RAM booting issue

[Elrau]

Hello,

I have a weird issue with my build, I hope someone will have an idea.

I'm based on a Supermicro X11-SSL-F with Kingston Server Premier 8Gb Hynix sticks, let's call them RAM1/RAM2/RAM3.

Initially, when I plugged RAM1 + RAM2, I have the"x08=mrcWriteLevelingError" with a bip sequence from my motherboard. After some research I tried several things like disabling the SMBus to PCI but nothing changed.

After that, I discovered the following :

Only RAM1 -> Boot perfectly
Only RAM2 -> Same error

I was able to confgure and run TrueNAS only from RAM1 but it let me think RAM2 perhaps had a problem.
I just received this morning a replacement, RAM3, and I have the same issue :

Only RAM1 -> Boot perfectly
Only RAM2 -> Same error
RAM 1 + RAM3 -> Same error
Only RAM3 -> Same error

But at this point, having two broken sticks seems unlikely and at the same time one stick works.

Did you already see such a behavior ?

Thank you.

 

[Tigersharke]

I am no expert but have you considered that possibly the socket(s) may have issues? Maybe they are dirty or there is a flaw? Check that all devices are seated completely, maybe remove them and re-install the cards etc that may be involved.

I know there is a something about matched ram, which would be a pair of ram which is intended to work together, but I am not sure how critical this is though I have avoided using ram dimms which were not sold as a multi-unit pack.

I can only assume your RAM1, RAM2, RAM3 are perfectly identical with frequency size and timings, disparity among them could be an issue I suppose.

 

[Elliot Dierksen]

I would also look at the CPU sockets to see if there are any bent pins there. No guarantees, but as @Tigersharke mentioned this sounds more like something on the motherboard than the RAM sticks.

 

[Elrau]

Thanks for your answers.

It was a rookie mistake to buy them separately yes, didn't know it was important in that case, used to "lambda" RAM for normal/gamer setup.

I unmount everything and CPU pins seems fine but I'm not 100% sure about one. I tried to fix it but same error with the tests.
What is misleading me is, if it's related to CPU, I'm expected to have the same behavior with every RAM stick. Perhaps I'm missing some knowledge here.

I'm started to think I should go with the working stick as I don't need much RAM on my set up

 

[joeschmuck]

You are missing some very important data... Exactly which slots were you populating? This can prevent some motherboard from work at all.

The slots are to be populated in the following order: DIMMB2, DIMMA2, DIMMB1, DIMMA1. Is this how you have them installed? With all three modules installed, slot DIMMA1 (closest to the CPU) should be empty. I using only two sticks, both blue slots are filled.

 

[Elrau]

Yes I'm using this order according to the documentation (in my case A1 is the closest)

 

[joeschmuck]

(in my case A1 is the closest)

Ageed.

To make sure I understand correctly, the stick identified as RAM1 works in Slot? by itself?
RAM2 placed into that same slot, and still the only stick of RAM fails to work?
RAM3 has the same results as RAM2?

If you are plugging into DIMMB2 slot then install RAM1 stick. Run MemTest86 on that one stick, make sure it passes the test 4 times.
If it passes the test then my conclusion is RAM2 and RAM3 are bad or incompatible.

Before you test anything, exactly what is the part number on each stick (include all numbers/letters). Just want to look up the specs on those sticks to see if two are just not compatible or if they are bad. In some situations you can underclock the RAM speed a tiny bit and get a stable system. Not saying this is the case here but it could be.

 

[Elrau]

I'm not sure about if it's the correct info but here it is

RAM 1 : ETMH09B2390 // 0000010117536-T000444
RAM 2 : ETMH09B2340 // 0000010159136-T000293
RAM 3 : ETMH09A2390 // 0000010111864-T000369

It seems I have a mix of 3 ETMHxxxxxxx, close but not equal

 

[joeschmuck]

The numbers provided are not the part number, they look to be more of an ID of the chips used and manufacturing date.

The following link is a RAM decoder for Kingston, see if there is a match.

Kingston® Memory Part Number Decoder - Kingston Technology

Learn how to read Kingston® memory part numbers including Kingston FURY™, Server Premier™, ValueRAM®, HyperX®, DDR5, DDR4, DDR3, DDR2, and DDR memory product lines to help you identify modules by specification.

www.kingston.com

What I came up with is: ETMH = Hynix, 09 = 9 ICs per stick, B = November, 23 = 2023. I have no idea if this accurate, take it as a possibility unless you purchased these 5 months ago or they were used.

 

[Elrau]

For all of them it's KSM26ES8/8HD, I don't have a correspondance for the last D in Kingston Server Premier DDR4

I also have this ID and I think date ? (YYMD ?)
RAM 1 : ID 0U32 // DC 2344
RAM 2 : ID 0U36 // DC 2346
RAM 3 : ID 0U13 // DC 2347

 

[joeschmuck]

Here is your RAM spec sheet: https://www.kingston.com/datasheets/KSM26ES8_8HD.pdf

Yours -
Capacity: 8GB
Type: 288-Pin DDR4 SDRAM
Speed: DDR4 2666 (PC4 21300)
Voltage: 1.2V
Dimm Type: ECC Unbuffered UDIMM
CL: 19

I looked up on the QVL and it gave me this Supermicro part MEM-DR480L-HL01-EU26, which is identical, but the Kingston RAM was never officially tested by Supermicro. I think you are fine to be honest.
Capacity: 8GB
Type: 288-Pin DDR4 SDRAM
Speed: DDR4 2666 (PC4 21300)
Voltage: 1.2V
Dimm Type: ECC Unbuffered UDIMM
CL: 19

What Next:
1) Using the one stick that works, run Memtest86 for 4 complete passes.
2) If that works, no failures, remove it and one last time, place the other sticks in one at a time and see if either will run.
3) Assuming they do not work, if you just purchased these new, return them for a replacement. Or you can RMA these as they often have a lifetime warranty.
4) If they by magic work, test each one out like in step 1. Make sure they pass. If they do then they are good as well.
5) If the two sticks pass and you place them into the other slots and it takes a dump again, there could be a CPU pin issue, but I'm fairly certain the CPU is a ball, not pin type which means your motherboard would be damaged.

The point is, if one passes and the others do not, it must be the RAM that has failed.

Hopefully I have made this clear, sometimes I ramble.

Best of luck to you.