13900k W680 vs 7900x - DDR5 ECC support

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
S

Styp

Member
Aug 1, 2018
69
21
8
I am looking into building a new workstation for mainly two purposes: Lightroom and Davinci Resolve
According to the Pugetsystem benchmark, getting into a more expensive platform doesn't make sense because single-thread performance is still crucial.

I am considering going for a system with DDR5 ECC memory and I am unsure which platform works better:

- Intel: 13900k + Asus W680 ACE + Kingston DDR5-4800 ECC UDIMM
- AMD: 7900x + X670 / B660 + Kingston DDR5-4800 ECC UDIMM

Is there a better memory? How is the ECC stability on AMD compared to Intel? I think Intel is a pretty safe bet at this point? Are there any 48GB modules on the market yet?
 
T_Minus

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,646
2,063
113
Not that I looked into it, as I went AMD + ECC for fastest single-core in a business setting (not desktop), but I don't recall seeing much testing of intel consumer CPUs and ECC -- has there been enough of that now to consider it working?

AMD has been shown to work to the prev generation, and iirc mostly intel ECC was rare i3 or i5 not the top CPUs?

I'm curious if intel's now all fine on ECC. Could be game changer for small servers where people still want ECC and ramp up that single-core perf for a good \affordable $
 
B

BlueFox

Legendary Member Spam Hunter Extraordinaire
Oct 26, 2015
2,098
1,513
113
  • Like
Reactions: T_Minus
S

Styp

Member
Aug 1, 2018
69
21
8
@T_Minus: As far as I undestand, W680 with Core-i 13th gen is replacing the former Xeon W-12xx offering. The Z680 vs. W680 chipset distinguishes the ECC support.

I am just a little afraid that AMD and ECC will not be stable. Who has experience with that? Is it even officially supported?
 
J

jei

Active Member
Aug 8, 2021
153
82
28
Finland
I've been using X570 + AMD 5700X/5900X with ECC (Windows 10) since 2019 without a hitch. The motherboard (Asus Pro WS X570-ACE) explicitly supports ECC "Reliable and responsive ECC memory suitable for mission critical tasks"
 
S

Styp

Member
Aug 1, 2018
69
21
8
jei said:
I've been using X570 + AMD 5700X/5900X with ECC (Windows 10) since 2019 without a hitch. The motherboard (Asus Pro WS X570-ACE) explicitly supports ECC "Reliable and responsive ECC memory suitable for mission critical tasks"
Click to expand...
Yeah but this is DDR4 with AM4 - I am looking into AM5 with DDR5, which is confusing at the moment anyway with the on-die ECC terminology.
 
  • Like
Reactions: jei
codgician

codgician

New Member
Dec 14, 2022
15
6
3
La La Land
On-die ECC is not equivalent to ECC RAM.

From my understanding, on-die ECC happens within each RAM die. It is needed because DDR5 RAMs usually have very high clock frequencies and the possibility of on-die errors are no longer negligible.

Unlike ECC, on-die ECC can not correct bit flips that happens outside of die (e.g. bit flips happened between the RAM and CPU).
 
R

RolloZ170

Well-Known Member
Apr 24, 2016
5,416
1,632
113
codgician said:
It is needed because DDR5 RAMs usually have very high clock frequencies and the possibility of on-die errors are no longer negligible.
Click to expand...
they say due to smaller processes (nanometer) the very close data is not safe without.
and we don't need another rowhammer issue.
 
unwind-protect

unwind-protect

Active Member
Mar 7, 2016
422
157
43
Boston
I think onboard ECC without reporting is making things worse, not better. Since you cannot test these things without internal ECC anymore you could be sitting on a DIMM with a permanently flipped set of bits and get instantly hosed on another random error.
 
S

Styp

Member
Aug 1, 2018
69
21
8
codgician said:
On-die ECC is not equivalent to ECC RAM.

From my understanding, on-die ECC happens within each RAM die. It is needed because DDR5 RAMs usually have very high clock frequencies and the possibility of on-die errors are no longer negligible.

Unlike ECC, on-die ECC can not correct bit flips that happens outside of die (e.g. bit flips happened between the RAM and CPU).
Click to expand...
I was referring to 'confusing' because most online stores don't know how to categorize them, so it is difficult to find the correct RAM currently.
Some sites I checked have faulty or incomplete categorization.

Am I right the only speed currently available is DDR5 4800 CL40?
Did somebody see a 48GB stick in the wild?
 
M

MichalPL

Active Member
Feb 10, 2019
189
25
28
Talking about "true" ECC DIMMs ?

EDIT: DDR5 unbuffered and DDR5 registered is different.

