Help finding compatible ECC RAM for B450M-A

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
Hi All,

I have an Asus Prime B450M-A mobo w/ a Ryzen 3600. I'd really like to run ECC memory, but I'm having trouble finding 16GB sticks that are compatible.

Ideally I'd like 2x16GB of ECC, with the option to upgrade to 4x16GB ECC.

  • Currently I'm using 1x16GB of CORSAIR Vengeance LPX DDR4 3000 CMK16GX4M1B3000C15. This boots well.
  • My first attempt was with 1x16GB Supermicro (M393A2K40CB2-CTD) ECC DDR4 2666 Model MEM-DR416L-SL04-ER26. This did not boot.
  • My second attempt was with 1x16GB and 2x16GB Kingston ECC DDR4 2133 Model KVR21R15D4/16. I was more confident this would work given the slower speed and many reviews. Unfortunately it did not.

I don't like to have to return items as it costs the business money and both of us time. I'm hoping my next order will be a success.

Does anyone know of what ECC memory models are combpatible with the B450M-A? I do realize I may have to settle for 2x8GB ECC, but the more memory I can get the better!

Thanks for reading

Edit: Seems like Kingston KSM26ED8/16ME 16GB DDR4 2666 is promising looking at what some have run on other Ryzen boards. It's just a bit pricey, like double the price!
 
Last edited:

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
Ahhh I see, I'm quite new to ECC memory. So likely these failed because of being registered...which I see is denoted by RDIMM as opposed to UDIMM.

Thanks Blue Fox, I will continue researching.
 

pitingres

Limp Gawd
Joined
Jul 25, 2018
Messages
144
The motherboard might accept ECC memory, but whether it will actually error correct, or just ignore the error correcting bits, is another matter. Unfortunately the B450m-a specs don't say for sure either way.
 

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
The motherboard might accept ECC memory, but whether it will actually error correct, or just ignore the error correcting bits, is another matter. Unfortunately the B450m-a specs don't say for sure either way.

gah...

How can one test this or ascertain how it behaves?
 

pitingres

Limp Gawd
Joined
Jul 25, 2018
Messages
144
I suppose you could ask Asus. The only other way I know is to try it, i.e. install some ECC memory, do something to induce memory errors (I've heard of carefully hitting it with a hair dryer or other heat source), and then look at the OS error correction counters to see if errors were corrected.

I was going to install ECC memory in my X470 Taichi (ultimate), but so far I haven't had any observed issues with regular non-ECC memory. I'd love to use ECC and it's really annoying that it's such a pile of unknowns. At least Ryzen gives you a chance at it, unlike Intel.
 

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
https://hardforum.com/threads/i-need-flawlessly-binned-ddr4-ram-whats-the-good-stuff.2000132/

I looked into EUDIMMs too, ECC Unbuffered/Unregistered DIMMs, but a lot of opinions seem to hint that it's not really that necessary for home use.
I think proper ECC RAM really ought to be RDIMMs, and that's why it's way more popular, and is the stuff that's actually used on servers.

The research I did pointed to the fact that most RAM errors are not caused by "cosmic rays," but are just defects in the silicon.
Better silicon means less defects, so I went with Ballistix Max. Why? Made by Crucial, which is owned by Micron.
I figure they save the best chips for their own brand, so for my NAS Router machine, I chose the best stuff by Crucial, and then plan to actually downclock it with a similar CAS timing to what it advertises.

That should basically eliminate all chances of errors, but still provide me with speed that ECC stuff isn't made for.
Although it is still possible to overclock ECC RAM. But why would you? ECC RAM is meant to be slow, boring, and stable. =p

The reason for ECC is simple:

If you have a massive server farm cluster that's built of say, 128 computers, and each computer has 128 GB of RAM, then the chance of a critical memory error is 128 times higher than a single computer with just 128 GB of RAM and not clustered. And a computer with just 32 GB of RAM has four times less chance of a critical RAM error than a computer with 128 GB.

So basically, ECC RAM is best when dealing with huge amounts of RAM, where all of it, all "16384 gigabytes" of it, as my example provided, must work perfectly all the time, with no critical RAM errors.

