ECC is not only enabled on Ryzen, but it works! ...Sort of.

[MACMAC]

While there has been countless confirmations that ECC is supported by Ryzen processors and AM4 motherboards, and even one or two confirmations that it was actually enabled. I pretty sure this first confirmed tests that show that it's actually working, more or less.

 

Quote

In conclusion, what is currently available on the AM4 platform is an incomplete implementation of ECC. This is very likely why motherboard manufacturers have been relatively hesitant about claiming that their products support ECC memory in ECC mode. Based on our findings, there is clearly some level of ECC functionality that is working right now, but it does not cover the full spectrum of memory error detection and correction. Having said that, the status quo is arguably better than nothing, especially since single-bit errors are much more likely than multi-bit errors (which are often caused by a failing memory module), so I suspect that many people will still want the extra protection that is available right now.

 

Article: http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/75030-ecc-memory-amds-ryzen-deep-dive.html

[ARikozuM]

I can't wait until users start building servers with Ryzen only to lose all data just because the ECC didn't work with X board.

[dalekphalm]

7 minutes ago, ARikozuM said:

I can't wait until users start building servers with Ryzen only to lose all data just because the ECC didn't work with X board.

Well, that should never happen, because:

1. OEM's like Dell and HP will for sure certify any Ryzen based servers with ECC - or simply not offer it as an option if they cannot validate it.

2. Anyone who is building their own server - especially for a production (meaning: live business) environment, should 100% only be buying RAM that is on the validation list for their motherboard.

 

Now, if nothing has been validated for ECC yet? Then users should avoid using Ryzen in such situations.

[MACMAC]

7 minutes ago, ARikozuM said:

I can't wait until users start building servers with Ryzen only to lose all data just because the ECC didn't work with X board.

To be clear, they would have lost data without semi-ECC too, in whatever scenario you're creating.

The current situation for most prosumer types isn't buy Ryzen or buy a proper ECC validated platform, since those platforms are out of the financial reach of most people.

I wrote this article exactly so people wouldn't be buying into this blindly, which is what was happening before.

[ARikozuM]

Just now, dalekphalm said:

Well, that should never happen, because:

1. OEM's like Dell and HP will for sure certify any Ryzen based servers with ECC - or simply not offer it as an option if they cannot validate it.

Almost everyone in this community would prefer their own server over those by HP/Dell (aside from @Electronics Wizardy).

Just now, dalekphalm said:

2. Anyone who is building their own server - especially for a production (meaning: live business) environment, should 100% only be buying RAM that is on the validation list for their motherboard.

Since when do users check? 

Spoiler

That was a joke.

Just now, dalekphalm said:

Now, if nothing has been validated for ECC yet? Then users should avoid using Ryzen in such situations.

Agreed.

[MACMAC]

2 minutes ago, dalekphalm said:

Now, if nothing has been validated for ECC yet? Then users should avoid using Ryzen in such situations.

As I wrote in the article, validation is never going to happen. But that hasn't stopped people from successfully using ECC on previous AMD platforms.

[rjfaber91]

17 minutes ago, MACMAC said:

(...) ECC is supported by Ryzen (...)

AMD does have something of a longstanding tradition of not bothering with the kind of artificial deactivation of features that Intel is such a fan of, so I'm assuming that the limited ECC support on consumer-grade Zen chips is simply because it's already worked into the chips to make creating server-grade Zen chips later a lot easier, and AMD simply didn't bother to disable that functionality.

[MACMAC]

1 minute ago, Aleksiandrovich said:

X99 should have done this, i hope X299 allows for such compatibility with 12C16T skews.

To be fair, you can put a Xeon on an X99 motherboard and it will work with ECC memory, even Registered ECC memory, I've tested it.

[dalekphalm]

Just now, ARikozuM said:

Almost everyone in this community would prefer their own server over those by HP/Dell (aside from @Electronics Wizardy).

Since when do users check? 

  Reveal hidden contents

That was a joke.

Agreed.

For here, yes. At work? Fuck no.

 

I'll take the Dell ProSuppport warranty with guaranteed 4 hour response time, and next day parts/onsite tech delivery, thank you very much

 

Since when do users check? Well, if they are building a server with ECC RAM, they should damn well be checking every time. Though I suspect, especially users on here, not so much lol.

[MACMAC]

1 minute ago, Aleksiandrovich said:

Yeah but i meant like 6800K so on, non xeon chips.

That would be the dream, but Intel will never allow that to happen, too much money to be made on Xeon chips...

[ARikozuM]

3 minutes ago, MACMAC said:

To be fair, you can put a Xeon on an X99 motherboard and it will work with ECC memory, even Registered ECC memory, I've tested it.

Registered RAM won't work in unbuffered slots... 

[Notional]

I wonder if the issue resides in the motherboards, the MC in the CPU, or the microcode, not unlocking the full support?

 

That being said, shouldn't ECC be standard about now? It cannot be that expensive to implement it, compared to flashy metal shrouds, ugly RGB lighting and stupid chrome limited edition crap. I'd rather see some nice looking ECC memory.

[ARikozuM]

Just now, Notional said:

I wonder if the issue resides in the motherboards, the MC in the CPU, or the microcode, not unlocking the full support?

 

That being said, shouldn't ECC be standard about now? It cannot be that expensive to implement it, compared to flashy metal shrouds, ugly RGB lighting and stupid chrome limited edition crap. I'd rather see some nice looking ECC memory.

