after each other, I'm not surprised at all if an unstable main PSU in turn causes secondary voltage regulators to not working as well. But this is failures that can be hard to diagnose, because measuring the ripple voltage (aka. the quality of the output) of a PSU is not something you can just have a look at the bios menu to figure out.

Actually - when thinking about it because your 3. point - Try to figure exact where the sound comes from. You sure it is the main PSU? Because if Yes, then the it can look like the PSU is the problem.

 

[Veedrac]

Thanks @Grobe, I'm glad to hear you've seen this sort of issue before and I appreciate the details.

I managed to borrow another PSU from a friend, and the issue is still there (though the coil whine is quieter), so it sounds like the motherboard would be the culprit. Here's hoping a new motherboard fixes it.

 

[Veedrac]

Update: it wasn't the motherboard either.

I think this about leaves me with a faulty CPU as the only good culprit.

 

[Veedrac]

After replacing the CPU also, finding it had no effect, then checking the software and OS were good seven different ways, and then warbling to the gods of computation,

The motherboard wasn't setting the RAM to the correct rated voltage by default when enabling XMP.

I didn't know this was possible. Build guides never mention that XMP is anything other than a 1-step process. I had even checked the timings before this just to be on the safe side, since I'm not a total dunce, but those were fine. The voltage error wasn't even that it was too low; it was that it was a notch too high!

It turns out motherboards have this thing called a QLV list that nobody particularly cares about. I get the impression it should circumvent these issues.

E: Further, although instability seems to be greatly reduced at the correct voltage, it still does not seem to be fully stable. I am likely going to replace the RAM with another, explicitly supported brand in the near future.

 

[Grobe]

Thank you for posting updates. This may be helpful for other people finding this forum post.

Hope replacing RAM will help out the issue for you.

 

[Veedrac]

Late update:

Being somewhat burned by Ryzen, I switched to a 12700k and MSI X570-A PRO. That solved the memory errors and gave me a better GPU than my Quadro 2000 (which I had modded to remove the fan, so it was pretty jank).

It did not solve the audio problems. This means I've had audio problems with three motherboards, three CPUs, two memory kits in many configurations, two power supplies, different drive configs, different GPU configs... one would think it has to be faulty audio equipment, except that it works fine with two different laptops.

I tried a bunch of things to fix the audio, but the thing that worked excellently was a Topping HS01. This adds a new power/ground signal to the USB, direct from a wallwart, and it does some other isolation stuff that is of smaller importance. You can also get an iFi iDefender+, which is the same but red, and might be like one yoctometer lower end.

I also tried a cheap PCIe USB extension, since it was cheaper, someone said it worked for them, and I wanted more USB-C ports anyway, but it didn't work for me.

This is not how my first PC build was meant to go, but I'm glad to have frontloaded the pain. As long as Amazon doesn't get angry at me for refunding too much XD.