I had first used regular RAM, then switched to ECC RAM.
All my previous servers before that did also not have any ECC in.
Depending on the power grid, quality of RAM etc., usually 2 months before I started to see random issues showing up. Under Windows systems, you don’t care, as you reboot the machine regularly anyway. Under Linux, where the machine can be running rather long (I had my 386DX33MHz server running some 900 days without a hitch). That stability stopped when the CPU’s and RAM became way faster.
Note that using a good PSU + battery backup as power-grid filter helps a lot too.
So - why are datacenters using ONLY ECC RAM? There must be a reason TBH.
So - why are datacenters using ONLY ECC RAM? There must be a reason TBH.
Different use cases. Datacenters don’t expect downtime, so when it happens, it is highly unusual and needs to rectified immediately. Unless you are a hobbyist or just prone to leaving your computer on all the time, you aren’t going to be using a consumer-grade laptop in the same manner.
Now, would it be nice, if Framework supported ECC RAM on some of its motherboards? Sure, but–putting aside benefits to the user–how much extra would that cost and how hard would it be to source parts?
I just did a cursory search for “SODIMM ECC RAM DDR5” on Amazon and Newegg. There are almost no options available that the average consumer would recognize, what is available is easily 20% more than an equivalent non-ECC RAM offering, and it’s ONLY “unbuffered”. I don’t know if that counts as “server-grade”, but that’s what’s there is.
I have Kingston 5600MT ECC SO-DIMM in my AMD 13, just because I bought it at a time where there was still ECC support mentioned on the AMD specs page for the processor and the FW13 was still under specification embargo. Since then that mention has been removed from the AMD pages and even with trying to manually enable it that feature does not work.
I have lost files to corruption on several NAS systems build on consumer hardware without ECC. Are memory problems to blame? HDDs? Bad powersupplies? Software or filesystem bugs? I can’t know.
But, what I do know is that since I use consumer-level hardware with ECC for my NAS and desktop (since 2013) I have not lost a single file due to corruption. Those systems were all not really that much more expensive than the ECC-less alternative.
Especially since AM4 well working ECC in the desktop space has become really affordable.
I reboot my laptop only for kernel updates, I do not see why I should do that any more often. Since I do mainly consumption on my notebook and any real work takes place on my desktop I live with the risk of memory corruption, but I am absolutely sure that there is a tangible risk to my data without ECC.
A short search on a german price comparison website find Kingston, Mushkin and Micron SO-DIMM with ECC.
All consumer adjacent platform will use unbuffered ECC which is not in itself a bad thing. The “buffering” on R-DIMM is used to lighten the load on the memory controller in systems with 8,12 or even more parallel memory channels. The dual channel in consumer-platforms are easily handled by the memory controllers there.
I had a similar experience to this. My desktop used to give me nonstop problems using 4 dimms. Reinstalling OSes all the time because they would stop booting, or things would stop opening. I moved to using Mushkin ECC memory that is “overclocked” from the factory and I have since had 0 issues using 4 dimms.
Now, I want every device to have it. Just having the feature available can contribute massively to framework’s mission. They talk about enabling software longevity and supporting the devices, but a device without ECC is a massive sacrifice to its longevity.
Think of how many people have issues with their devices and think it’s time to throw them away for new ones? Tablets, desktop hardware, laptops, etc. How many times have you experienced a mysterious issues that cause you to want to get a new device, or start replacing things to try and fix it?
ECC is a massive tool for sustainability. Systems with it not only experience less problems due to memory issues, but after their life as a consumer workstation it can live on as something such as a NAS. People that truly want their data to be intact will always look for something that has the feature.
Doesn’t even have to be that. If you keep rolling the die, you will get days where even hours after boot the computer just blue screens. Maybe it could be blamed on crappy drivers, but the difference between my ECC and non-ECC systems have been the relative lack of surprise crashes on ECC systems.
I want to be able to be on the go without stupid preventable things interrupting my work.