ECC requires an extra chip, a compatible memory controller from the CPU, and compliance (for lack of a better word) from the chipset. Price has been coming down lately but I don't see the value of ECC in gaming systems when the speeds are usually the base speed of the RAM (i.e DDR3-1333 or DDR3-1600). You're not going to find 3000MHz ECC memory anytime soon.

[Notional]

1 minute ago, ARikozuM said:

ECC requires an extra chip, a compatible memory controller from the CPU, and compliance (for lack of a better word) from the chipset. Price has been coming down lately but I don't see the value of ECC in gaming systems when the speeds are usually the base speed of the RAM (i.e DDR3-1333 or DDR3-1600). You're not going to find 3000MHz ECC memory anytime soon.

As long as ECC is workstation/server only, then yeah, only standard speeds. But ECC should be very easy to sell to gamer nerds, with a little price premium. Sounds like it won't require a whole lot from AMD to make it work in Ryzen either.

[MACMAC]

14 minutes ago, ARikozuM said:

Registered RAM won't work in unbuffered slots... 

Well as I wrote - I tested it - and it worked fine in an ASUS X99-Deluxe II. Much to my surprise, believe me.

[MACMAC]

13 minutes ago, Notional said:

I wonder if the issue resides in the motherboards, the MC in the CPU, or the microcode, not unlocking the full support?

 

That being said, shouldn't ECC be standard about now? It cannot be that expensive to implement it, compared to flashy metal shrouds, ugly RGB lighting and stupid chrome limited edition crap. I'd rather see some nice looking ECC memory.

I think AMD has to release microcode that will allow BIOS manufacturers to unlock additional ECC-related settings, once that happens and a bit of driver/kernel work is done, then even 2-bit ECC should work in both Linux and Windows.

[MACMAC]

Just now, huilun02 said:

Unacceptable. How can AMD's consumer processors/chipsets not support ECC when Intel's does? 

Actually, like I wrote in the second paragraph (lol), most consumer Intel processors and chipsets don't support ECC. No Core i7s or Core i5s support it. A few Core i3s do (at least previous generations), and some Pentiums.

[Ronda]

Is halting the machine really standard operating procedure for multi bit errors? If you value data integrity above all else, I suppose you'd want that, but if availability of a service is also a consideration, the way it works now on Linux 4.10 seems to be the best approach to me.

[MACMAC]

12 minutes ago, Ronda said:

Is halting the machine really standard operating procedure for multi bit errors? If you value data integrity above all else, I suppose you'd want that, but if availability of a service is also a consideration, the way it works now on Linux 4.10 seems to be the best approach to me.

Yes, it is.

 

A multi-bit error is so rare that when it happens it means something has gone fatally wrong, so you don't want to the machine to keep operating if you value your data at all, or if you want to be able to trust the results of whatever program(s) you're running.

[Dabombinable]

Considering the cheap VIA Apollo Pro 133A chipset from 1999 had full support for ECC RAM...and was found in even some of the cheapest motherboards......

[Doobeedoo]

It's there but those needing should be seeking validation for then.

[Valentyn]

11 hours ago, ARikozuM said:

I can't wait until users start building servers with Ryzen only to lose all data just because the ECC didn't work with X board.

Why would you use Ryzen instead of Naples?

It's like using Z270 for a server and then bitching it all went up in smoke; despite having a proper server platform available.

[dalekphalm]

10 hours ago, MACMAC said:

Yes, it is.

 

A multi-bit error is so rare that when it happens it means something has gone fatally wrong, so you don't want to the machine to keep operating if you value your data at all, or if you want to be able to trust the results of whatever program(s) you're running.

Indeed, some Google Engineer wrote a paper about RAM bit-errors, and he calculated that a single bit error occurs approximately every 1.5 years (assuming Google Data Center workload) per stick.

 

He also calculated that under normal operation, we'd all be long dead and gone before a multi-bit error happens - exceptions being things like a defective or damaged RAM stick, etc).

[ARikozuM]

5 hours ago, Valentyn said:

Why would you use Ryzen instead of Naples?

Cost... Simply cost... We've all been there.

5 hours ago, Valentyn said:

It's like using Z270 for a server and then bitching it all went up in smoke; despite having a proper server platform available.

Plenty of first-time server builders use consumer-grade (nothing wrong with that) hardware without checking what their primary needs are. Using a Gaming 5 motherboard is all good for a server since it has the features you may need, but you don't need the majority of those features. You don't need an audio chip in a server. You need/want something that is dependable, offers networking (IPMI, WoL, etc.), ECC (for critical data and RAID), extra GbE LAN ports if necessary, and you want it to consume as little juice as possible if it's a 24/7 server. 

[Valentyn]

Just now, ARikozuM said:

Cost... Simply cost... We've all been there.

Plenty of first-time server builders use consumer-grade (nothing wrong with that) hardware without checking what their primary needs are. Using a Gaming 5 motherboard is all good for a server since it has the features you may need, but you don't need the majority of those features. You don't need an audio chip in a server. You need/want something that is dependable, offers networking (IPMI, WoL, etc.), ECC (for critical data and RAID), extra GbE LAN ports if necessary, and you want it to consume as little juice as possible if it's a 24/7 server. 

How many Intel Z170 and Z270 boards support ECC though?

That's what I'm getting at, buying consumer components can't guarantee enterprise/workstation features.

 

Applies to Intel and AMD; which is why the former charges even more for X99 so that those can support it all without going straight into Server components.