Possible to mix ECC Memory?

ejn1111

Dabbler
Joined
Jul 24, 2020
Messages
21
Hi, I have one 8GB Kingston ECC stick of 2133 ECC 1.2v CL15 for my new build. I wanted to get 16GB in the build so was looking online to purchase another identical Kingston 8GB stick. The price was almost $120 if memory serves me and I found a Samsung 16GB identical spec stick for like $150 so just got that (it was on the Supermicro approved memory list for the motherboard). I was going to sell the 8GB stick or keep as a spare but now wondering if I can just add the Samsung stick for a total of 24GB of RAM? Any issues in doing this with Freenas or the Mobo (4 memory slots)? Appreciate if someone could let me know.
Thanks!
 

ejn1111

Dabbler
Joined
Jul 24, 2020
Messages
21
It would depend on the motherboard, but in general I wouldn't expect this to be an issue.
It's a Supermicro X11SSM-F. Thanks for the reply.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
It's not a great idea, memory should ideally be perfectly matched. However, it happens all the time, and as long as the specs are similar, it's likely to work. To be safe, set it up in memtest86 and let it run a minimum of several days.
 

Yorick

Wizard
Joined
Nov 4, 2018
Messages
1,912
I’ll argue there’s some tribal lore in that cautionary note, likely passed down from non-ECC consumer boards.

I’d trust SM’s guidance and the ECC: If there is an error, IPMI will let you know. A memtest burn in could still identify “infant mortality” in a DIMM, but that’s independent of running different sizes.

The motherboard manual on pg 33f. states:

Populating these DIMM slots with memory modules of the same type and size will result in interleaved memory, which will improve memory performance.

and

Mixed DIMM speeds can be installed. However, all DIMMs will run at the speed of the slowest DIMM.

You’ll want to populate the blue slots first, B2 and A2. You won’t get interleaving, but TrueNAS performance is not gated by memory performance.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
I’ll argue there’s some tribal lore in that cautionary note, likely passed down from non-ECC consumer boards.

Well, I deal with about fifty server boards for every "non-ECC consumer board" that comes thru the shop here, so, no, there's no "tribal lore", and I can tell you that there are absolutely combinations of things that do not work even where you might think they should.

For example, Intel Ivy Bridge E3 CPU's are incompatible with four Elpida based DDR3 DIMM's, but you can use two Elpidas in the first bank and two Samsungs in the second and they're fine.

If you thought all my scolding people to do proper burn-in testing was hypothetical, you're wrong. It's because edge cases absolutely do exist. Some of us do this stuff professionally and we obsess over these things because we want the correct outcome. Since the poster already had the part on hand, there's not really any harm in trying as long as it's tested. I *expect* it will work. I wouldn't bet a hundred bucks on it though.
 

Yorick

Wizard
Joined
Nov 4, 2018
Messages
1,912
So noted :).
 

ejn1111

Dabbler
Joined
Jul 24, 2020
Messages
21
I’ll argue there’s some tribal lore in that cautionary note, likely passed down from non-ECC consumer boards.

I’d trust SM’s guidance and the ECC: If there is an error, IPMI will let you know. A memtest burn in could still identify “infant mortality” in a DIMM, but that’s independent of running different sizes.

The motherboard manual on pg 33f. states:

Populating these DIMM slots with memory modules of the same type and size will result in interleaved memory, which will improve memory performance.

and

Mixed DIMM speeds can be installed. However, all DIMMs will run at the speed of the slowest DIMM.

You’ll want to populate the blue slots first, B2 and A2. You won’t get interleaving, but TrueNAS performance is not gated by memory performance.
Thanks for the advice and appreciate the slot guidance, I was wondering about that.
 

ejn1111

Dabbler
Joined
Jul 24, 2020
Messages
21
It's not a great idea, memory should ideally be perfectly matched. However, it happens all the time, and as long as the specs are similar, it's likely to work. To be safe, set it up in memtest86 and let it run a minimum of several days.
Many thanks for the input jgreco... Do you think the extra 8GB Ram is worth the additional risk/headache? sounds like it may be but thought I would ask your view.
 

Apollo

Wizard
Joined
Jun 13, 2013
Messages
1,458
Many thanks for the input jgreco... Do you think the extra 8GB Ram is worth the additional risk/headache? sounds like it may be but thought I would ask your view.
If successfully tested thoroughly, risk/headache shouldn't have their place.
If you stick to the 8GB sticks (pun intended) and run with 16GB only, then you can keep an eye or your "ARC size" and "ARC Hit Ratio".
Depending on the workload, the types of files and how often data gets accessed, it could take minutes or weeks before seeing drastic changes in the ARC behavior.
If ARC size is mostly constant, it means your system doesn't really need more RAM. If ARC is erratic, then adding more RAM could help.
 

ejn1111

Dabbler
Joined
Jul 24, 2020
Messages
21
If successfully tested thoroughly, risk/headache shouldn't have their place.
If you stick to the 8GB sticks (pun intended) and run with 16GB only, then you can keep an eye or your "ARC size" and "ARC Hit Ratio".
Depending on the workload, the types of files and how often data gets accessed, it could take minutes or weeks before seeing drastic changes in the ARC behavior.
If ARC size is mostly constant, it means your system doesn't really need more RAM. If ARC is erratic, then adding more RAM could help.

Good advice. Assume ARC size and ARC hit Ratio are in the FreeNas Dashboard?
 

Apollo

Wizard
Joined
Jun 13, 2013
Messages
1,458
Good advice. Assume ARC size and ARC hit Ratio are in the FreeNas Dashboard?
Should be under "Reporting" / "ZFS" section.
 

jgreco

Resident Grinch
Joined
May 29, 2011
Messages
18,680
If successfully tested thoroughly, risk/headache shouldn't have their place.
If you stick to the 8GB sticks (pun intended) and run with 16GB only, then you can keep an eye or your "ARC size" and "ARC Hit Ratio".
Depending on the workload, the types of files and how often data gets accessed, it could take minutes or weeks before seeing drastic changes in the ARC behavior.
If ARC size is mostly constant, it means your system doesn't really need more RAM. If ARC is erratic, then adding more RAM could help.

Well put.
 
Top