Basically, the difference between you and a server farm is that you have a "32-in-16384" chance of critical error compared to that server cluster I made up. 1 out of 512 chance then, compared to a cluster.
Of course, that's not to say ECC RAM wouldn't still protect you from an error. It really just depends on what you're doing and why you need ECC. Some errors aren't critical.

Even on big server farms, how common are RAM errors really? Well, I doubt companies want to tell us that, because then common consumers would get nervous if the number is too high. ;p

I recommend you do some additional research before going the ECC path, and if you DO go ECC UDIMM, let me know your results, and I'll let you know mine, and then we can see who made the better choice.
 

ThreeDee

[H]F Junkie
Joined
Sep 5, 2001
Messages
10,988
OP wants to know what ECC memory will work on his current setup .. only DDR4 UDIMM's will. Whether he "needs" ECC or not is up to the OP to decide.

If they were to ask if they "need" it .. I'd say , "well what are you doing with your hardware?" .. if they are making a server out of it and using an OS with ZFS file system, then I'd say yeah, it would be a good idea .. if they are installing Windows 10 and going to use their setup for gaming .. then I'd say no
 

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
https://hardforum.com/threads/i-need-flawlessly-binned-ddr4-ram-whats-the-good-stuff.2000132/

I looked into EUDIMMs too, ECC Unbuffered/Unregistered DIMMs, but a lot of opinions seem to hint that it's not really that necessary for home use.
I think proper ECC RAM really ought to be RDIMMs, and that's why it's way more popular, and is the stuff that's actually used on servers.

The research I did pointed to the fact that most RAM errors are not caused by "cosmic rays," but are just defects in the silicon.
Better silicon means less defects, so I went with Ballistix Max. Why? Made by Crucial, which is owned by Micron.
I figure they save the best chips for their own brand, so for my NAS Router machine, I chose the best stuff by Crucial, and then plan to actually downclock it with a similar CAS timing to what it advertises.

That should basically eliminate all chances of errors, but still provide me with speed that ECC stuff isn't made for.
Although it is still possible to overclock ECC RAM. But why would you? ECC RAM is meant to be slow, boring, and stable. =p

The reason for ECC is simple:

If you have a massive server farm cluster that's built of say, 128 computers, and each computer has 128 GB of RAM, then the chance of a critical memory error is 128 times higher than a single computer with just 128 GB of RAM and not clustered. And a computer with just 32 GB of RAM has four times less chance of a critical RAM error than a computer with 128 GB.

So basically, ECC RAM is best when dealing with huge amounts of RAM, where all of it, all "16384 gigabytes" of it, as my example provided, must work perfectly all the time, with no critical RAM errors.

Basically, the difference between you and a server farm is that you have a "32-in-16384" chance of critical error compared to that server cluster I made up. 1 out of 512 chance then, compared to a cluster.
Of course, that's not to say ECC RAM wouldn't still protect you from an error. It really just depends on what you're doing and why you need ECC. Some errors aren't critical.

Even on big server farms, how common are RAM errors really? Well, I doubt companies want to tell us that, because then common consumers would get nervous if the number is too high. ;p

I recommend you do some additional research before going the ECC path, and if you DO go ECC UDIMM, let me know your results, and I'll let you know mine, and then we can see who made the better choice.

I do understand your reasoning.

My desire for ECC stems from wanting to avoid possible data corruption. Perhaps I'm thinking about it incorrectly but if there's data corruption in a file stored in memory, then when written to the disk the corruption would follow as well. Is that correct to you?

Still if this is hardly an issue today, and if the best silicon is ultimately the best prevention, I do see the merit of getting the best crucial ram possible. What is their best these days anyways?

Thanks for the lengthy reply btw!
 

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
I do understand your reasoning.

My desire for ECC stems from wanting to avoid possible data corruption. Perhaps I'm thinking about it incorrectly but if there's data corruption in a file stored in memory, then when written to the disk the corruption would follow as well. Is that correct to you?

