4x DIMMs of the same model, chips are 2x Micron and 2x Hynix.
Was running just fine for more than a year.
Top-quality PSU, well-ventilated case.
Always used at stock voltages, permitting for “Auto” in UEFI settings.
What has happened:
Noticed some new errors in dmesg and occasional forced reboot under Windows.
As the system was unchanged, I have decided to proactively update BIOS…
There was a new version with several vulns fixed, so why not?
Issues became worse. Memtest86 revealed multiple memory errors. Downclocking memory to XMP profile helped to reduce the quantity and frequency of the errors, but hasn’t fixed the problem completely.
Current state of affairs:
Leaving 2x DIMMs out and running at XMP resolves errors completely.
Swapping for another pair of DIMMs resolves the problem as well.
Using either pair of memory slots (1-3 or 2-4) does not change anything.
Questions:
What is at fault here?
How do I unambiguously identify the culprit and persuade supplier that the component is indeed faulty?
did you ever try plugging it all back in?
I’ve had a few cases where a stable system started wonking out, and reseating the ram a dozen or so times fixed it. DDR4 doesn’t seem to seat very well, I had trouble getting it stable in another AMD system, which again came down to reseating a few times.
Can you try the RAM at the lower speed, but manually loosen the timings? I know the 3000 series should have a better memory controller than previous generations, but it might be having problems running all four sticks. I know when I tried my R7 1700x it didn’t like and that’s what I had to do. I upgraded to a 5600G last year and just swapped the memory over and can get all 4 sticks of 3200MHZ RAM running @ 3600MHZ, where it would only work @ 3000MHZ with loose timings on the 1700x in the same motherboard.
There’s also been a little bit of talk of AM4 silicon dying, and the R5 3600 in particular was brought up recently.
I cleared UEFI/BIOS so SOC voltage is stock. I haven’t messed around to overclock the iGPU yet. It’s using 4 x 8GB sticks like the ones here:
All CPU’s with an integrated memory controller have an officially supported memory ‘speed’. I looked up similar sticks to see what looser timings they were running at 3600MHZ, and plugged in those numbers. I had considered buying new RAM, but after watching several videos with similar conclusions, it wasn’t worth the cost of getting lower latency RAM vs running 3200 MHZ RAM at the higher speed with looser timings. There’s a bunch of budget oriented YouTube channels out there that have tested this and my results fall in line with what I saw in several videos.
I haven’t had any stability issues. The memory controller for 5000 series Ryzen is substantially better than the first generation. The big uplift in single core performance made the most noticeable difference for me, but tweaking the memory made a difference for my main use case, which was for playing 7 Days To Die. Since I already had everything else on hand, dropping in a new processor given the cheap price was a no-brainer.
I don’t know how they are nowadays, but in the past ASUS boards have been known to be a bit picky. On top of that, RAM that isn’t bought in a kit, even with the same part number, can sometimes use different RAM chips under the heat spreaders. Add in the fact you can’t run faster than the slowest chip, and it can be more difficult to get that much RAM running at full speed.
You could try what I did and manually set the speed and the timings that would be appropriate for that speed. If that doesn’t work you could loosen the timings a bit further and see if that does it. Especially with that many cores, usually the increased bandwidth greatly offsets the extra latency of looser timings. If the machine had been running with 2 sticks first, there could be dust in the RAM slots. I had to clean mine out really well.