CPU support for Supermicro X12, X13

[mindio]

The specification should read "ECC support for APUs is only with Ryzen PRO".
And for non-APUs it now gets complicated, with some of the last 5000non-X lacking ECC support, but that happened long after the board was released and manfacturers rarely update product pages.

I can confirm Ryzen 5 3600 is working with MC12-LE0 and ECC RAM:

Now the question of the topic remains, if the lowest spec. LGA1700 CPU with ECC support i5-12500 (?) will work on Supermicro X13SCH-F? :)

 

[ChrisRJ]

I can confirm Ryzen 5 3600 is working with MC12-LE0 and ECC RA:

This is the main misconception we see in this context. There is a fundamental difference between "it starts up and works normally" and "the full ECC functionality is there". Which scenario we are looking at here I cannot say. But I wanted to make people aware of this non-obvious difference.

Anyway, I very much hope that things work for you!

 

[mindio]

This is the main misconception we see in this context. There is a fundamental difference between "it starts up and works normally" and "the full ECC functionality is there". Which scenario we are looking at here I cannot say. But I wanted to make people aware of this non-obvious difference.

Good point. How can we test this?

 

[ChrisRJ]

Good point. How can we test this?

Perhaps I am wrong here. But I cannot imagine how to do this without very special hardware. But Google is probably your friend here ...

 

[Ericloewe]

Memtest86 (the commercial one) claims to support error injection on many platforms - if the system firmware hasn't disabled that option, that is.

 

[bdkjones]

I just built a system using the X13SCH-F board and the Pentium G7400 with 32GB of Micron 4800 UDIMM ECC in a single DIMM. TrueNas Scale is reporting ECC memory on the dashboard, but I can't find any mention of it in the bios. `dmidecode -t memory` reports single-bit ECC, but I don't see any references to it in logs.

The usual "Supports ECC RAM" line is completely missing from the ark page for the Pentium G7400 (as opposed to showing an explicit "No"), but the G7400E (embedded) ark page shows "yes" for ECC support. So I contacted Intel and....they have no idea if the G7400 supports ECC. They're going to get back to me in a few days after "consulting the engineers".

In the meantime, I'm incredibly impressed with the board's efficiency. With a Seasonic TX-700 (700W) PSU and a Samsung 870 EVO 500GB SATA SSD connected (boot disk), this build is idling at 18W measured at the wall. The PSU is 72.6% efficient at that 2% load, so about 30% (4.2W) is power-loss. If you assume another 3-4W for the BMI, the board+chipset+CPU is using around 10W at idle. (Where I live, electricity is $0.60/kWh so I'm highly motivated to keep things efficient.)

 

[Ericloewe]

60 cents per kWh? Oof. Not fun.

 

[nabsltd]

I just built a system using the X13SCH-F board and the Pentium G7400 with 32GB of Micron 4800 UDIMM ECC in a single DIMM. TrueNas Scale is reporting ECC memory on the dashboard, but I can't find any mention of it in the bios. `dmidecode -t memory` reports single-bit ECC, but I don't see any references to it in logs.

All DDR5 has "single-bit ECC" internally. This detects multiple bit errors and corrects single bit errors. It is necessary to allow the RAM to perform at 2x the speed of DDR4 while retaining the same small package and thermal characteristics.

This is all done internally on individual RAM chips, and no information about errors is reported over the DMI bus. True ECC memory uses parity across the set of chips, which ensures that the data leaving the DIMM is valid. This type of ECC is reported and can be logged.

 

[bdkjones]

All DDR5 has "single-bit ECC" internally. This detects multiple bit errors and corrects single bit errors. It is necessary to allow the RAM to perform at 2x the speed of DDR4 while retaining the same small package and thermal characteristics.

This is all done internally on individual RAM chips, and no information about errors is reported over the DMI bus. True ECC memory uses parity across the set of chips, which ensures that the data leaving the DIMM is valid. This type of ECC is reported and can be logged.

Right. I'm aware of the difference and that's why I'm hesitant to trust the TrueNAS dashboard's report that ECC is in-use. Grepping the boot logs for "EDAC" turns up just one line that lists a version (instead of multiple lines reporting ECC).

And I'm pretty sure that `dmidecode` will report ECC memory because it *IS* ECC memory. Although the data with is 64 and the total width is 72, which is exactly as it should be if ECC is active.

Regardless, I'm not convinced that ECC is *actually* working with the Pentium G7400, which is why I contacted Intel to get a definitive answer. My current plan is to switch to a Xeon E-2434. It has a slightly slower clock, but 4 cores instead of 2. It also has hyperthreading, but I'll likely disable that because it's overkill and consumes power. (The lower range E-2414 lacks hyperthreading, but is a 2.6Ghz base clock instead of 3.4 and I do want some processing ability when I need it.)