Still if this is hardly an issue today, and if the best silicon is ultimately the best prevention, I do see the merit of getting the best crucial ram possible. What is their best these days anyways?

Thanks for the lengthy reply btw!

Yeah, that's exactly how I see it too. I want my memory to be solid to avoid errors on the disk when writing. But what sort of data do you plan on writing? If it's photos and videos, then it matters significantly less than if it's either encrypted data where a single out-of-place bit can make it unreadable, or if it's data that's worth serious money to you.

Photo/video data, if it has a single bit error in memory or on the disk could change a single color of a single pixel just a tiny bit. From a hexadecimal color like #5600ff to #5600fe. The actual difference then is literally nothing to our human eyes. It is insignificant and doesn't matter. But if it's an encrypted chunk like 0110110110110110 with a password that decrypts it only if it matches the file structure exactly, then if one digit turns to 0 when it should be 1, perhaps the entire file will be unreadable after that. Although to be honest, I'm not sure that's how modern encryption works. It might be fine even if there was a single bit difference, like if the password accepting area was only at the beginning of the file. But then what if that small area is the area in memory affected by the error? Bad news then, eh?

Despite that, I still can't help but continue to feel wary about EUDIMMs. If EUDIMMs were more popular, I'd feel more comfortable getting it. For instance, if RDIMMs worked with my consumer motherboard, I would actually get that. But EUDIMMs just don't seem to have any market presence. Sure, it exists, but companies seem reluctant to produce or market it. I have to look rather hard to find it.

Apparently the guy who does Anand Tech clearly says you don't need ECC with ZFS. Maybe it's recommended by some people, but it seems to be a lot of hype by people who are very likely as knowledgeable as you or I, but not actually a professional data integrity expert. They might not actually have access to real-world memory failure statistics.

If you're just a typical geek who likes computer stuff, but doesn't necessarily manufacture their own silicon or own a data center filled with massive RAM disks containing precious customer data and financial information, then you're on my level, and we just can't know how common RAM errors are, when on the scale of a 32 GB system, simply because they don't actually happen very often. And certainly not as often as what you'd see in a 128 terabyte RAM data center.

My personal solution when building my NAS and internet routing machine was to buy Ballistix Max. I was looking at their no-heatsink desktop memory, but I didn't get the feeling it was well binned. I *really* don't like the idea of using "gaming" equipment, but a good binning is a good binning.

Truth be told, I'm not an expert, and will have to see what happens. It hasn't even arrived yet, so I'm at the same level as you, and possibly working on a very similar project.

To test data corruption, why not create a RAM disk on your current computer, and create some large, complex files, such as an 8 GB image, like a picture you made in GIMP or Photoshop of a bunch of different colors and stuff. Nothing fancy, just make sure there's tons of gradients and no easy way to compress the file, you know? Then take note of the checksum of that file after you've saved it to the RAM disk. SHA256.

Then spend some time gaming, processing data, browsing the web, watching videos, and don't shut your computer down for a week.

Then recalculate the checksum for that file you stored in memory. Did it change? If yes, then in just one week, you've proven that RAM errors can certainly happen and corrupt data on consumer equipment. But if the data is fine, then perhaps the risk is not so big, at least if the data isn't extremely valuable.

And that's with your normal computer RAM.

What I did was buy some Crucial Ballistix Max, 4000 MHz RAM at CAS 18.
Then I plan to clock it manually to 3200 MHz at CAS 16, or maybe even 3600 MHz at CAS16, which still seems slow for it, just to give it some breathing room.

It's like taking an Olympic sprinter and putting them on a nice path in a beautiful park. If they can handle going absurdly fast in a serious race, then it should be a piece of cake to put them on a gentle forest path, with little rabbits hopping around and birds tweeting. Uh, I'm just being descriptive here for fun.

To be clear, if your data is VERY important, and worth true value to you, where a single bit error could be VERY bad, then you really ought to buy some used server equipment and RDIMMs, as well as use a server processor. Processor binning is the act of analyzing silicon and checking for defects. The more defects, the lower it goes in the bins. Intel makes one type of silicon wafer, and then bins it for i9 if it's good, i7 if it's alright, i5 for the mid quality stuff, and i3 for wafer scrap that's barely functional.