The whole fixation on reliability being the exclusive privilege of some other chassis form factor and haute compute class of users is frankly annoying. Why should laptops not have it? HP even offers Intel Core (not even Xeon) laptops with ECC RAM.
It may be small market, but the people in the particular market are concentrating themselves right at Framework’s doorsteps. It’s not as if Framework needs to expend enormous marketing effort to find enough people to justify making it happen. “Take my money!” ―somebody.
Every ECC memory thread on any forum (here, NAS, etc.) attracts trolls.
“You don’t need ECC.” “Why do you need ECC?”
These trolls need to get jobs or something.
These threads aren’t for you. Go away!
Every PC in my home that I built have ECC memory.
R7-9800X3D 64GB, R7-5800X3D 64GB, R5PRO-4650G 64GB, R3PRO-4350G 32GB, R3PRO-2200GE 32GB, R3-3300X 64GB, R7-2700X 16GB.
I also use server-grade fans (Sanyo Denki) and quality power supplies.
I love my PCs. I don’t care about trolls’ opinions.
SHIPPING with ECC is probably a “nope you wont find it on the face of earth”, but buying your own and chucking them in after you purchase is more or less a “yes”.
I imagine Dell make .. one. They make like, 1,000 SKUs. At least something should fit your bill.
Well it’s not that Framework is not aware, it’s just that the chip (7040 series) simply does not support ECC.
And even on AMD’s desktop platform, only CPUs without integrated GPUs support ECC.
Intel’s ecosystem is a lot nicer. Chipset features, drivers, etc.
Interesting. It does support ECC. That’s new. 3600 speed on 4-sticks though is quite the murder.
I dont think they are particularly worth chasing after. But at least on Digikey they are not too expensive, either.
Sanyo, uh .. Nidec. Sunon, Foxconn. AVC, CoolServer.
For me a nice case is more important. But the modern towers just aint cutting it. Its a lot more interesting (literally) to get a used prebuild and use its case, though for ATX it’s getting more difficult.
I view fans more as consumables. Eventually they die, you swap new ones in.
E3 xeons (now just called E-series) are a bit interesting, as they use the same consumer sockets. In your case (a v2), a 1151, I think. Or a 1150. Or a 1155. Idk.
3rd gen equivalent probably. Since a v3 is 4th gen equivalent.
However in the most Intel way possible, you can only use these with server chipsets (such as C236 for 7th gen. I forgot what is for the 3rd/4th gen).
AMD doesnt really have a equivalent to that. They don’t have (at least as of recent) a entry-level server, like Intel does with the E-series xeon.
There also used to be Cores that are “server” grade, like the i5-750, with 3-channel memory and ECC support. But Intel seem to have changed the wording around that.
It’s all just .. names .. and marketing. It doesnt mean much.
You’re misinterpreting their requirements. Using iGPU as differentiator is incorrect. Regular CPUs all support ECC (if motherboard does), whereas monolithic APUs will support ECC only in PRO form.
For example, Ryzen 7 5700X (Vermeer) will support ECC, but Ryzen 7 5700 will NOT, because Ryzen 7 5700 is Cezanne APU with iGPU disabled. Same problem with Ryzen 5 5600 (Vermeer) vs Ryzen 5 5500 (Cezanne). It gets complicated, so it’s best to check which die is actually used for each model. It was pain to obtain various PRO APUs, especially early on.
I disagree with Intel’s ecosystem comment. AM4 supported Summit, Pinnacle, Matisse, Vermeer, Vermeer-X, Raven, Picasso, Renoir, and Cezanne, now going for 8 years of good support.
I intentionally bought a motherboard with 2 DIMM slots, as I never plan to put 2 DIMMs per channel on any DDR5 system. 64GB DIMMs are now available, so 128GB is possible with just 2 DIMM slots. Neither Micron nor Kingston sells 64GB EUDIMMs yet, but it’s just a matter of time. Also, all EUDIMMs on the market are JEDEC spec, so they’re all maxing out at 5600MT/s with CL46.