Hacker News Reader

Comment by throwawaaarrgh | Hacker News Reader

"Why pay more for something good when you can pay less for something not as good?" - every consumer

I believe this is more about Intel who for probably a decade have sold consumer CPUs with memory controllers capable of using ECC memory but either disabling the feature in the chip, or more recently with Alder Lake, locking the feature behind an enterprise motherboard chipset despite the chipset not being relevant to ECC support.

Having to pay a 20% premium on RAM for the stability of ECC is one thing, having to pay double, triple, or quadruple the system price to disable arbitrary locks is another.

With AMD that limitation is somewhat lifted with normal consumer CPUs and quite a few motherboards having support. Unfortunately the RAM itself has very poor availability. Very few manufacturers offer it and it tends to stick to the standard speeds without any of the overclocked-from-factory parts available with ECC. That last bit is strange because ECC is reportedly very good as a validation tool for overclocking builds so it seems like the RGB-lighted market could end up valuing it as a high-end feature too. One of the RAM manufacturers should rebrand ECC to "Extreme Clocking Capacity" and start selling it as a feature to differentiate their RAM from the competition.

> Unfortunately the RAM itself has very poor availability.

That doesn't match up with my experience, even though I've seen it mentioned several times by people as if it's the case.

For example, when I went looking for ECC UDIMM sticks for my Ryzen 5600X build a year or so ago, I went looking at the website of my local computer supplier.

Many potential options available:

https://www.scorptec.com.au/product/memory/ecc-&-registered

Now, there aren't as many different options as for non-ECC stuff. eg:

https://www.scorptec.com.au/product/memory/ddr4 (many pages of sticks available)

But it's not like theres any kind of availability problem. And the Kingston ram I bought happily overclocked to 3200MHz without any effort on my part. Using an ASRock B550M Pro4 motherboard for reference, if that's useful.

It will depend on where you are. I'm in Europe and it used to be that the usual retailers had no DDR4 ECC options at all. Now there are 1 or 2 available and still no options for DDR5 ECC. Looking at those pages your suppliers seem much better than what I've found so far and yet also have no DDR5 options. Meanwhile DDR5 non-ECC is very broadly available and DDR4 non-ECC has great and mature options.

> That last bit is strange because ECC is reportedly very good as a validation tool for overclocking builds

Well, yes, but it's not the only tool. In practice you'd use some validation software suite to find a reasonable stable configuration and afterwards prey. Overclockers, who rather than spending extra money on a CPU which assuredly delivers requested performance stress their hardware beyond specifications, are least likely to pay for the extra bits.

If non-ecc RAM is good enough for most fault sto only cause noticable problems rarely, its possible to sell slightly less perfect modules than with ecc, where even otherwise unnoticable problems are detected and reported. The ratio of modules returned to the manufacturer is probably higher for ecc than non-ecc. Unless ecc modules are already better selected than non-ecc at the manufacturer.

Who want overclocking the ECC RAM? You choose ECC because of the stability over performance. Why do you want to reduce the stability?

Because given the same chips you can push ECC RAM to speeds beyond what non-ECC RAM can do before corrupting. You can push beyond what the chips are capable of and rely on the error correction to keep your system stable.

ECC in common desktops/workstation/servers will correct all single bit errors and detect all 2 bit errors.

So sure you can run dimms every so slightly faster and fix the occasional single bit flip, but even a single double bit flip and an process or your kernel crashes.

Seems much saner to go for a safe, robust, and reliable system at standard clocks in ECC instead of trying to get slightly more performance which increases the chances or errors, corruptions, crashes, and shorter service life.

How many seconds do you need to think about this to come up with counter examples? There are entire businesses based on selling things which are better or helping buyers find them, suggesting that the rush of perceived superiority isn’t warranted.