Only Xeons are the stuff that they consider flawless enough for enterprise. And even those get binned.

Intel isn't very popular right now, but they certainly make for a good example of how to bin stuff. The same will happen with memory chips, and I wouldn't be surprised if the best stuff goes into RDIMMs, because that's what server producers actually buy. The memory that "gamers" buy is also very likely lower quality than what they would sell to a professional company that's buying memory for their server. Corruption in a customer's credit card details or personal information could be a lot worse than a memory error while gaming.

I'm sure all companies would love to sell flawless silicon to everyone, but the reality is that some silicon wafers, no matter if it's memory or processing substrate, is going to contain defects, and companies can still find a use for it, even if it has a little flaw.

When it comes down to it, I don't know who is buying EUDIMMs. If I was a serious company I'd buy Epyc servers with RDIMMs and Quadro GPUs. I wouldn't mess around with some consumer level device that can't even run a normal RDIMM. Even if EUDIMMs work, I can't find enough information on that entire class of product to feel comfortable buying it. No one is marketing it, and I had to search really hard to find that stuff. If I had a proper budget, I'd just test the two side by side, and see which configuration is worth the money.

Downclocked gaming RAM vs ECC. Is there a true benefit? It's more likely the error occurs on the disk drive itself anyways.

When it comes down to it, these things are an issue of money, right? Otherwise we'd all just buy the most powerful server hardware and do our gaming at 240+ frames per second on our high-end huge monitors, we'd have petabytes of data storage, all with multiple backups, and we'd all have 100 gigabit ethernet.

But instead, I've gotta look at things pragmatically. If ECC is just chasing ghosts and doesn't provide serious real-world benefit, other than "it'll prevent the occasional error," then I just can't see it being useful compared to much more versatile RAM like some downclocked Ballistix Max.

It's a matter of a lack of serious research by others, as well as having to navigate marketing segmentation and other dumb hurdles. Am I paying more money for "Max" just because it's "1337 g@ming" or am I actually buying a good product? Is their desktop/office RAM, although boring as heck, actually better than gaming RAM, but just sold without an XMP or DOCP profile, and no fancy heatsink?

Is ECC RAM actually going to protect me from errors? Do these errors actually happen for real? Of course they happen, but are they "catastrophic," in the sense that you or I imagine? I asked the same questions you did, just the other day in fact, so I'm still wondering if I made the right choice by sidestepping around ECC UDIMMs and doing this "downclock the good stuff" maneuver.

But yeah, if you do that experiment with the RAM disk I mentioned, keep a single large file in pure memory for a week, while also doing other stuff in memory, just on regular desktop RAM, will even ONE RAM error occur? And if it does occur, does it matter in something like an image?

Well, that's all for now! Uh, I just like talking about this stuff, so sorry for the paragraphs. =p
 
Last edited:

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
For the record, this is the stuff I would have bought if I had bought ECC UDIMMs.
https://www.newegg.com/crucial-16gb-288-pin-ddr4-sdram/p/1X5-001S-002T4

This is the stuff I bought instead.
https://www.crucial.com/memory/ddr4/BLM2K8G40C18U4B

It's more versatile because I can always switch the NAS drives out and turn my NAS Router into a gaming machine if I want. It's all consumer hardware. But I'm happy that AMD lets us use EUDIMMs if we want. Maybe I'll switch to EUDIMMs someday if the Ballistic Max doesn't work out.

I got the ROG Strix B550-F because it has built in Wifi 6 and a 2.5 gigabit ethernet port. My friend who's funding this happens to be an ITX nerd, and can't upgrade to 10 gigabit, so I picked out a motherboard that'd hook up to their 2.5 gigabit ethernet ITX B550 board with perfect speed compatibility. But I also plan to install a four-port 10 gigabit NIC too. I went with a 3600x, which might seem like overkill, but the binning!! I'm sure a chip with an X at the end has better silicon, and that just means less chance of errors, especially if downclocked a little.

