Mixing RAM brands

JCH

Member
Jan 2, 2017
42
6
8
23
Okay... so I usually never mix RAM but I bought a combo on eBay that only included 2x 16GB DDR4-2666 ECC unbuffered RAM by Crucial. I had a spare 2x generic kit that’s also DDR4-2666 ECC unbuffered that works when mixed in the server.

I also had 2x Kingston 16GB DDR4-2666 ECC unbuffered RAM as well.. so I tried mixing that with the Crucial sticks, to no avail. Am I stupid and am not seeing something or does my board just not like these two brands mixed together?

The board POSTs with the individual brands themselves just fine.

I'm getting this when mixing the two together:
0x19=mrcRoundTripLatencyError
0x0C=mrcMiscTrainingError

Crucial: CT16G4WFD8266
Kingston: KSM26ED8/16ME

 
Last edited:

JCH

Member
Jan 2, 2017
42
6
8
23
i am mixing 3 different brands on my epyc 7551 setup (h11dsi, rev.1.01).

try to update bios ???
BIOS is already on latest from Supermicro, R1.2c

Just weird that it works when I mix with generic RAM but not with these Kingston sticks
 

blinkenlights

Active Member
May 24, 2019
133
48
28
Could be the motherboard, or could be a faulty module. Do you have all of the memory settings configured to "Auto" or similar? I think that error message indicates mismatched CAS latency (CL), but according to the Kingston spec sheet it also is CL19: https://www.kingston.com/datasheets/KSM26ED8_16ME.pdf

In general, any proper (JEDEC compliant) memory will work as long as ranks, banks, and latency match on a channel. Frequency is best practice, but don't worry - the motherboard should downclock to the slowest module speed ;)
 
  • Like
Reactions: JCH

JCH

Member
Jan 2, 2017
42
6
8
23
Could be the motherboard, or could be a faulty module. Do you have all of the memory settings configured to "Auto" or similar? I think that error message indicates mismatched CAS latency (CL), but according to the Kingston spec sheet it also is CL19: https://www.kingston.com/datasheets/KSM26ED8_16ME.pdf

In general, any proper (JEDEC compliant) memory will work as long as ranks, banks, and latency match on a channel. Frequency is best practice, but don't worry - the motherboard should downclock to the slowest module speed ;)
I actually got this error too:
0x0C=mrcMiscTrainingError

HOWEVER, I just tried again this morning and the system posted..... I have no idea why it decided to work now. The system had been running with the Crucial RAM and I just powered it off, added one of the Kingston RAM sticks at a time and it posted both times. I didn’t change a thing from when I tried it the first time :eek:

BIOS shows 64GB as well. I am running memtest86 to test these sticks just to be sure they won’t crap out on me
 

ReturnedSword

Active Member
Jun 15, 2018
198
48
28
Santa Monica, CA
Generally mixing RAM isn't an issue nowadays. The BIOS should default to the lowest common denominator, however it might take a failed POST or two (or three) for the auto-detection to work. Both your Crucial (Micron) and Kingston sets utilize Micron DRAM, however the PCB design is a bit different. This should not matter though.
 
  • Like
Reactions: JCH

JCH

Member
Jan 2, 2017
42
6
8
23
Generally mixing RAM isn't an issue nowadays. The BIOS should default to the lowest common denominator, however it might take a failed POST or two (or three) for the auto-detection to work. Both your Crucial (Micron) and Kingston sets utilize Micron DRAM, however the PCB design is a bit different. This should not matter though.
So with this in mind, I went ahead and tested the system more. First off, it passed memtest86 with flying colors, no errors in all 4 passes.

I went ahead and decided to reseat the RAM and change the order that they were in. Originally, they were ordered as, crucial, kingston, crucial, kingston. I changed the order to kingston, crucial, kingston, crucial. Just like that, the board wouldn't POST anymore. Removing the fourth stick (crucial) remedied the problem, but as soon as it was inserted the system no longer POSTs. I power cycled 3-4 times and got the same error all 4 times. It would not POST.

I went ahead and booted each of the four sticks in each slot on the board just to be sure I didn't have any dead sticks nor DIMM slots and each time the system POSTs, so nope.

Lastly, I changed the order of the RAM back to, crucial, kingston, crucial, kingston, and the system did NOT POST. Again however, I power cycled the system, and on the second attempt the system POSTed, much like you had said where it probably needed some time to "auto-detect".

All-in-all, I might just swap out 2 of my sticks for the same brand of the other 2.. I want to put this server into production and don't want to risk having any problems
 

gb00s

Active Member
Jul 25, 2018
169
45
28
Malta
In my dumb opinion this is not a Dimm issue but an initialization issue of the ram from the motherboard. I remember having similar issue with an Intel S2600CW. Used it with 2 different brands and significant different timings. Once I got the board to initialize everything it ran flawless. Changing just one ram stick in order, let the next boot fail. Two more cycles and everything was fine again.
 

blinkenlights

Active Member
May 24, 2019
133
48
28
In my dumb opinion this is not a Dimm issue but an initialization issue of the ram from the motherboard. I remember having similar issue with an Intel S2600CW. Used it with 2 different brands and significant different timings. Once I got the board to initialize everything it ran flawless. Changing just one ram stick in order, let the next boot fail. Two more cycles and everything was fine again.
@gb00s that is not a dumb opinion at all. I thought I read this was a Supermicro X11 board.. my experience with X9/X10/X11 boards is they are finicky on the best days. Just had an X11SSH-F (sister board to this one) crater itself yesterday evening after updating the BIOS and IPMI firmware. Solid boards once you get them working, just leave them alone...
 

blinkenlights

Active Member
May 24, 2019
133
48
28
With 2133 mem my SM boards where usually (i.e. most) quite solid, but as soon as 2666 comes into play...
@Rand__ I have experienced that with Supermicro boards, although my server build (Gold 5218 on an X11SPi-TF) seems to be rock solid with Crucial 2666 LRDIMMs. I have also had far less issues with LRDIMM than RDIMM, and try to use them where possible.