ECC and DDR5
etbe.coker.com.au
etbe.coker.com.au
DDR5 on-die ECC detects and corrects one-bit errors. It cannot detect two-bit errors, so it will miscorrect some of them into three-bit errors. My understanding is that the on-die error correction scheme is specifically specially designed such that the resulting three-bit errors are mathematically guaranteed to be detected as uncorrectable two-bit errors by a standard full system-level ECC running on top of the on-die ECC. But I've never found a real authoritative reference that directly says that.
One nightmare scenario is a 3-bit error being detected as a correctable 1-bit error upstream, resulting in data corruption. I agree it’d be really nice to know that could never happen!
Wonder if they could be used to introduce “correctable” errors…
Is anyone here good at navigating patent records? I would be surprised if there isn't something to be revealed in a patent but I don't know how to find it.
But it's definitely possible, I have an ECC setup on my desktop and it's caught a memory error or two over the last few years.
Or just get an AMD Ryzen.
I have DDR4 ECC memory on my home server. Last week I got a crash... logged into my BMC and saw that one of the modules had "uncorrectable ECC". Then I looked over the logs and saw about 200 or so errors for the exact same module but "correctable ECC". They would happen over a short period of time followed by 4 months of no issues. Thanks to ECC and logging it was quick and easy to catch the issue right on the first occurrence, no data corruption. I have 4 modules in total, and only this one was having issues, so clearly a hardware fault.
Now imagine it was a regular non-ECC memory. There won’t be any crashes due to uncorrectable ECC; another 3 or more years would pass before corruption would become frequent enough for the issue to be noticed. At that point, there would be hundreds of corrupted files.
And 9 times out of 10 (probably much more) it will be faulty storage, not RAM.
I'd say you are way overestimating impact of no-ECC.
Single digit - sometimes none - errors per year in systems with consumer amount of RAM.
One time I took a working workstation from my home to my parents' home to give them. When I booted it up it gave a lot of strange kernel errors which I googled and discovered were reports of ECC correction. I reseated all DIMMs and it was fine. If that gave errors a few times a year instead of thousands of times on the first boot I wouldn't have noticed.
Some crappy flash drives might not do this, though. Seems to behave like a floppy disk at times.
I've seen this in DC grade SSDs, NAS grade hard drives, and SSDs sold in high end laptops.
I don't think so. I've got my important data backed up on offline HDDs and SSDs and my most important data lives on a RAID system (ZFS) on a Xeon with ECC RAM but...
... there are simply a huge lot of checks everywhere making it so that bit flips on non ECC systems simply aren't that serious.
We live in a world of checksums and retries at every layer and distributed systems that are just about everywhere.
I went on for decades without ECC memory and nothing bad happened. Yes, I had the occasional bitflips. I did found out faulty mem using memtest (or weirder stuff, like some RAM stick identical to the other three, same chip everywhere, same sticker: but reported by memstick as being some other mem).
But I don't think that the "one bitflip cascading into corruption of your life's data" is something necessary to freak about.
It's not happening. Never happened to anyone.
If you (or other) are really concerned, you can use my scheme:
- dsc09874.jpg
- dsc09874-b3-73b0835221.jpg
About 1/3rd of all my files are "tagged" this way, directly in the filename: this survive different medium and different OSes.
I then have my custom rsync backuping script doing a dry-run and warning me in a "stop the world!" fashion if it detects a bogus checksum (either on the source or on the dest: during the dry-run any modification to a file with a checksum is verified and any checksum mismatch means "stop the world").
I then also have, in addition to my custom backuping script, another script that allows me to verify either all files with that naming convention or a percentage I pick of those files (so basically some random sampling).
It's not mutually exclusive with ZFS / a system with ECC RAM.
> but not all our files are replaceable, and you forever and always might have corrupted files you are unaware of that got written during this time
I hear you but... How much do you really need, years later, a file that you haven't needed for years? That one file is unlikely to be a big loss.