On top of that, I'm using a single 2 TB NVMe drive for main storage, with an automatic "when-data-is-modified" backup drive, which is a 4 TB Ironwolf Pro HDD. Based on my research, I don't think RAID is necessary. Instead, a more expensive Pro version of Seagate's main NAS drive was what I went with. WD had that drama about their shingled NAS drives recently, so I don't trust 'em much anymore. But I wouldn't be surprised if they all did it. Honesty and integrity is not obsolete. Just underutilized.

NVMe SSDs are chips, and that means it's binned for silicon defects. In other words, the more free of defects it is, the better for avoiding data corruption. DOA type defects are the only thing I imagine could happen with something quality like that, because if it works a little, it works a lot. It's not like an HDD, so it won't explode just from a single bump. If it does die suddenly, it'll probably be from heat during use, so I'll be there when it dies, and I can just manually replace it, then store the files I'm transferring on the backup HDD in the meantime.

I don't think NVMe in RAID is even recommended, and although I'd have to check, I'm not sure it's even possible. I believe that the controller on the NVMe SSD itself is actually a RAID-type device linking the multiple memory chips. That's why Phison's E16 NVMe controller apparently has multiple small ARM CPUs on the SSD. The SSD is literally powered by MULTIPLE processors! Phison's the brand I've been researching lately because they seem to have some serious potential, especially with their upcoming E18 controller.

I don't know Phison's secrets, but I bet it's very similar to a RAID configuration, only it's hardwired to use the chips on the NVMe drive, instead of HDDs. That's why another layer of RAID using two NVMe SSDs doesn't work very well. But you'd have to ask an expert.

However, I'm actually pondering Crucial's P5 drive for this. It's actually very new, so although it's not tested for "reliability" yet, it's still built by an established brand that makes memory already. NVM does stand for "Non-Volatile Memory," so for this, I would trust the "reliable" brand of memory, which happens to be Crucial/Micron. I don't get paid to recommend them, but I probably should be. =p

Anyways, this was just meant to be a small paragraph but I guess I really like researching and writing about this stuff.
I hope you enjoyed it.
 
Last edited:

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
Aegir haha awesome, people like you are why I keep coming back to HF for so many years.

Okay so my important data is, as you seemed to have guessed, pictures and videos. I did not realize corruption would only be a tiny change on pixels! That makes it entirely insigifnicant.

I do play games and have some VMs and what not, but nothing critical. I'm sticking with Windows, at least for now.

All of the data is one OneDrive as well, which idk if that will do error correcting in and of itself. Do you have any idea about that actually? Does cloud storage prevent data corruption because of the cloud storage presumably error correcting itself? One would think so...but then I still think a corrupted file on my machine would just get uploaded to the cloud and replicated there............................

This however does lead me to another idea. What if I used a linux VM with ZFS that synced with OneDrive! That would perhaps then avoid data corruption and let me use normal RAM! I had not thought of ZFS at all, it's been a while since I've messed around with that.

The cost of 64GB of more normal RAM and the cost of 32GB of ECC RAM can be quite similar, and I see more general use for the 64GB RAM.

To test data corruption, why not create a RAM disk on your current computer, and create some large, complex files, such as an 8 GB image, like a picture you made in GIMP or Photoshop of a bunch of different colors and stuff. Nothing fancy, just make sure there's tons of gradients and no easy way to compress the file, you know? Then take note of the checksum of that file after you've saved it to the RAM disk. SHA256.

I may try this as well. Though how to create a 8GB image file(lol) is something I would need to figure out as well!

Thanks for your help and time Aegir
 

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
Aegir haha awesome, people like you are why I keep coming back to HF for so many years.

Okay so my important data is, as you seemed to have guessed, pictures and videos. I did not realize corruption would only be a tiny change on pixels! That makes it entirely insigifnicant.

I do play games and have some VMs and what not, but nothing critical. I'm sticking with Windows, at least for now.

