Also, a couple other comments. When you get the new RAM, try it with
your existing BIOS. Or, at least, examine all the BIOS release notes,
to see what side effects a BIOS upgrade could have. For example, the
1016 has a chipset driver dependency in the OS. If the chipset
drivers are already higher than the stated revisions below, then it
would be safe to upgrade the BIOS to 1016 or higher.
http://support.asus.com/download/download.aspx?SLanguage=en-us&model=A8N-SLI Deluxe
"Please do read the description before updating BIOS
To avoid crashing file system, please do update the chipset driver to below
version prior to this Bios.
For WinXP 32bit system, please download and update chipset drivers V6.65 or later
For WinXP 64bit system, please download and update chipset drivers V6.65 or later
For Win2k/2003 32bit system, please download and update chipset drivers V6.66 or later
For Win2003 64bit system, please download and update chipset drivers V6.66 or later"
As for the ECC comment, ECC is error checking and correcting. When a memory
DIMM has a (ninth) parity chip on it, it allows the detection of multi-bit
errors, and the correction of single bit errors. When a large quantity of RAM
is involved, it provides a means of the user having errors corrected in the
background (for the single bit ones), or an error showing on the screen, to
inform you of the potential for a calculation to be corrupted. Without ECC,
you might not notice that a bit error has occurred in the RAM.
ECC support is required in two places to work. The DIMM has to have the ninth
chip populated on both sides, giving a 72 bit wide data array, compared to the
normally 64 bit wide DIMMs. The chipset also must have the provision for the
extra data bus wires, and have an ECC logic block to do the ECC calculation.
The Athlon64 family has the ECC chipset support built in, so any Athlon64
should be a candidate (check the manual to verify the support is there, to
be absolutely sure).
The Athlon64 has two error correction modes. It has the normal SECDED, which
corrects a single bit in error, in a 64 bit data word. In other words 64 bits
data plus 8 bits parity, allows correcting any single bit in error in the
entire 72 bit quantity (even an error in the ECC section can be fixed). If
two bits are in error, that fact can be detected but not corrected. If more
than two bits are in error, then statistics come into play, as only some error
patterns will be detected, while the undetected ones are no worse than using
non-ECC RAM.
The second mode is called "chip kill". It relies on dual channel configurations
of RAM. It treats the data bus as being 144 bits wide. It fetches data from
both DIMMs, so there is 128 bits of data and 16 bits of parity. Chipkill mode
allows up to four bits in error, clustered together, to be corrected. On
server systems (probably not that useful in your case), where x4 nibble wide
memory chips on registered DIMMs are quite common, it allows a single x4 wide
memory chip to completely die, and the pair of DIMMs will continue to work as
if nothing happened. (If you've ever worked on a system with bad memory, with a
dead chip or data bus bit, with ECC enabled, you'll realize there is a cost.
The system continues to run, but it may crawl. I had a Unix machine that did
manage to continue running without a problem, until maintenance arrived to swap
out the RAM, but it did slow down noticeably. We kept it running, to see
if it would crash or not, and it stayed up and you could still work on it.)
Now, in your particular situation, it is a bit much to ask you to remove the
two, non-ecc RAM you currently own, and buy four sticks each equipped with
ECC and install them. All RAM has to have ECC, for you to be able to enable
it in the BIOS. If you were buying RAM from scratch, then suggesting ECC for
that much RAM would make sense. Again, it is a call on your part, as to whether
you feel the extra reliability of ECC protection, is worth the extra hassle or
not. I cannot make that call for you. If you are a banker doing accounts,
then you'd add it
If a gamer, then the occasional reboot, or crash out
to desktop, is not the end of the world. Some users just are not that
demanding of enhanced reliability, to go the extra mile. The machine I'm
currently working on, is ECC capable, but doesn't have ECC RAM. If it
falls over once every couple months, I really don't care
Paul