Now I'm not saying you're wrong: I ended up with my naming scheme (appending a partial cryptographic hash to the filename) after finding a series of about 20 pictures that were corrupted. And on all the backups I first checked, they were also corrupted. I ended up finding a very old data CD on which that series of .jpg files were fine. And I decided "never again, I want such SNAFU to be detected in the future".
But truth is: with 40 000 pictures, life without those 20 pictures would still be fine.
Normal good DIMMs, at least when new, should not have errors more frequently than one error per many months.
Frequent errors may appear with memories that are not seated well in their sockets, or which are old, at least several years old.
Frequent errors may also be caused by more general computer problems, like a bad power supply unit.
(Correlated bit-flips are likely from a manufacturing error... or, at least in one case, from excessively radioactive ceramic packaging.)
The more atmosphere you have to randomly absorb high-energy photons, the better.
Higher density memory drops fewer electrons in each well, which means a lower-energy photon can change the state.
Corolary: submarine datacenters are much better than orbital datacenters. Cheaper to cool, construct, shield and access.
This gets said a lot but with little evidence. The mario speedrun bit flip has been replicated with marginal connector insertion.
Whether it is the primary source of errors, depends on the design of the DRAM die.
In a badly designed DRAM there could be many other error causes.
The designers of a DRAM certainly attempt to minimize all the error causes that they can control. If they succeed, the cosmic radiation would remain the primary error source.
Without access to internal documents of the DRAM vendors, we cannot know if they have indeed minimized all other error causes.
In modern DRAM chips, a major error cause can also be the disturbance of some memory cells when some other cells are accessed in their neighborhood, which is the basis of the Rowhammer attacks, but which can also happen during the normal use of the memory, when certain access patterns happen by chance.
The DRAM vendors do not provide details about the behavior of their devices, with the hope that this makes harder the task for someone who wants to design a variant of the Rowhammer attacks, but this also makes impossible for the owner of the memory to predict whether an application program will not perform by chance an access pattern that will unintentionally flip some memory bits, which without ECC will not be detected and corrected.
Avoiding returns for ECC DIMMs is a harder challenge when they are used in an environment where the server will have a management system that tracks errors over all time and when there might be a network management system tracking all errors over all systems.
Avoiding returns for non-ECC DIMMs is easy when a few errors per month won't be noticed by most users who assume that Windows always crashes anyway.
Has any DRAM company tested chips and made non-ECC DIMMs from the ones that have lower quality?
As an aside are the 2^N*3 sizes of DDR5 DIMMs (like 24G and 48G) made from chips that had errors in one section and got reconfigured to have 3/4 the capacity to not use the bad parts?
Are you disputing that the Earth's atmosphere absorbs and reduces the amount of ambient cosmic radiation? Or where is the skepticism coming from?
Muons are nasty little buggers.
Yea, except datacenters also need connectivity. And submarine connectivity, while not particularly expensive, is quite vulnerable - even more so than on land.
If you only need 100M deep water then you don't need to go far offshore, there are places where it's only about 1KM out. 12M of water is used for storing spent nuclear fuel assemblies and less than half of that is needed for protection so 5M should be adequate for stopping cosmic rays and that depth is common in harbours for recreational boating, such harbours shouldn't have issues of cable breakage.
But the real issue for comparing submarine and orbital computers is cooling. Space is cold but doesn't allow easy transmission of heat.
I would think most servers and workstations would be RDIMM (Registered DIMM) by now and consumer stuff uses soldered down memory. Memory failing because it's old is definitely a thing, and very possible in this scenario, but I feel like I haven't seen errors due to physical insertion, that were not caught immediately by POST, in years.
But maybe it's just me. Happy to be lucky. :)
The computers in plastic cases can sometimes be affected by appliances with electrical motors that are used nearby them.
The cosmic radiation is attenuated only inside a big building, e.g. one with many stories above you. A really great attenuation is obtained only underground.
