4
PaaPsik
5y

how many of you developers use ECC ram? any noticable difference? Any advantages?

Comments
  • 5
    why would you....? ECC is meant for servers, 24/7 running machines. They require more energy (hence shorter batt life) and they are fucking expensive. And seriously, how likely are you to catch a freely flying neutron with your DIMM? What impact would that cause? BSOD? Business impact?

    I think it's not worth it
  • 2
    Yes, I have ECC RAM, and no, they are not expensive. Servers need buffered/registered RAM, but for desktop mobos, you use unregistered because you only have four RAM slots. Also, unregistered doesn't have additional latency. However, it may not be as fast as the fastest non-ECC RAM available.

    Yeah, ECC RAM for servers only.. that's what people say until they realise that their write disk caches are also in RAM.

    RAM flips do happen, read up on the famous Google study.
  • 1
    @Fast-Nop yes, RAM flips DO happen (e.g. rowhammer attack). However I've never encountered any problems regd naughty bits. In theory -- yes, shit can go wrong. In practice - ... I'm sorry but I cannot confirm any of this.

    I did not get your reference regarding disk caches. Is there smth wrong with caching in non-ecc dimms? Also I did NOT say that ecc is for servers only. I said it is meant to be used in servers. Asshole is meant for shit to come out. We all know very well it's not the only way people use it.
  • 2
    @netikras what happens when a bit flips in the write cache? You get file corruption, but you won't usually notice that unless you have files with checksums. If the file system is using automatic defragmentation, that can even happen with files that you aren't using. It's just that after years, you open a file, and it has artifacts or so.

    You'll never know because without ECC, you don't have diagnostics. That's also why you can't say that you havn't encountered problems with naughty bits because that claim lacks any valid basis without ECC.

    And no, ECC isn't meant for servers only because then unbuffered ECC DIMMs wouldn't exist. Actually, the reason not to use ECC came when sales people realised they could sell RAM cheaper if they omit the 9th chip.
  • 0
    @Fast-Nop, @halfflat - for how long have you been using ecc? And how many errors has it approx corrected so far? I wonder what's the error rate in PCs.

    Also do you live close to radiation sources? Airports, powerplants, chernobyl, etc..?
  • 0
    @netikras For about 8 years, and the WHEA hasn't reported any error so far. However, I have my BIOS set to a mode where the chipset does more than just ECC. It cross-connects the RAM somehow so that autocorrection is possible, I think it was called "chipkill" back then, and it does an automatic background scrub through the whole memory. Yeah, it was one of the most expensive desktop mobos back in the day.

    So Windows probably wouldn't notice an error because a single bit flip is likely to be corrected before Windows gets around to it.
  • 0
    @Fast-Nop ECC DIMMs have an ability to report corrections to OS. At least Linux (RHEL) and various flavour of Unixes do report ecc reparation triggers in dmesg (OS kernel log). I could only assume Windows has this ability as well.

    Over my sysadmin career I've seen 3 or 4 reparations reported. If you do not know how many flips has your ECC recovered, how do you know it's helping? Or even working at all...?
  • 0
    @netikras back then, I have checked that it is activated, dunno what the tool was. Whether it's really working - I don't think it's actually testable without injecting an ECC error, and I wouldn't know how to do that without some sort of specialised debug RAM hardware.

    While I'm at it, maybe the copper cooler sinks around the RAM bars also provide some level of protection against some kinds of error causes.
  • 0
    @Fast-Nop So simply put you do not know whether ECC feature of your dimms has been of any use for you nor how much errors has it saved you from. It could be 0 you know..
    Which makes it kind of unfair to blindly claim that ECC is of any actual use for PCs, don't you think?

    You use ECC - you haven't noticed bitflips
    I use non-ECC - I haven't noticed bitflips as well.

    Neither in disk write cache, nor in apps, kernel, wtv.
    So once again... Is it worth to get an ECC DIMM over non-ecc (with lower latency, lower power consumption) for a PC? Or does it actually make more sense to leave ECCs for devices that are used vastly, like 1000s of customers facing servers? because.. you know.. an accidental bit flip would not assign a $100k loan for your bank account instead of $2k topup :)

    From what I see I can conclude without a doubt: No, it's not worth it for PCs.
  • 1
    @netikras if building a used pc it acutally makes even more sense - older xeons + ddr3 used ecc ram is far cheaper than buying traditional ram, because theres an abundance of server ram on the market
  • 0
    @pionell be careful - desktop boards only take unbuffered DIMMs while servers usually have buffered ones, and they are not compatible.

    Reason is that in a desktop board, you only have four slots, so no buffering. With 24 slots or so on a server board, unbuffered wouldn't work.
  • 1
    @Fast-Nop if you're buying a xeon you most probably get a server board of some sort along with it. But thanks for the tip
Add Comment