Some people can’t afford to buy anything but the cheapest but in most cases it comes down to not thinking that they use it enough to matter (e.g. the common homeowner advice to buy a cheap tool the first time & replace it better if you break it), not having a good way to tell whether something is actually better, or the market being such that there is a huge gap between the product bands. ECC falls into the latter two categories: the average buyer isn’t familiar with the issue and probably thinks the outcome would be a crash rather than silent data corruption, and Intel’s market segmentation means that you don’t have a choice in the consumer space and have to move into far more expensive and limited categories. It’s not reasonable to say price-sensitive consumers are the problem when that’s also saying “stop buying laptops”.

> when that’s also saying “stop buying laptops”.

My 4 years old laptop has ecc. To see if anything has changed I just checked the recent models. The manufacturers website didn't provide any possibility to filter for ecc or mobile Xeon, but an explcit text search returned some results. Ecc still exist in laptops, its just a bit difficult to find.

On the plus side, 19:10 screens are back.

How much more for how much better?

Anyway, try to take the technical specification of any consumer oriented motherboard and discover if it allows ECC.

Why pay for something I don’t need when there’s the option to not pay for it? - me

Just yesterday, I bought an extra 64GB for my home Linux PC. I absolutely couldn’t care less about it crashing or calculating the wrong result every blue moon (in practice: never), but I did choose the RAM sticks that were $10 cheaper.

That's totally fine. The problem is people that do need it not having the option to pay the extra without getting a totally different "workstation-class" computer.

The problem is the lack of choice, not your choice. If you were willing to pay $10 more for ECC DRAM, there would be no way to use it.

Plz tell me your IP, I'll block torrents from you :D

torrents have checksums on the blocks, which your client already checks, because the internet doesn't guarantee packets arrive without corruption. it doesn't matter where the corruption comes from.

I’d go dither than that and claim that almost everything of importance has checksums: Google Docs, Google Sheets, Git submissions, all my important web accounts, the traffic with my bank website and so forth.

When I look at my daily home computer usage, it’s remarkable how little I calculate on my local computer that’s actually with protecting.

Yes, I know. It was a joke :)

How many people complains about memory corruption? This is a ridiculously rare issue, which impact is minimal, if not null, for so many people. There's an actual cost to provide ECC memory though, there's actually more hardware necessary to cover it. Obviously if more would need it, it may lower the cost because of the scale, but it will still be more. For some it's important, and for sure they will pay for it, so let them pay for it...

It's not like it wasn't available, Linus just forgot to put a reminder to buy some once available (backordering them would have been even more prudent).

Ridiculously rare? That's why CDs, DVDs, blurays, caches, SSDs, spinning rust, etc all have ECC, right? Even ancient file transfer protocols like xmodem and zmodem has error correction.

Bitflips aren't uncommon, it's just compression makes them much more noticeable. Without compression users might just see an occasional application crash, and just relaunch the proces.s Seems common for people collecting photos, music, etc for years go back and find some corrupted. It's hard to say exactly what happened, but with compression a bit flip it turned into a seriously corrupted file.

Keep in mind that, radiation caused bit flips are rare, some failure in the chip, pin, dimm, dimm connector, motherboard, CPU pin, CPU are much more common. It's MUCH nicer to see "error on dim X, row Y, column Z" then a randomly crashing machine. Even Linus sounds like he spent hours tracking it down, and thought he was hitting a kernel bug.

Isn't it worth 1/9th more memory chips to make the system robust in the face of wide variety of errors that can corrupt memory, which can lead to corrupted storage?

More like "Hey, nice mission-critical database you have, deep-pocketed enterprise customer. It would be a shame if a bit-flip happened to it".

Why buy mac pro for $5000 when you can buy trash can for $5. Even if it's not as good, it's much cheaper and does not look much different anyway.

"Why are all products so unreliable now!?!?" -- Also every consumer, when their appliance breaks one day out of the 2 year warranty

While I think that is true, do we have much choice in this case?

The computing world is full of more and less featured products.

Reply on news.ycombinator.com