But overall, pc cases are probably too thin to appreciably affect the radiation levels. Cosmic rays and their secondary radiation is fine with going through buildings so a 1mm thick case isn't going to do much.
Google and Facebook report a few unreliable CPU cores out of thousands of systems. Meaning maybe a 1/10,000 rate of repeatable CPU errors.
https://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf
Google reports 8% of DIMMs being affected by errors. While the differences in metrics prevent comparing the two things directly it seems that DRAM errors are more common than CPU errors.
With normal sticks I’d not have know that there is a potential issue.
With non-ECC RAM running a memory test and getting any errors means it’s not stable. The memory tests run at maximum bandwidth constantly so a few days of full bandwidth memory testing might produce as many transmissions as your months of regular server use. Seeing bit errors (even if corrected) means something is not right.
The ECC can correct it for a while but I would want to RMA that stick if it’s in warranty.
At some point I should try running it at slightly lower speed i guess. It’s very hard to troubleshoot something that only happens every month or two though
Never seen it fail a memtest but I can see it in the ECC stats since those cover weeks/months.
It was a pain to select a compatible board and source the RAM but glad I did, in June 2024... the RAM prices went up immediately after.
non-ECC should be a kernel taint in Linux. Newer versions of Windows (12?) should also refuse to boot without ECC. It should be mandatory in DDR6. The insanity must end.
That's because the on-die ECC cannot detect errors in the transmission of data from the DIMMs to the CPU. It seems like on-die ECC is meant to "hide" some level of errors to make less reliable RAM chips appear good. I have no idea if that's a reasonable way to do it, but it seems like it could be misused. Like TFA suggests, we need to know how the two levels of ECC interact.
Caveat that I was building a gaming PC at that time, I might have gone differently for a home server. And, this was before the memory market went nuts for unrelated reasons.
The "gaming" DIMMs are available in much greater speeds, which are "overclocked" speeds, i.e. there exists absolutely no guarantee from Intel or AMD that they work at the advertised speed, even if they frequently do work.
However, with such overclocked DIMMs, you do not know how well they work. They might work most of the time, but e.g. in a hotter room they might have an error or two per day without you noticing, especially if the computer is used mainly for games.
For someone doing anything professional on a computer, overclocked DIMMs are not an acceptable choice.
Without ECC, nobody knows if their memory is working or not. Regardless whether it is running a standard JEDEC speed or an overclocked one.
If you buy a complete gaming computer which includes overclocked memory, then hopefully the vendor has done a burn-in and has tested for some time the computer, as sold.
However, if you buy the CPU yourself, neither Intel nor AMD provides any guarantee that the CPU will work with overclocked memory.
It is upon you to test your assembled computer, but true tests would require a very long time and a climatic chamber, to offer any kind of certainty that the combination CPU-DIMMs works at the desired speed.
If you have ECC memory, you can try to overclock it and at least in this case you will know for sure if the memory works at the higher speed, or not.
Overclocking ECC memory is actually a method frequently used to verify whether the hardware ECC support and also the operating system EDAC driver work OK.
There's no guarantee that it works with standard clocked memory either.
If the CPU does not work at that speed, you are entitled to a replacement or a refund.
For instance, AMD Ryzen™ 9 9950X3D is guaranteed to work with DDR5-5600, but with nothing faster. No AMD non-server CPU goes above this limit for DDR5.
Some of the Intel CPUs are guaranteed to work with faster memories, e.g. Core Ultra 9 285K with up to DDR5-6400, and the very recently launched Core Ultra 7 270K PLUS with up to DDR5-7200.
This is not true[1]. These DIMMs are not guaranteed to work with all Threadripper CPUs, but they are guaranteed themselves to be stable at their advertised frequencies, which are higher than JEDEC. (And from my experience, 9000-series threadripper is perfectly happy to drive 8-channel 1Rx8 at 7200MT/s.)
