ECC Error Injection

[awitko]

I have a new PC build on Supermicro X11SCA motherboard with Intel XEON E-2288G and 64GB of ECC Ram.

The only reason I paid for the pro version of Memtest64 is that it offered the additional ECC error injection feature.

After installation, Memtest64 reports that although ECC is enabled, injection is not available on my system.

I see that ECC Error Injection is in the E-2200 Datasheets so it is apparently available for this CPU.

No explanation is provided as to why it is not available. I think I saw somewhere mention that sometimes it is disabled in BIOS by the manufacturer but without any details. Any information as to what are the relevant BIOS settings and whether they are available for modification on the Supermicro AMI BIOS to make ECC error injection available?

I would like to test my memory with ECC error injection. Any comments as to whether this might be doable? If this is not doable, I would like a refund if possible.

Also I ran the standard memory test through most of 4 passes overnight (10 hours). Had to end it prematurely part way through test 13 on the 4th pass this morning and it was reporting 0 errors. Does that mean zero errors after ECC correction or would it report ECC correction counts in the on screen status display (which I took a snapshot of). Otherwise some memory sensitivities may be being masked by ECC and I would imagine I should probably run test with ECC disabled to expose that potential problem. Welcome any thoughts on this and other suggestions to thoroughly test the memory.

awitko

 

[Solaris17]

Hm, I don’t know about this feature specifically but I know that ECC generally holds cpu/mobo requirements. I have had good luck speaking with supermicro support regarding my server boards, I would genuinely ask them.

 

[awitko]

Thanks Solaris17. Yes ECC generally requires a Xeon Processor and the motherboard is specified to work with ECC and it works perfectly with the ECC memory. ECC is supposed to correct one bit errors seemlessly, but they are relatively rare and ECC injection is meant to artificially insert errors to test that the ECC memory detects the error and can correct it. Again the Intel manual states that this processor has that capability but memtest64 says it is not available. I have been in discussion with Supermicro about other issues and am submitting an email about this. But to date they have been hesitant to provide a useful response to anything that touches on a third party. I could imagine they will say we don't know why it doesn't work you have to ask memtest64. And memtest64 will say the same about following up with Supermicro. I am hoping someone on this community can help...

 

[Ware]

My understanding is that this is disabled in BIOS because it can only cause problems. You would need some kind of "testing" BIOS, since this is probably disabled in production BIOS. I think you can trust that Supermicro tested the ECC functionality.
I don't see Coffee Lake support listed here: https://www.memtest86.com/features.htm

 

[Solaris17]

Thanks Solaris17. Yes ECC generally requires a Xeon Processor and the motherboard is specified to work with ECC and it works perfectly with the ECC memory. ECC is supposed to correct one bit errors seemlessly, but they are relatively rare and ECC injection is meant to artificially insert errors to test that the ECC memory detects the error and can correct it. Again the Intel manual states that this processor has that capability but memtest64 says it is not available. I have been in discussion with Supermicro about other issues and am submitting an email about this. But to date they have been hesitant to provide a useful response to anything that touches on a third party. I could imagine they will say we don't know why it doesn't work you have to ask memtest64. And memtest64 will say the same about following up with Supermicro. I am hoping someone on this community can help...

I know what ECC is, as I said I run it on my boards, you are not the only one that uses server components. I was speaking specifically about "Error Injection". Its odd you have not had good experience with them, they have fielded some pretty technical questions for me. I would probably leave mention of memtest out of the equation and ask about the feature specifically. Maybe it will yield a better response.

 

[awitko]

Thanks Solaris17. I was not making a judgement - just reading your statement where you said ECC has motherboard requirements and took that to suggest you did not understand details of it. My mistake. Good suggestion regarding not mentioning memtest to Supermicro. I have had mixed experiences with supermicro. The person who I have dealt with who specializes in the X11SCA board there told me it was normal that the BIOS did not recognize the 970 Pro NVMe drive but it turned out to be a socket problem on the board. He pushed me to contact Samsung.

Thanks ware. I understand why you wouldn't have this enabled in a production system, but there are jumper related to "manufacturing mode" for example. I don't see why they could not make this conditional on a mode like that if they did not want it in the production system.

I understand it doesn't explicitly list Coffee Lake but I would imagine the injection process is the same across Intel processors so if it works for older generations it should work for the newer ones. The requirements are pretty basic so once they work out the scheme to do this I don't see why it should change. Not certain to be the same, but likely.

Alex

 

[R-T-B]

Just to be clear, are you talking about memtestx86, a bootable product we do not produce, or memtestx64, our freeware windows based memory test?

I don't think a pro version of Techpowerups memtestx64 exists...