Did someone test ECC RAM on new Ryzen 3000 ?

Jandor

Gawd
Joined
Dec 30, 2018
Messages
665
Would be interesting to know the specs and the speed. Also mind to say what motherboard and CPU.
I'm used to put only ECC on all my computers that can. And Am3 was compatible and now Ryzen is.
 
Multiple X570 motherboard vendor specs list ECC support (always unbuffered on AM4), doubt anything changed.

Only non-Pro APUs are still left out.

FWIW testing and proving it is truly working is actually kinda hard, speaking from experience. Reading various windows/linux kernel flag bits is not solid proof, you need a logged and corrected error. It is possible for false negatives as some configurations will correct but not properly log the error.


As for speeds, if you use good b-die you can crank up the speeds like any other kit, though ECC is currently not binned or given fancy "custom" pcbs and heatspreaders so maybe not as high. I hit 3200C14 single rank on Zen1 with random unmatched 2133/2400 used server pull kits, b-die of course.

If these blingshit flashy memory reseller vendors were serious all their newer AMD-tested kits would be ECC, but then users might find out their so-called premium binning is sometimes just unstable trash.
 
I'm wondering when they're just going to start selling 'better' memory as it performs, listing a range of speeds, timings, and voltages, versus selling different kits of the same stuff with different XMP presets.

As far as 'ECC everything' goes... it's a nice idea, but it doesn't make a lot of sense. Very few will actually need it and it straight up costs more to produce.

Now, flipside, the utility for 'stable overclocks' using supported systems does make sense. Just expect to pay more for it over current 'enthusiast' kits if it ever materializes.
 
I'm wondering when they're just going to start selling 'better' memory as it performs, listing a range of speeds, timings, and voltages, versus selling different kits of the same stuff with different XMP presets.

As far as 'ECC everything' goes... it's a nice idea, but it doesn't make a lot of sense. Very few will actually need it and it straight up costs more to produce.

Now, flipside, the utility for 'stable overclocks' using supported systems does make sense. Just expect to pay more for it over current 'enthusiast' kits if it ever materializes.
It should cost around less than 9/8 of non ECC (same board cost one RAM chip more of 8), that is what it costs to produce.
It should be nonsense to use non-ECC chips when it's supported.
 
It should cost around less than 9/8 of non ECC (same board cost one RAM chip more of 8), that is what it costs to produce.

This would only be true if the same number of the enthusiast ECC part were produced and sold, so as to even out every other cost. As it stands they'd sell so few that they'd have to charge significantly more then the difference in BOM.

It should be nonsense to use non-ECC chips when it's supported.

It's both poorly supported (look at trying to prove that it is even enabled) and outright just not needed. It would be nice if support were firmed up and modules were available with correct markings for those that want them, but neither are the case today unfortunately.
 
I blame industry inertia more tbh, and some intel segmentation. Along with luck: the fact joe blow doesn't actually load ram 24/7 or notice errors as much. There are a lot of bits you can flip that don't crash or corrupt a data operation completely.

Every data bus in the world worth using has built in error encoding, except for consumer ram for some reason.

Ancient USB 1.x has better random error protection (8b/10b) than non-ECC dram ffs.

There are metric fartloads more server ram being built than consumer too. (mobile outnumbers all by now though)

Pick up a commodity naked dimm and look at the pcb, it probably has the empty silkscreen spot for the 9th chip, maybe even the spot for a register too. Yet these tweaker aftermarket OEMs have the money to make "total custom" pcbs, embedded leds, blingshit slivers of machined aluminum etc but nope that "custom" pcb is just for 8 chips instead of 9? Pure lazy and shortsighted. Also selling you a lot of binning myths, lots of thick glue and even etching the ICs so you can't see what they are actually being made with.

Not as bad as the monitor market, but the branded ram market fucking sucks.
 
I'm wondering when they're just going to start selling 'better' memory as it performs, listing a range of speeds, timings, and voltages, versus selling different kits of the same stuff with different XMP presets.

As far as 'ECC everything' goes... it's a nice idea, but it doesn't make a lot of sense. Very few will actually need it and it straight up costs more to produce.

Now, flipside, the utility for 'stable overclocks' using supported systems does make sense. Just expect to pay more for it over current 'enthusiast' kits if it ever materializes.
You can buy sets with an XMP preset for high clocks/low timings. But it costs like $300+
 
Not as bad as the monitor market, but the branded ram market fucking sucks.

Quoting in agreement. Although right now one can get an i3 at retail that supports ECC- and a Ryzen APU that doesn't. Neither make a ton of sense, and it would be nice for ECC to always be an option from a CPU and chipset perspective, even if it won't be used in some markets due to BoM or even potentially power draw.
 
You can buy sets with an XMP preset for high clocks/low timings. But it costs like $300+

They could cut down on confusion, support calls, and returns if they simply sold stuff with profiles that make sense, i.e. a JEDEC profile guaranteed to boot and an XMP 'performance profile' suited to whichever system, and then released a 'timing table' with up and down steps of speeds, voltages, and timings for various platforms.
 
i3 needs a server chipset, C246 or C242 for 8th/9th gen.

AM4 needs ECC support on motherboard (nearly all ASRock and Asus) and any Ryzen or PRO APU.
 
Will buffered ECC dimms work or are they electrically or physically incompatible?
 
Will buffered ECC dimms work or are they electrically or physically incompatible?

Epyc is registered, TR & AM4 are unbuffered.

Its possible with a firmware update (cpu itself most likely, not just bios code) that it would work, but only AMD would know for sure.