AMD (and possibly Intel, I don’t have any experience with their recent offerings) also lets you overclock memory completely on WRX90. So if you buy JEDEC sticks, you’re welcome to try to push them as hard as you can.
[1] https://v-color.net/products/ddr5-ocrdimm-amd-wrx90-workstat...
Without the reported ECC on Infinity Fabric I would have had no idea that the errors were happening and affecting performance.
The main reason there aren’t more options here is that a) high-frequency memory usually isn’t sufficiently stable, unless binned, and b) there’s very, very, very little demand.
[1] https://v-color.net/products/ddr5-ocrdimm-amd-wrx90-workstat...
It is if you're doing something where corruption is detectable after the fact, and happens rarely enough that redoing affected work adds less than 3% overhead.
Without ECC, you need a custom operating system, which will compute some error detection code, e.g. a CRC, for each read-only memory page, i.e. including all code pages and constant pages, and all cached but not dirty file system pages, and which will check the CRC codes periodically in the background.
That would still leave the read-write data segments unprotected, though on those some errors may be benign, if the locations are overwritten later without being read again.
From that time, I have a few old computers with 64 GB or 128 GB of DDR4 ECC memory.
Unfortunately, after the passage to DDR5, DDR5 ECC UDIMMs have become hard to find, and even when you could find them, the price difference could be as high as 30% to 50%.
Aliexpress had $50 single socket motherboards for Xeon E5-26xx CPUs and $150 dual socket motherboards last time I checked that use DDR4 RDIMMs.
Before the prices went up due to AI foolishness DDR4 RDIMMs were $1.20AU/G means means something like $US0.80/G.
On-die ECC was added in DDR5 because the error frequency has increased too much.
Its only effect was that now DDR5, as seen externally, has a reliability similar to that of the older memory generations. On-die ECChas not brought any serious improvement in reliability. It has just prevented the degradation of the reliability.
The errors generated internally are caused mostly by the ionizing radiation from the environment, which discharges the storage capacitors.
However there are many errors that appear during the communication between memories and the CPU, which are caused by the electrical noise from the environment.
These external errors are much less likely to appear for soldered DRAM. Because of this, for soldered LPDDRx memory it is less important to have end-to-end ECC than for socketed modules, which are much more vulnerable to noise.
The DDR5 standard has some options for using some error detection for the communication link, but it is impossible to know whether a given computer or motherboard has implemented such options or if they are enabled by the firmware when the motherboard has the physical support.
The only way to be certain that the memory works fine is to use ECC that covers completely the circuit from the CPU memory controller to the values stored in the DRAM and back to the CPU memory controller and your operating system has the appropriate device driver for obtaining the error reports.
The imperfect contacts are equivalent to adding series resistors, possibly in parallel with parasitic capacitors, on the link traces, allowing a greater amplitude for the noise pulses.
When you do not own a great number of servers, you cannot provide anything else except anecdotal evidence.
Because the vast majority of PCs do not have ECC memory, even the companies that have a great number of PCs have no idea about the reliability of the memory used in those PCs, because it is very difficult to distinguish memory defects from the huge number of software bugs.
Moreover, the frequency of memory errors is proportional with the amount of memory you have, so those with tiny amounts of memory, like 8 GB, are much less likely to encounter memory errors than those who have 32 GB or 64 GB of DRAM in their PCs.
I have tried to always use ECC memory in my computers, whenever possible. Even my laptop is an older Dell Precision, with ECC memory.
But I can also offer only anecdotal testimony.
I have also seen cases like that mentioned in TFA, where the existence of ECC allowed to discover that some modules were not seated well in their sockets, so reseating them prevented the reappearance of errors.
I have also seen a case when a laptop was not used for a long time and it was stored in a rather humid place, so the contacts in the SODIMM sockets had oxidized, which resulted in frequent memory errors. Scrubbing vigorously the contacts and reinserting the memory modules solved the problem.
