ECC Ram Support Question?

"Athlon II and Phenom II families apparently support ECC only if it is un-buffered AND un-registered"

-Some Guy from another forum i found on google
 
I believe they all support it if the bios supports, UNREG, UNBUFFERED only.
 
I guess the next question is, does ECC ram do anything useful in a home server environment if it is not unregistered and unbuffered?
 
I guess the next question is, does ECC ram do anything useful in a home server environment if it is not unregistered and unbuffered?

If its "not unregistered and unbuffered" and you use it in an AMD system it does do something - it prevents the system from booting at all. :).

The AMD memory controller supports unregistered, unbuffered ECC DIMMS. All this means is that the signalling tools used improve timing (registration) and signal levels (buffering) are not supported by AMD. These are physical properties of the DIMM and if you use the wrong type it just won't work at all - think of it like trying to use DDR3 memory in a system that only supports DDR2.

However, ungregistered and unbuffered ECC DIMMs still provide the extra bits required for ECC. If the CPU supports them and the memory controller knows how to do the ECC calculations (and all Athlon- and Phenom-series CPUs do) then the CPU supports ECC.

The problem arises because some MB manufacturers either (a) don't include traces on their MB for the extra ECC bits or (b) don't provide the hooks in the BIOS to turn it on. They do this either to save a few (fractions of) pennies or to avoid competing with their "server class" motherboards that support ECC (and cost more...).

Many AMD motherboards do offer support for ECC. YMMV. Check the forums to confirm support as the manufacturers will often tell you no even if the facts are yes (e.g., Biostar will almost always tell you it is not supported even though most of their AMD motherboards offer full support).

You can always test. Get the AMD MB/CPU you want. Install Unbuffered, Unregistered ECC DIMMS. Fire up Memtest86+. If you've got ECC support, it will tell you (almost always yes for an AMD CPU if the MB traces are there). It will also tell you if it is "enabled" or "disabled" in the memory controller. If its not "enabled", check your BIOS options to see if you can enable it and if you can, test again.
 
why wont they just make all ram ECC and registered? Wouldnt that make computers more stable?
 
One of the procs in question would be mated to an ASUS M4A88T-M LE AM3 AMD 880G HDMI Micro ATX AMD Motherboard.

This board supports DDR3 1866(O.C.)/1333/1066 ECC,Non-ECC,Un-buffered Memory (AMD AM3 100 and 200 series CPU support up to DDR3 1066MHz.)

It would be for a home-based RAIDz2 ZFS solution.
 
why wont they just make all ram ECC and registered? Wouldnt that make computers more stable?

yes, it would make computers more stable, over time especially. memory breaks down more and more over time, so ecc helps more the older your computer gets.

differentiation is how profits are maximized in business. that is the answer to your first question, not memory bandwidth.

by selling multiple kinds of memory, the memory companies make more money. by selling multiple kinds of chips and boards, the chip and board companies make more money. AMD is the underdog, so they give away ECC support for free. intel charges extra for it by disabling it on less expensive platforms.

microsoft published a paper when vista came out, requesting everyone use ECC to prevent hardware errors (because microsoft gets blamed for those errors). as you can see, hardware manufacturers did not respond positively.
 
I can confirm it works fine, using a Phenom II X3 720 BE with 8GB of DDR3-1333 ECC unreg/unbuf for my server testing/virtualization lab. Tangible benefits? Depends on how much magnetic interference you have around your machine and how long you plan on keeping this machine around with high uptimes.
 
On $$$ work-stations and servers that I've used, all memory ECC corrections were logged, so we could replace the ram or machine (motherboard) when corrections started popping up a lot. Not sure how the logging/notification works on a Windows or Linux box though.
 
Not sure how the logging/notification works on a Windows or Linux box though.

On a server board these show up as machine check exceptions. On linux you see these in mcelog. I used to be very familiar with this because my dual core dual processor Opteron 285 system used to flag several memory machine check exceptions (single or multi bit correction) daily when both processors were installed.
 
AII and PII support it. I used ECC in many AMD powered Openfiler and Freenas builds for clients wanting high performance low cost NAS solutions. I would have used intel as well however their lower cost desktop chips do not support ECC and AMD does.
 
why wont they just make all ram ECC and registered? Wouldnt that make computers more stable?

You would be surprised how often RAM encounters bit errors. However ECC is a little slower compared to non ECC and also cost a little bit more so most motherboard manufacturers do not implement the function into desktops due to that whole: Lets deliver the absolute minimum cost of manufacturing PC with maximum volume and profit mentality to the consumer.

Here is a study conducted which is good reading for DRAM errors:
http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf
 
So which one to pick for a server environment?

X2 560
$95
3.3 Ghz x2
80W
L3: 6MB

X4 640
$100
3.0 Ghz x4
95W
L3: N/A
 
Back
Top