I do know that in general ddr dimm slots with all traces are electrically compatible, as an example X99 boards work with either kind of ram (but no mixing) if you have an E5 Xeon.
 
Will buffered ECC dimms work or are they electrically or physically incompatible?

They will not work unless AMD releases a firmware update that allows this.
 
I blame industry inertia more tbh, and some intel segmentation. Along with luck: the fact joe blow doesn't actually load ram 24/7 or notice errors as much. There are a lot of bits you can flip that don't crash or corrupt a data operation completely.

Every data bus in the world worth using has built in error encoding, except for consumer ram for some reason.

Ancient USB 1.x has better random error protection (8b/10b) than non-ECC dram ffs.

There are metric fartloads more server ram being built than consumer too. (mobile outnumbers all by now though)

Pick up a commodity naked dimm and look at the pcb, it probably has the empty silkscreen spot for the 9th chip, maybe even the spot for a register too. Yet these tweaker aftermarket OEMs have the money to make "total custom" pcbs, embedded leds, blingshit slivers of machined aluminum etc but nope that "custom" pcb is just for 8 chips instead of 9? Pure lazy and shortsighted. Also selling you a lot of binning myths, lots of thick glue and even etching the ICs so you can't see what they are actually being made with.

Not as bad as the monitor market, but the branded ram market fucking sucks.

this should have been done 15 years ago when you still had to research your ram/mobo/cpu to make sure they were compatible. it's gotten so easy to build a pc now, that pretty is what sells. IMO the market missed that exit, and it isnt likely to make the turn any time soon.
 
One of the good things about competition is having the market leader try harder to provide value.

Hey Intel - Ryzens and Threadrippers provide ECC support... just sayin’ > . >
 
One of the good things about competition is having the market leader try harder to provide value.

Hey Intel - Ryzens and Threadrippers provide ECC support... just sayin’ > . >

If I have a good gripe about Intel's product segmentation, this is largely it. Let me run ECC on an overclocked consumer CPU if I damn will want to, damnit.

:)
 
Besides memtest, are there any other ways to make sure the ECC RAM is actually working to correct errors, and not just 'enabled' in the bios?
 
In linux you can see a log of the corrections (likely there won't be any) with edac-util

Code:
jmd0 ~/shell-scripts # edac-
edac-ctl   edac-util
jmd0 ~/shell-scripts # edac-util -v
mc0: 0 Uncorrected Errors with no DIMM info
mc0: 0 Corrected Errors with no DIMM info
mc0: csrow0: 0 Uncorrected Errors
mc0: csrow0: mc#0csrow#0channel#0: 0 Corrected Errors
mc0: csrow0: mc#0csrow#0channel#1: 0 Corrected Errors
edac-util: No errors to report.

I did some more testing 2 years ago when I got my Ryzen 2700 and the ASUS X470 Prime PRO + 16 GB Crucial DDR4 2400 EUDIMM Kit (ECC Unbuffered) for my linux pvr / server:

https://hardforum.com/threads/x470-ecc-support.1960005/#post-1043662242
 
In linux you can see a log of the corrections (likely there won't be any) with edac-util

Code:
jmd0 ~/shell-scripts # edac-
edac-ctl   edac-util
jmd0 ~/shell-scripts # edac-util -v
mc0: 0 Uncorrected Errors with no DIMM info
mc0: 0 Corrected Errors with no DIMM info
mc0: csrow0: 0 Uncorrected Errors
mc0: csrow0: mc#0csrow#0channel#0: 0 Corrected Errors
mc0: csrow0: mc#0csrow#0channel#1: 0 Corrected Errors
edac-util: No errors to report.

I did some more testing 2 years ago when I got my Ryzen 2700 and the ASUS X470 Prime PRO + 16 GB Crucial DDR4 2400 EUDIMM Kit (ECC Unbuffered) for my linux pvr / server:

https://hardforum.com/threads/x470-ecc-support.1960005/#post-1043662242


Thanks. I'm building a new rig with an AsRock X470 taichi and a Ryzen 7 3800X. I'm trying to decide whether it's worth getting ECC RAM for it. I'd prefer to know if ithe ECC memory will actually correct any errors before I decide. If there are no error logs created, and no way to make sure, then it's probably not worth it.

I'm on Windows. Would Linux in virtualization be able to generate error logs?
 
Would Linux in virtualization be able to generate error logs?

It appears not in ubuntu 20.04 WSL2.

Code:
jdrescher@jmdws:/mnt/c/Users/jdrescher.RADIMG$ uname -a
Linux jmdws 4.19.104-microsoft-standard #1 SMP Wed Feb 19 06:37:35 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux
jdrescher@jmdws:/mnt/c/Users/jdrescher.RADIMG$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 20.04 LTS
Release:        20.04
Codename:       focal
jdrescher@jmdws:/mnt/c/Users/jdrescher.RADIMG$ edac-util -v
edac-util: Fatal: Unable to get EDAC data: Unable to find EDAC data in sysfs
jdrescher@jmdws:/mnt/c/Users/jdrescher.RADIMG$ sudo edac-util -v
edac-util: Fatal: Unable to get EDAC data: Unable to find EDAC data in sysfs

I have 32 GB of ECC on my windows workstation listed in my sig. However I am not sure how to see this info in windows.
 
Last edited:
Sounds like windows logs error corrections to the event viewer. Statistically you start needing ECC as the amount of RAM goes up and assuming you actually utilize the RAM for data like financial or scientific data. It is not really needed for media creation or consumption. It’s more likely to be helpful at 64GB and up.
 
Back
Top