One thing you can try, is use the Auto settings, then use a Windows
utility to examine the hardware settings being used. That will
give you a starting point. CPUZ from cpuid.com is an example,
and you might also find Everest from lavalys (formerly called aida32)
or Sandra from sisoftware helpful as well, when it comes to
listing system properties.
You can also look up timing information from the product datasheet.
The datasheet links are on this page. A typical number might be
2-3-2-6-1.
http://www.kingston.com/hyperx/thelines/default.asp?type=khxu
This page has a slide show about memory. Slide 24 shows the order
these numbers are listed in (CAS-tRCD-tRP-tRAS-commandrate):
http://corsairmicro.com/corsair/products/tech/memory_basics/
On Athlon64, command rate, DDR clock rate, and number of sticks
go hand in hand. The command rate of the memory is always
basically 1T, so in a way, including that parameter is a
misnomer. If you have four sticks of RAM in a dual channel
system, it is likely the BIOS will set the command rate
automatically to 2T, which reduces memory bandwidth by perhaps
20% or so on benchmarks. (With only two sticks, the BIOS might
use the 1T setting, as there is less electrical load.) The 2T
setting means two clock cycles are used to pass a command to
the memory, while 1T means one clock cycle is given instead.
The memory basically always takes a command per cycle, and
when the memory controller uses 2T timing, is basically wastes
a cycle to improve memory timing and allow the address/command
more time to settle.
With any of these settings, start with the values returned in
CPUZ, with the board all set to Auto. Get a copy of memtest86
from memtest.org and use that to test the memory. The memory
must be error free, no matter what final settings you decide
on, because otherwise the system will crash at some point in
time. Tighten up a timing parameter, then re-test, to make sure
you haven't pushed the memory too far.
The datasheet for the memory will also have a suggested voltage
setting. 2.5V is the memory chip voltage rating for memory
slower than DDR400, and 2.6V is the number used for DDR400 or
higher. Those are minimum acceptable values. The maximum
allowed voltage is determined by which of two devices has the
lower breakdown voltage - the memory controller or the memory
chips on the DIMM. Some memory chips are intolerant of voltage
increase, while others can take a lot more. For example, the
Winbond memory BH-5 can be pushed all the way to 3.3V, and not
too many other chips can take that kind of voltage. In your case,
your memory controller is in the processor, and section 7.1 of
the processor datasheet, says 2.9V is the max the memory interface
pins on the processor can handle.
(see section 7.1 "Absolute Maximum Ratings" for S939 processors)
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/31411.pdf
Generally, the Asus voltage range listed in the BIOS, should not
go high enough to damage a motherboard component, but only you
the user can decide whether or not to use more voltage than the
value listed in your memory DIMM datasheet. For example, the
Kingston datasheet mentions 2.6+/-0.2V as the voltage spec,
meaning you can set the BIOS to no more than 2.8V for the memory.
Thus 2.75V would be as far as I would push it, as there can be
some setting error in the motherboard circuits. (Generally, I'm
very impressed with the precision of the Asus circuits I've
measured with a multimeter, so it won't be too far off.) A
setting of 2.75 is less than the stated 2.8V limit of the
memory and 2.9V limit of the memory controller on the processor.
While in the private forums, you can find overclockers applying
more voltage than the stated maximums, the results of such
experiments can be unpredictable (burnout the memory overnight or
kill the processor).
You can purchase third party devices, like the OCZ DIMM booster,
and these devices can override the voltage limits of the motherboard
circuits. Using such a device generally means a "mad overclocker"
doesn't give a damn if they burn out a component while testing.
HTH,
Paul