I have also seen many cases where certain memory modules degraded after many years of use, e.g. 5 or more years of 24/7 use, and they began to have frequent errors, e.g. multiple errors per day, even if when they were new the error rate could have been of one error per year or even less.
In such cases ECC was extremely useful, because the DIMM that had become worn out could be identified and replaced and the server could work fine some more years.
Boy did I learn that the hard way. I found an ASRock server motherboard that fit my needs for a NAS perfectly. I just assumed all ECC memory were registered so I didn't even bother to check. I had already put everything together minus the memory. I had actually purchased the memory and had to return them because they were RDIMMs and get the ECC UDIMMs for quite a bit more. Luckily this was just before the AI induced price spike.
I hope that my blog posts on these topics will save some people some money.
> We need ECC RAM to be more widely used. Ideally we would have some government action to force this given the ongoing cost to society in corrupted data and lost time due to RAM hardware errors. I think that at minimum we need sufficient taxes on non-ECC RAM (and EC4 RAM for DDR5) to make it more expensive when bought new than ECC RAM.
ECC RAM is not needed for every application and mandating it for every gaming rig and machine used to run Chrome will just increase prices unnecessarily. 99% of the time it's also not a "cost to society" but a cost to whoever owns/uses that machine. If they want to eat that cost, it should be up to them.
The price increase will be very small if manufacturers are no longer able to segment the market to make a higher margin on those who need it. No one wants even games or web browsers to crash randomly.
> If they want to eat that cost, it should be up to them.
They need to be given the choice. Most people do not know they even have the choice, and they are not offered it at a reasonable price.
A cost to people in general is a cost to society.
I'm not sure making it illegal to sell consumer grade hardware is the boon for the people you imagine
https://ripe92.ripe.net/programme/meeting-plan/sessions/112/...
Restricting risky items is standard practice for all governments. All the "use by" dates on food etc are from government regulation. The current experiments with "raw milk" shows that you can get away with flouting food safety regulations and not have immediate problems.
AMD doesn't have this segmentation. Where's the hoard of people building systems with ECC ram? At best you have a few people on HN saying they're doing it, but in practice nobody actually does it.
More people would use them, if they were available. RDIMM's do not work on normal consumer boards, so obviously people won't be using them.
Whether a filesystem could be designed to operate in the face of random memory changes would be an interesting computer science research topic. But for getting working systems having RAM just not fail is a better option.
The rest of this article is great. This, however, is absolute garbage and nonsense. The last thing you want is more regulations and "government certified systems". Take a look at the bullshit with age verification bills and operating systems.
I noticed the difference because in a Hardware forum someone wrote that the framework desktop does not have ECC while an HP G2 Workstation also featuring the Ryzen 395 Max but in the ECC supporting variant paying a premium of +600 bucks.
Why is it so often Australia with the "We must force unnecessary things upon people" attitude?
That's what we mean when we say that people with your point of view are promoting nanny-statism. It amounts to infantilization of adult consumers and subsequent disempowerment.
Occasionally such legislation can be justified, as when genuinely hazardous products are involved, or products whose use involves wide-ranging externalities, but usually not. People can and should be educated -- and expected -- to make their own decisions about things like whether they need ECC RAM.
I think my knowledge of the topic is pretty good but I learned from this discussion that almost all AMD CPUs support ECC RAM and you apparently just need a motherboard to match. When accompanying people who were buying "gamer" PCs (after they rejected my recommendation for a second hand workstation grade computer with ECC RAM) I never witnessed store staff saying "hey you could get a motherboard with ECC support".
Just because you can allow it, doesn’t mean you should. I don’t want to live in a world where you need a PhD to avoid being scammed by increasingly sophisticated ploys. Data isn’t free anymore and the right questions are becoming harder to ask.
Ridiculous. Access to knowledge has never been as widely and freely available as it is now (no thanks to the nanny-statists.)
