- Joined
- Jul 2, 2024
I highly recommend getting ECC UDIMMs. At least where I live I can get a stick of KSM32ED8/32HC for roughly the same price.Motherboard: MSI PRO B550M-VC WIFI Micro ATX AM4 Motherboard ($169.99 @ Amazon Canada)
Memory: Corsair Vengeance LPX 32 GB (4 x 8 GB) DDR4-3600 CL18 Memory ($117.99 @ Amazon Canada)
The default timings are very bad, but you can lower them a lot. I've got 4 sticks (should've gone with 2 in hindsight) at IIRC 3666MHz 17-20-20-38 @ 1.4v.
I probably could push it harder if I used only 2, and maybe got a new PSU too.
With ECC memory you will detect much sooner is some component is shitting itself. My previous PC would freeze at random times and I only realized my RAM became unstable (probably aging PSU) because I happened to have a disk with ZFS which was getting more and more checksum errors, and even then only when the amount of errors became ridiculous.
A big issue with non-ECC RAM is that corruption is entirely silent. If e.g. a HDD with a somewhat robust filesystem fails it's usually binary: you either have all your data or you lose all of it and need to restore from backup.
Silent corruption can go undetected and hence will eventually spread to your backups, ruining your backups too. Checksums help but will not catch everything since the data might already have been corrupted before you could calculate the checksum.
Of course you'll need a motherboard that supports ECC too. I have an ASRock one and according to
dmesg
it works fine. I've also seen corrected errors with edac-util -v
but only very, very occasionally so far.