Search results

Regardless it is good to know that an X13SAE(F) will boot a 14th gen out of the box (at least with 2.1).

@brained can't speak to the issues with virtualized IOMMU you are having, but the split-lock warnings are a VM hypervisor problem. Split-locks by default are logged in the kernel, because they are highly discouraged at this point. There is a setting you can change to silence this, but this is...

X13SAE. 4 sticks and 128 GB memory total.

Agree with both of you @JanR and @RolloZ170.

There is no compatible EDAC driver for the W680 chipset for Linux yet. I think severe memory errors will get logged through ACPI and/or SMI/machine check alerts though.

I wanted to alert folks that Supermicro has released BIOS version 3.0 for the X13SAE(F). They did not provide a changelog, but my guess is that it enables compatibility with 14th gen Raptor Lake CPU's. I also noticed it added higher speeds in the memory speed selection drop down, up to 5600 MHz...

Ah. You are right about who enforces the frequency reduction standard based on the load factor. In this case it is the CPU manufacturers, because the memory controller is actually on the CPU. I would advise that if you don't follow the rules specified by JEDEC and the CPU, then the reliability...

See https://forums.servethehome.com/index.php?threads/lga-1700-alder-lake-servers.35719/post-370867. The JEDEC spec requires that the speed decline as the number of slots per channel/dimms per channel/ranks per DIMM increase. Most people do not know this. Supermicro follows the JEDEC standard...

I have four Micron ECC sticks for 128 GB total working fine in my X13SAE, but I am running them at JEDEC spec speed, which means 3600.

You mean the X13SAE-F right? I have the non-F model, but hopefully this will be useful. This is with OpenSUSE 15.4 (5.14.21-150400.24.60-default) The non verbose version: 00:00.0 Host bridge: Intel Corporation Device a700 (rev 01) 00:01.0 PCI bridge: Intel Corporation Device a70d (rev 01)...

Make sure you are using the BIOS for the correct board. The X13SAE and X13SAE-F (with the onbaord BMC) are different. I was having a problem when I originally built my workstation, and it turns out I was trying to apply the X13SAE-F bios ver 2.0 to my non-F board.

Anyway. I am pretty convinced we are getting real ECC with these modules, even if the reporting is a little screwy.

It is really crazy isn't it? Kingston's own spec sheet for those modules says 72... https://www.kingston.com/datasheets/KSM48E40BD8KM-32HM.pdf

The socket dimensions are standard and the clearances appear to be just as good as with gaming boards. I am using the Noctua NH-U12A cooler.

Also I have been running my S13SAE with a PL1 of 225W for several months now. Rock solid. You can use the PL overrides in the BIOS to do this (overriding the default). Just make sure you have proper cooling in place.

ECC modules have historically been 72 bits wide. It is only with DDR5 that we are seeing a 2x40 = 80 bit wide option. Both are "true" ECC. Both offer 1 bit correction. The confusion is around the reporting of errors (multi-bit detection). I also have a feeling the tools are misreporting things...

On the Supermicro vs Asus. It doesn’t surprise me that the Asus board outperforms the Supermicro board, because Supermicro tends to be very conservative with their timings in favor of stability. They also strictly enforce the default PL1 and 2 power limits whereas Asus may not. You need to check...

It is more complicated than that. See the long LGA 1700 Alder Lake "Servers" thread. There actually are 72 bit wide and 80 bit wide DDR5 ECC modules. It appears as if most, if not all unbuffered DDR5 ECC modules are actually 2x36 and not 2x40. The registered DDR5 ECC modules appear to be 2x40.

That is a good idea. I took close-up pictures of the modules, but only the side that has the serial number label, and it obscures the parts. I will have to find a time to shut the machine down and take out a module to do this. Right now it is busy... :)

I need the 128 GB of ram for what I am doing, so ... 3600 it is. I do not want to play with fire by trying to run it above spec for fully populated.