Since you are new here - I suggest re-reading the first post of this topic.
Originally Posted by gpacster
Note: *SERVER* ram is typically 'Registered' which means electrically it works slightly differently then "regular" desktop system ram.
If I were you - check the model # from that ram to determine what kind it is, if its some kind of Unbuffered DDR2/3 - then you're OK.
The magic keywords here are UNBUFFERED - ECC or not.
AM3/AM3+ ECC Support
Hate to bump my own thread; but new information is here!
AM3+ Boards that support Unbuffered ECC
-AMD 900 Series chipsets (970/990)
What is the recommend method for watching RAM for ECC errors?
Are they logged in /var/log/messages? Or do I have to watch /sys/devices/system/edac/mc/?
What do you use?
A single bit error will generate a Machine Check Event which is shown in dmesg and captured by your system logger. Some BIOSes have a log of them too.
A non-correctable error will cause a Machine Check Exception and lead to system halt by default (but this can be overridden).
ECC is unofficially but fully supported with a lot of tweaks in BIOS - like memory scrubbing etc. I use it as a NAS platform (because of 10 Sata3 ports).
Evidently its something ASUS missed. This information is usually from a manual / BIOS - but ASUS distinctly listed the 800G as not supporting ECC on the board's webpage... probably a typo or mistake.
Originally Posted by Cres
I've heard about other Gigabyte boards supporting ECC, good that the 990FX series is there - is this a modded BIOS or a later revision?
Two of our developers' workstations have ASUS mainboards with 880G chipsets (ASUS M5A88-M EVO), and they support ECC just fine. One curious difference is that with a Phenom X6, Linux' EDAC subsystem reports S4ECD4ED in /sys/devices/system/edac/mc/mc*/csrow*/edac_mode, and with an FX-4xxx CPU reports only SECDED instead. Does anyone happen to know why that is?
Last edited by colo; 04-11-2012 at 08:19 AM.
Yes, I just ordered some ECC/unbuffered (ECC/UDIMM) for my Proliant Microserver.
I decided to test them in my GigaByte 970A-UD3 motherboard and they worked. They were still manually configured for my old 1600MHz desktop RAM, but I could see an ECC [enabled] option in dram configuration.
I loaded system defaults in the BIOS and rebooted so they would auto setup themselves and when I went into the BIOS it said ECC Support was disabled and about a half-dozen ECC options were all grayed out (set to [disabled].
After enabling the ECC option I was able to enable the top halve, while the bottom half wanted me to set timings that I have no idea what they should be. Hopefully somehere can explain what they are what they should be.
Here's a screen shot of BIOS:
If you know of any links for extra help/explanations that would be great
MCE sound like machine check exception
Chipkill set to off since it is iirc. only on very expensive IBMs, even better than pure ECC, but you need the expensive HW.
Dunno about redirection and all the single scrubbing options, would have to look that up on the net myself. I guess it is the part that goes regularly around and checks for errors and corrects them and stuff. Though I thought something like that would also be done by the kernel. Basically that is one core component of ECC so you might want to turn it on unless strange things happen afterwards.
All the bottom half is good old traditional stuff. Normally you just set it to SPD (that is a small chip on you RAM, the normal + 1 additional membanks - and then a tiny thing, which does the timing for you). So SPD will set it and you should be fine. These options are only meant in case there is no SPD or you want to manually speed up things but then you have to carefully try step by step to keep your system stable. And since stability is your main concern when using ECC anyway you should just leave it to the SPD.
Tags for this Thread