I am not clear here but does using slots 1 and 3 apply to single sided ram
too ?
Yes, the "1 and 3" rule spreads out the loading on the bus, so the
transmission line impedance doesn't dip as much as it would if the
sticks are close together. This should cause less of a reflection
from the mismatch that a DIMM causes. Single sided DIMMs cause less
of a reflection than a double sided DIMM, but spreading them out
still helps them.
I am using this board too and my original stick of ram was a very generic
stick of Chinese ram, DDR400 @ 512 Mb. My CPU is an Athlon XP Tbred-b 2400.
Running Win XP Pro and using the latest AIDA 32 Enterprise version I was
getting a ram benchmark putting my system only 1 notch above a Duron 900
system.
I returned the single stick and got 2x s/sided sticks of Kingston DDR400
making 512Mb and this lifted my AIDA benchmark to the same level as in the
next paragraph. I expected higher as this A7V8X-X system is a tad faster
than my A7V333 system.
The confusion really sets in when my AV7333 system using XP 2000 chip and 2x
s/sided sticks of DDR333 ram @512Mb gives me a benchmark in the right area
for my system but puts the A7V8X-X in the same spot.
I am sorry for rambling on but I can see where Devast8or is coming from.
It just seems to me that if you can get the ram right, these boards are
quite fine.
You are mixing in memory timing issues. The effects I am referring to
are analog electronics - they either make or break the interconnect.
If the analog is bad, then you'll have constant random memory errors
happening.
The memory chips themselves have timing parameters, as to how long each
stage of a memory transfer takes. The parameter are stored in a little
chip on the DIMM, called the SPD EEPROM. If you set a motherboard memory
to "AUTO", these suggested timings are read out of the SPD and used
to set up the memory. If you buy a CAS3 stick, then expect it to give
lower memory bandwidth numbers than a CAS2 stick, at the same memory
clock frequency (but the CAS3 will be cheaper). Both of the DIMMs will
be equally affected by the "analog electronics", but the silicon inside
the memory chips on one of them is able to give data faster than the
other one. So, they are two very separate issues.
You can set the memory to "Manual", and tune the memory timing numbers
yourself. Doing so harvests whatever margin for uncertainty that the
manufacturer included when they made the DIMM. Some DIMMs have been
shown to have tremendous potential for higher performance (I read of
a DDR266 stick that went all the way to DDR500!). But the DIMM market
is very fluid, and you cannot expect the same chips or PCBs to be used
in memory products for more than a few months (many companies don't make
their own product - there is even a company that just makes standard
blank DIMM PCBs for other companies to put chips on).
Another behind-the-scenes effect, is the BIOS on some motherboards
automatically slows down the memory clock when you add more DIMMs.
You have to check the BIOS screens, or use a utility, to find out
what is really happening to the memory clock rate and memory timing
parameters (i.e. 2-3-3-7 1T)
I have even found the part numbers for DIMMs stored in the BIOS code.
This means that Asus checks for certain brands of memory, and ignores
some of the timing info in the SPD when they see those particular
sticks of memory.
Another thing that affects memory benchmarks, is when the processor
clock and the memory clock are in a so-called "1:1" or synchronous
relationship. Some digital retiming logic can be removed from the
memory path when this special timing relationship exists. This is
called "PAT" by Intel, on the 875, but the effect exists in many
Northbridge chips. So, sometimes running a FSB at 333 and memory at 333
can give better performance than FSB at 333 and memory at 400. You
might have to boost the memory to 500 before the faster memory clock
compensates for the extra delay that was added to the memory timing
path.
So, there is a lot of monkey business that goes into making memory
work. And good pictures or tutorial articles are hard to find.
HTH,
Paul
P.S. If you want to see what memory bus signals look like, there is a
5.9MB download on the Micron web site, and the back few pages of
the document shows various test cases that Micron was using. The
square box in each picture measures what is called the "eye opening",
and the bigger the box is, the more error-free the memory will be.
Very subtle changes in the construction of memory chips or printed
circuit boards can affect how "wiggly" these signals are - so sometimes
whatever "rules" we come up with, just don't seem to work out. The
more DIMMs you use, the more ringing you see in the signals, and
more the "eye opening" shrinks, leading to the possibility that
you'll see random memory errors. Slowing down the memory clock
when using many DIMMs, gives the signals more time to settle down
between data bits. This is the "analog side" of memory design.
http://download.micron.com/pdf/technotes/TN4607.pdf (see back pages)