The government imposed a 12% GST on all computer gear just because they felt like it. Would making ECC RAM GST exempt satisfy you?
So you think its realistic to expect the average consumer to assess the pros and cons of using ECC RAM?
We could solve this by better information which addresses your concerns about it being nanny statism. We would need it to be something that consumers understand - e.g. We could require all computers without ECC RAM to have a warning label saying something like "this computer uses cheap memory so is prone to crashing and losing your data". I strongly suspect the impact of that would be very similar to that of just banning non-ECC RAM.
I would also argue that having reliable RAM is a reasonable extension of existing legal requirements for consumer products: e.g. that they are of satisfactory quality and fit for purpose (to use the current UK terminology).
I just do not think that banning non ECC is an example of nanny statism.
Forcing the vendors to sell only products that do what they are advertised cannot be considered in any way as "forcing unnecessary things upon people".
Any computer without ECC memory is a computer that does not compute from time to time and you have no idea when that happens.
Originally, any IBM PC or compatible had memory error detection.
All this problem of the computers without ECC memory has been created by some evil Intel employees, who have conceived around 1994 a scheme for market segmentation between "Pentium" and "Pentium Pro".
So they removed memory error detection support from the Pentium chipsets (Intel Triton), so that this would remain a marketing feature exclusive to the "Pro" Intel products.
Obviously the removal of this feature from "consumer" products was done without any price reduction, which made happy also the memory vendors, because this trick increased their profits.
So in the next years this market segmentation method became the "standard" business practice, preying on the naivety of most computer users.
A consumer protection law should force that any computer or motherboard must support ECC, which adds a completely negligible cost.
The end users may buy non-ECC modules, if they choose so, but non-ECC modules should be sold like cigarettes, with a visible advertisement that using such modules may cause unpredictable errors in your computer.
So are the computers with ECC memory. It's not an absolute protection either.
I hate to break it to you but the computer I had in the 80s didn't come with ECC. Non-ECC computing was a thing WELL BEFORE Intel tried to think of market segmentation. Revisionist history is not welcomed here.
The DDR4 ECC RAM has 72 bits total and 64bits of data which means the same 9/8 ratio of data storage to usable data as the original PC/XT.
The Z80 and 6502 based computers in the 80s typically didn't have parity RAM. The Amiga and Macintosh were higher performance systems in the 80s which also didn't have parity RAM.
We could and should have gone straight from parity RAM to ECC RAM for everything.
Absolutely all IBM PC originals or clones until those with 486 and with the early Pentium variants supported parity on the motherboard.
The user could choose to not install the parity chip (or when the first memory modules appeared there were also slightly cheaper modules without the parity chip), but that was the user's decision, all the motherboards supported parity.
On IBM PC compatibles, the lack of protection against memory errors had never existed before Intel segmented the market between "non-Pro" and "Pro", by launching the Triton chipset, which no longer supported parity memory.
As long as the memory modules were byte-wide, only parity could be implemented. ECC memory could appear only later, after the passage to wide memory modules.
it is now $500
when the AI/datacenter bubble finally bursts we are going to have some amazing hardware deals
Of course an AI is going to tell you AI isn't a bubble.
Please stop repeating this old myth, it is - and always has been - BS and needs to die.
See for example: https://kldload.com/zfs-wiki/myths
Claiming that every issue you've experienced is from seating errors? Come on man, at least deliberately reproduce that scenario once. I've tortured motherboards for fun and never managed that on purpose. We did pry every cap off of an old K6 board while it was on and... it just kept running. It didn't come back after we turned it off
The amount of time it takes to reseat memory, even if the case is already open and ready, is also more than enough time for it to cool back down.
If you have decent memory running at what it's rated for and you're seeing more than a couple per year, good odds are it's getting too hot or the load line calibration is not aggressive enough.
E: I have never, and I mean literally never, seen age or abuse play a factor.