All of the data is one OneDrive as well, which idk if that will do error correcting in and of itself. Do you have any idea about that actually? Does cloud storage prevent data corruption because of the cloud storage presumably error correcting itself? One would think so...but then I still think a corrupted file on my machine would just get uploaded to the cloud and replicated there............................

This however does lead me to another idea. What if I used a linux VM with ZFS that synced with OneDrive! That would perhaps then avoid data corruption and let me use normal RAM! I had not thought of ZFS at all, it's been a while since I've messed around with that.

The cost of 64GB of more normal RAM and the cost of 32GB of ECC RAM can be quite similar, and I see more general use for the 64GB RAM.



I may try this as well. Though how to create a 8GB image file(lol) is something I would need to figure out as well!

Thanks for your help and time Aegir


I don't know much about OneDrive to be honest. I mostly avoid cloud stuff, unless it's my own cloud. I would say that they probably do their best to keep your data safe, but there's really no way to know how they operate.

As for ZFS, I'm just using it as an example, due to it being popular to say "ECC RAM is required if you use ZFS!" And that statement is simply not true. Beyond that, I don't even plan to use ZFS for my own personal NAS. I'll use something that Windows can natively read probably, although I'll be using Linux on the NAS itself.

And if you want to create a RAM Drive, check out ImDisk. It's free and open-source.
 

Deathwish238

Limp Gawd
Joined
Oct 11, 2006
Messages
312
Aegir Thanks for your reply once more! ImDisk looks pretty nifty too, I appreciate the link.

I'm curious what you think of this study? http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf

It is a little old, but seems mostly relevant still...

Though it does appear to support what you've been saying that memory errors are mostly due to bad silicon.

But still this makes it seem way more common than expected:
Our per-DIMM rates of correctable errors translate to an average of 25,000–75,000 FIT (failures in time per billion hours of operation) per Mbit and a median FIT range of 778 – 25,000 per Mbit (median for DIMMs with errors)

Do you have any insight on the Ballistix Elite vs Ballistix Max? Is the Max that much better?
Elite: https://www.newegg.com/ballistix-16gb-288-pin-ddr4-sdram/p/N82E16820164153
Max: https://www.newegg.com/ballistix-16gb-288-pin-ddr4-sdram/p/N82E16820164164

Both are DDR4 4000 and it appears that the Max line replaced the Elite line.
 

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
Aegir

I'm curious what you think of this study? http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf


Do you have any insight on the Ballistix Elite vs Ballistix Max? Is the Max that much better?
Elite: https://www.newegg.com/ballistix-16gb-288-pin-ddr4-sdram/p/N82E16820164153
Max: https://www.newegg.com/ballistix-16gb-288-pin-ddr4-sdram/p/N82E16820164164

Both are DDR4 4000 and it appears that the Max line replaced the Elite line.

The Max is newer, and so it's probably better and more compatible with Ryzen.
I don't see any reason not to get it, considering.

And yeah, that's the exact study I was talkin' about.

I think many memory errors are actually already automatically corrected by the OS or the CPU usually, so the only thing ECC seems to protect against is "bitflip" from a cosmic ray or similar, which would go undetected by the rest of the system.

Either way, after even more research, I feel pretty comfortable with the Ballistix Max I bought. But it still hasn't arrived yet, so I haven't had a chance to run a test RAM disk and see if it can truly hold data for a long period without any problems.

I also only got the 16GB pack, instead of 32GB. I'm a bit sad, because my cool network card supports RDMA: Remote Direct Memory Access. A very advanced feature. But hey, I'm not exactly rich, so I can't just buy anything. Oh well! This sort of project is still very fun.
 

Aegir

Limp Gawd
Joined
Apr 15, 2020
Messages
442
Oh, and many NVMe SSDs are using ECC technology, so if written from the CPU down the PCIe lane, and into the NVMe drive, it will act like memory and automatically guard against memory errors, including bitflip. Pretty nifty, eh?

I think this also bypasses RAM entirely, so when writing directly from NIC to CPU to NVMe, I don't see any reason that your data would get loaded into RAM.
 
Top