Just Wondering wrote:
> Hi,
> I own an ECS mainboard model 755-A2. It has two DIMM slots (standard
> bank 0, 1 and the other is bank 2, 3). Try as I might, even with
> matching memory sticks that came in the same package, I cannot get the
> board to work with two DIMMs installed. Has anyone else had this
> issue? I flashed the BIOS to the latest version they ever put out
> (Sept 05). It was supposed to fix the SATA controller to be able
> handle hard disks greater than 137 GB and it did---I now have a have a
> 500 GB SATAII 300 drive--although I did have to hard code it via
> jumper to only pass data at the SATA I spec value of 150 MB per
> second. The BIOS update was also supposed to bring the memory
> capacity up to 4 Gig total from 2 Gig total---so I thought I could
> just return my two 1 Gig sticks and get a single 2 Gig stick to get
> the 2 Gig DDR RAM capacity I wanted on this (now) old computer. But
> Frys doesn't sell any 2 Gig sticks (except as sets of two matched 1
> Gig sticks and I tried that). So---has anyone else had any luck
> getting two sticks to work in their ECS Model 755-A2 boards?
>
> (BTW, the 755 board has the 754 pin processor connector for 754 pin
> AMD chips.--but I guess you guys would already know that).
>
> Please let me know.
> Thanks,
> Jerry
There is a table in the manual, a table for a three slot board.
Your board has two slots, but at least some of the principles
should be the same.
The Athlon64 has the memory controller inside the processor. It
has loading limitations - the more loading on one of its address
busses, the slower it goes. The S754 processor has two address
busses, and one data bus. On a three slot board, one address bus
goes to one slot, the second address bus goes to the other two
slots. The address value on each bus, is the logical complement.
(One bus has a 0, the other bus a 1, and the switching noise
cancels out.) One address bus is sent the correct value, the other
bus gets the complemented, wrong value. Only the "correct" bus gets
the chip select asserted.
A lot of nonsense, to make it work. The reason for a lot of this,
is the drivers on the bus cannot be too aggressive, as the switching
noise could potentially upset the operation of the processor core.
With two sticks, you can help it by dropping either the memory clock,
or setting the command rate (on other boards, the available options
would be 1T or 2T) for command rate. With two double sided sticks
installed, you'd expect it would not run at DDR400 Command Rate 1T,
but somewhat lower.
You could try dropping the "Max Memclock" in your BIOS quite a bit,
and see if you can get the two stick memory configuration stable
that way.
So the recipe would be:
1) Install just one stick.
2) Go to the BIOS. Set "DDR Timing Setting by" to [Manual].
Then the "Max Memclock" setting should light up. Set it to DDR200,
instead of the more normal DDR400, or DDR333 etc. The purpose of going
this low, is to prove that the two sticks will work under some conditions.
3) Save and exit from the BIOS.
4) When the board starts to POST again (you see some text), you can shut it off.
5) Power down and *unplug* the computer. Install the second stick of
RAM, with no power present in the computer. (If +5VSB is running,
the RAM can be damaged.)
6) Plug in and power up. First off, check that the board will POST with
the new memory setting. You could go into the BIOS again and verify
the settings are still applied (in case the BIOS is whacky, and changes
things on its own). You should be able to exit from the BIOS, without
saving changes.
7) Get a copy of memtest86+ from memtest.org and test the new two stick
configuration. (Can be run via floppy or CD.) Test for two full passes.
This may take an hour or two. While this test is not very thorough,
in terms of identifying problems, it is better than nothing.

At this point, I'd boot a Knoppix or Ubuntu Linux CD (no hard drive needed),
and then test with Prime95 from mersenne.org . That is a better test that
the memory is working well. While tuning a machine, I might run Prime95
error free for 10 minutes. If I think I've got my final, conservative
settings, I might test for 4 hours.
9) Fire up Windows again. Use a copy of CPUZ from cpuid.com , to verify what
your motherboard is using for the RAM timings (like 3-3-3-

, the
command rate (1T or 2T), and the actual clock being used for the RAM
(DDR200 in this example).
In terms of the aggressiveness of the settings, from most aggressive to least
aggressive, it goes:
DDR400 1T (typically works with one stick on an AMD channel)
DDR400 2T (may work for two sticks on a channel - slightly better performance)
DDR333 1T (may work for two sticks on a channel)
DDR333 2T (should definitely work with two sticks)
...
DDR200 whatever (if this doesn't work, RMA something

)
The timings are also important, and at DDR400, the industry standard
(i.e. if the advert doesn't say anything) would be 3-3-3-8. The first
number is CAS and affects performance the most. You don't want to
dial down those numbers too much (too tight), as that can also cause
errors.
To be able to tune the board for maximum performance, you really want
a BIOS that exposes the four timings, the command rate, and the
memory bus frequency. If some of those settings are missing, it
means doing the best that you can, with less than ideal values.
Another useful setting, is adjusting Vdimm. DDR400 memory officially uses
2.6 volts. DDR333 or lower memory uses 2.5V. In practice, if the motherboard
has an adjustment, you can probably apply 2.7V without a problem. (That
is what I'm using on my current board.) The allowed spec is 2.6 +/- 0.1V,
so 2.7V is right on the edge of that limit. And in practice, there is more
room than that to raise the voltage. (In addition, some motherboards don't
tell the truth - they may show 2.5, 2.6, 2.7, 2.8 in the BIOS, but the
hardware only has two voltage values as a result of those settings. So
the hardware capability doesn't always match the BIOS description.)
On your board, I see a "DIMM voltage adjust", and the manual is a bit
confused about what options it has. Either it will show a numeric
value, or [Enabled]. In any case, try 2.6V or 2.7V, and see if the
error rate in memtest86+ drops to zero as a result. You only want
to apply as much voltage, as is needed to maintain a zero error rate.
There is no point in "cooking" the RAM for nothing, unless it really
needs it.
HTH,
Paul