How important is ECC really?

[ArcticApe]

Hello!

I'm speccing out a home server, preferably using a Ryzen 1700x I found for dirt cheap on local equivalents to craigslist. Ive done some research but the topic of ECC always comes up. Some say building a server without ECC is pointless as your entire setup is shit poopoo without, some say its nice to have but not a dealbreaker. I'm a bit lost. I'd only be using the server to store video and text documents, could they suddenly just be rendered useless and gone because I didnt shill out 3x the amount of money to get ECC memory?

[Radium_Angel]

9 minutes ago, ArcticApe said:

I'd only be using the server to store video and text documents,

You won't need ECC then

[RONOTHAN##]

The reason ECC is recommended for a NAS is that the RAM is usually used as a cache, so in the rare event of a bit flip during the brief period between when the file is being stored in RAM while being written to the HDDs it can cause the file to be written improperly and be corrupted on the server. 

 

That said, it is incredibly rare that happens assuming your running your RAM at JEDEC specifications, so it's not really that big a deal for a home server to have ECC or not. If it's similar cost to go ECC than to go for non-ECC, yeah spend the extra little bit for it for peace of mind, but 3x the amount I wouldn't. 

[SupaKomputa]

For home server, yes it's kinda poopoo. Ryzen doesn't support full ECC (registered) and not all motherboard support the feature.

For full ECC support, i suggest getting an old Xeon, they are cheap, you can get a full system (cpu, mobo, memory) for less than $100.

[ArcticApe]

Just now, RONOTHAN## said:

The reason ECC is recommended for a NAS is that the RAM is usually used as a cache, so in the rare event of a bit flip during the brief period between when the file is being stored in RAM while being written to the HDDs it can cause the file to be written improperly and be corrupted on the server. 

 

That said, it is incredibly rare that happens assuming your running your RAM at JEDEC specifications, so it's not really that big a deal for a home server to have ECC or not. If it's similar cost to go ECC than to go for non-ECC, yeah spend the extra little bit for it for peace of mind, but 3x the amount I wouldn't. 

What if I were to use an NVME or sata SSD as a write cache, for example zlog or whatever they call it in truenas, would that basically get the same effect as ecc, or circumvent the risk?

[SupaKomputa]

1 minute ago, ArcticApe said:

would that basically get the same effect as ecc

No. ECC is error correction, got nothing to do with cache.

 

[Electronics Wizardy]

20 minutes ago, ArcticApe said:

I'm a bit lost. I'd only be using the server to store video and text documents, could they suddenly just be rendered useless and gone because I didnt shill out 3x the amount of money to get ECC memory?

Your desktop probably doesn't have ecc. How often do you have issues? Normally its pretty well and if there are large amounts of issuesses like crashing.

 

6 minutes ago, ArcticApe said:

What if I were to use an NVME or sata SSD as a write cache, for example zlog or whatever they call it in truenas, would that basically get the same effect as ecc, or circumvent the risk?

ECC won't affect this.

 

But if your using ZFS there isn't really a true write cache that would make sense for most uses.

[RONOTHAN##]

Just now, ArcticApe said:

What if I were to use an NVME or sata SSD as a write cache, for example zlog or whatever they call it in truenas, would that basically get the same effect as ecc, or circumvent the risk?

The risks are already super low, so that would be going from something like 0.0001% of a chance to a 0% chance with ECC, for general consumers it's basically unheard of to have a bit flip. It's not till you get into the enterprise space where you have 100s if not 1000s of PCs running that require 99.9999999% uptime where a bit flip realistically becomes a concern. Just use RAM as a cache, it doesn't matter for a home server.

[ArcticApe]

1 minute ago, RONOTHAN## said:

The risks are already super low, so that would be going from something like 0.0001% of a chance to a 0% chance with ECC, for general consumers it's basically unheard of to have a bit flip. It's not till you get into the enterprise space where you have 100s if not 1000s of PCs running that require 99.9999999% uptime where a bit flip realistically becomes a concern. Just use RAM as a cache, it doesn't matter for a home server.

Aight, that gives me a lot of peace of mind. Cheers

[porina]

1 hour ago, RONOTHAN## said:

The risks are already super low, so that would be going from something like 0.0001% of a chance to a 0% chance with ECC, for general consumers it's basically unheard of to have a bit flip.

Anyone have a credible estimated memory error rate? I had a value of 1 per GB per year in my mind, but I can't find a reference to it. I did find some other references that were orders of magnitude worse than that.

 

Keep in mind that even if you have an error happen in ram, it might not affect your running it if happens in a non-critical area, or even unused area of ram. In a game it might cause a glitch that appears and disappears again as quickly in a frame. I'm aware of a game speedrun where it was worked out a bit flip happened causing a fortuitous outcome. Bit flips might also contribute to those random unexpected crashes or errors we can sometimes get, and don't return again. It gets written off as one of those computer things.

[ArcticApe]

21 hours ago, porina said:

Anyone have a credible estimated memory error rate? I had a value of 1 per GB per year in my mind, but I can't find a reference to it. I did find some other references that were orders of magnitude worse than that.

 

Keep in mind that even if you have an error happen in ram, it might not affect your running it if happens in a non-critical area, or even unused area of ram. In a game it might cause a glitch that appears and disappears again as quickly in a frame. I'm aware of a game speedrun where it was worked out a bit flip happened causing a fortuitous outcome. Bit flips might also contribute to those random unexpected crashes or errors we can sometimes get, and don't return again. It gets written off as one of those computer things.

The TickTockClock upwarp?