Jack wrote:
> "Paul" <> wrote in message news:f6jjgk$144$...
>> Jack wrote: <SNIP>
>
>> Have you looked at the "percentage" setting for memory ? My Nforce2 board
>> offers "100%", "83%" and so on, and if my FSB is FSB400, that would give
>> DDR400
>> and DDR333 respectively. By using a lower percentage setting, you can
>> drop the DRAM speed a bit, and repeat your tests.
>>
>> Turbo, on some motherboards, sets CAS to CAS2. Which is not good, if the
>> RAM being used is actually a CAS3 product. Better to either allow
>> the BIOS to pick the (slowest) timing via an "Auto" setting for the
>> timings,
>> or examine the three sticks, and work out the timing numbers yourself.
>>
>> In terms of the relative speeds of the FSB and the memory, remember that
>> the processor FSB is 64 bits wide, while the two memory channels together
>> are 128 bits wide. The system is "balanced", if the memory runs at half
>> the
>> rate of the FSB. For example, if you had the processor at FSB400, then two
>> DIMMs at DDR200 in dual channel, would have the same theoretical
>> bandwidth.
>> So you can turn the memory down a bit, and try to get stability, and lose
>> a few percentage points in performance. It still helps to have the
>> memory subsystem a little faster than the theoretical balance point, since
>> the memory bus is not 100% efficient (has timing holes). But if you had to
>> run the FSB at FSB333 and the memory at DDR266 dual channel, that might
>> still be a viable solution.
>>
>> Using a "non-CPC BIOS", otherwise known as "Command Rate 2T", is another
>> way to improve stability (at the expense again, of memory bandwidth). But
>> modded BIOS are not available for all boards.
>>
>> You already know about the "single channel workaround", but that isn't
>> very useful when using three sticks. For three sticks, either drop the
>> memory clock, or find a "non-CPC" or non Command Per Clock version of the
>> BIOS.
>>
>> Paul
>
> Paul here's a response I just got back from Gigabyte:
>
> "hello,
> memory need to be same identical , to run 2G memory on dual channel mode :
> put in 2 stick 1G identical memory on slot 1 and 3 should work ."
>
If it is a dual channel capable Nforce2 chipset, all that you need to do
is balance the total memory on the two channels. 2x512MB on one channel,
and the 1GB stick on the other channel, will give dual channel. The DIMM3
slot is the one that gets the 1GB module, while DIMM4 remains empty (and
DIMM4 would not exist on other brands of motherboards).
I've even done an experiment, where I got the source for memtest86+ and made
a trivial 3 line change to the program. I set up memtest86+ to measure the
memory bandwidth at eight different spots in the memory. I installed
3x512MB on purpose, as it is "unbalanced". What I discovered is, the
memory operated at dual channel speeds for 2/3rds of the memory, and at
single channel speed for the remaining 1/3rd of memory. So in fact, the
motherboard will operate in dual channel mode, for as much memory as
is "matching" on the two channels. Any left over memory (where one channel
has more memory than the other), is operated in single channel mode.
OK, while I was writing that, I download the manual from here.
http://america.giga-byte.com/FileLis..._7n400p2_e.pdf
PDF page 51, shows the "percentage" setting in the BIOS for memory. Try
a value like 83% and run your tests again.
I don't know anything about EasyTune. I did all my testing on my Asus
Nforce2 board, using the BIOS. Unless there is something wrong with the
hardware, dropping the clock on the memory (using the percentage setting),
should give you stability. The JEDEC spec for PC3200 suggests 2.6V as the
voltage to use, and I don't recommend more than 2.7V unless you know for
sure the memory is rated for more than that. Some memory cannot take a
lot of voltage, while other kinds can. The champ for voltage, is Winbond
BH-5, which people routinely ran at 3.3V. My Ballistix memory would be
toasted in seconds, at that voltage.
Paul