Hacker News Reader

Comment by p1mrx | Hacker News Reader

I have an Intel C226 chipset with an i3 CPU, ECC RAM, and Linux, but as far as I can tell, there is no way to determine whether ECC is actually working.

I've heard of people collecting known-bad DIMMs for this purpose, or trying to blast their RAM with heat or radiation.

Why does this have to be so difficult?

View on HN

So this is the datasheet for the C226 (https://www.intel.com/content/www/us/en/products/docs/chipse...) and you have an i3-8100 or equivalent that does support ECC, then the BIOS will configure how it wants ECC errors reported. On my Ryzen board they are reported via SMI events and there is a chip that you can talk to on the SMBUS that will tell you how many events it has seen since the log was reset.

In my experience getting the information isn't "difficult" so much as it isn't "common". On my FreeNAS device (which has the Xeon) it also has an IPMI controller which lets me ask it over the network how many ecc errors have been seen. (and it can act like a remote console as well but we'll leave that for another time :-))

So find out how the motherboard is tracking ECC, then find the tool that talks to that system to ask it for the logs of ECC errors.

Well, non-ECC RAM can report "0 errors" for less money.

The problem is how to determine whether your system is actually capable of detecting/logging errors end-to-end. There needs to be a standard way to inject an ECC error and see what happens.

There are also BIOS setups that "mask" soft or correctable errors since nothing bad happens when they correct the problem. I've always preferred to not mask those errors as it was always a good thing to note on server hardware on clusters that if their soft error correction rate started rising then a DIMM was going to need to be replaced.

Overclock it one step at a time until it gets mad.

> Well, non-ECC RAM can report "0 errors" for less money.

"0 errors" is very different from "ECC not available," and that's a distinction I'd expect the operating system to be able to make.

For example, here's how Linux shows ECC and non-ECC memory reports:

    # ECC
    $ sudo edac-util -v
    mc0: 0 Uncorrected Errors with no DIMM info
    mc0: 0 Corrected Errors with no DIMM info
    edac-util: No errors to report.

    # Non-ECC
    $ sudo edac-util -v
    edac-util: Error: No memory controller data found.

It doesn't have to be that difficult. A crude yet effective method is to overclock the RAM beyond a reasonable frequency and a handful of correctable errors should pop up in no time.

In case the OS/BIOS does not support overclocking or fails report memory error properly, certain data pins of the dimm slot could be physically shorted with metal wire to add a persistent stuck bit. Without ECC the system will not boot. I've seen it done on multiple occasions but web search is drawing a blank for the moment. IIRC it was being discussed a lot when first gen Ryzen cpu came out and people were arguing left and right whether ECC support is present on motherboard X.

Finally some memtest software claims to be able to inject errors on a software level. I can't vouch for them as I have very little experience but it would be very simple to find out.

You can also play with voltage settings if those are more accessible than speed and timing.

Or play with rowhammer tests.

As far as I know Intel doesn't support ecc on their consumer platforms. You need a xeon for that.

Intel supports ECC on i3, but not i5-i7. At least that was the case for the Haswell generation; I haven't checked recently:

https://ark.intel.com/content/www/us/en/ark/products/77495/i...

i3-8100 cpus support ECC

See the comment from viraptor with the link to edac-utils.

Reply on news.ycombinator.com