DDR3 and DDR3 ECC can work on same motherboards (LGA1366/2011/some 2011-v3) and CPU (for example E5-2678 V3).
DDR4 and DDR4 ECC can work on same motherboards (LGA 2011-v3/3647/...) and CPU (for example also E5-2678 V3)

EDIT: DDR5 unbuffered and DDR5 registered is different.

ECC DDR5:
1682459286536.png
Non-ECC DDR5:
1682459300076.png
 
Last edited:
S

Styp

Member
Aug 1, 2018
69
21
8
It's crazy that even in this community, there are still discussions about non ECC (on-die) vs ECC UDIMMs.
The terminology is just bad, so bad.
 
R

RolloZ170

Well-Known Member
Apr 24, 2016
5,416
1,632
113
Styp said:
It's crazy that even in this community, there are still discussions about non ECC (on-die) vs ECC UDIMMs.
The terminology is just bad, so bad.
Click to expand...
its easy to indentify if you get the RAM in your hands, just count the chips.
edit:
DDR5 UDIMM ECC is only availabe in 2x (32+4) = 72bit.
simply check the pinout
 
Last edited:
  • Like
Reactions: Styp
A

abufrejoval

Member
Sep 1, 2022
39
10
8
RolloZ170 said:
they say due to smaller processes (nanometer) the very close data is not safe without.
and we don't need another rowhammer issue.
Click to expand...
I keep hearing from the inventors of the Rowhammer attack that it can only get worse because of physics: the closer the cells, the higher their effect on each other. They constantly keep finding new attack patterns and additional potential with new technology.

ECC completely stopped being a Rowhammer protection beyond DDR3 and it's sole remaining quality is reducing (not eliminating) the probability of memory errors having an effect or going undetected.
 
A

abufrejoval

Member
Sep 1, 2022
39
10
8
Styp said:
Yeah but this is DDR4 with AM4 - I am looking into AM5 with DDR5, which is confusing at the moment anyway with the on-die ECC terminology.
Click to expand...
The smaller and closer the DRAM cells, the more suceptible they are to cross-talk and energy events. On-die ECC was considered necessary to compensate the additional risks created from shrinking cell dimensions, much like flash generations built extra layers of checksums and block management to deal with the increase likelihood of defects.

It might actually result in DDR5 being less fragile than plain DDR4, but it doesn't reach the level of off-die ECC, which remains an extra effort, cost and security.

When I was faced with a similar decision a year ago, my problem was that DDR5 was crazy expensive, DDR5-ECC was simply not available at any price and W680 boards with DDR4 support never in stock. Supermicros with W680 and DDR5 were listed, but I didn't just want to purchase 128GB of non-ECC DDR5 RAM at 200% DDR4-ECC prices only to wait until DDR5 ECC came around. And then mainboard vendors didn't commit on Raptor Lake compatibility for their W680 mainboards.

So I went with Ryzen 5000, one 5800X3D, one 5950X and 128GB of DDR4-3200 ECC each, which has become as cheap as non-ECC DDR5 recently.

Of course that's now no longer top notch, especially for peak scalar power, but my workloads don't depend on that as much. Stability and long-term 24x7 power consumption as well as overall availability made me switch to AMD for that round and it has delivered so far.
 
S

Styp

Member
Aug 1, 2018
69
21
8
abufrejoval said:
The smaller and closer the DRAM cells, the more suceptible they are to cross-talk and energy events. On-die ECC was considered necessary to compensate the additional risks created from shrinking cell dimensions, much like flash generations built extra layers of checksums and block management to deal with the increase likelihood of defects.

It might actually result in DDR5 being less fragile than plain DDR4, but it doesn't reach the level of off-die ECC, which remains an extra effort, cost and security.

When I was faced with a similar decision a year ago, my problem was that DDR5 was crazy expensive, DDR5-ECC was simply not available at any price and W680 boards with DDR4 support never in stock. Supermicros with W680 and DDR5 were listed, but I didn't just want to purchase 128GB of non-ECC DDR5 RAM at 200% DDR4-ECC prices only to wait until DDR5 ECC came around. And then mainboard vendors didn't commit on Raptor Lake compatibility for their W680 mainboards.

So I went with Ryzen 5000, one 5800X3D, one 5950X and 128GB of DDR4-3200 ECC each, which has become as cheap as non-ECC DDR5 recently.

Of course that's now no longer top notch, especially for peak scalar power, but my workloads don't depend on that as much. Stability and long-term 24x7 power consumption as well as overall availability made me switch to AMD for that round and it has delivered so far.
Click to expand...

Which chipset are you using? I think that is an interesting option, to be honest.
 
You must log in or register to reply here.
Top Bottom