"J. Oram" said:
Board: ASUSTeK Computer INC. A7N8X-E REV 2.xx
BIOS: Phoenix Technologies, LTD ASUS A7N8X-E Deluxe ACPI BIOS Rev 1008
12/05/2003
BIOS settings in memory are "Optimal"
Bus Clock: 200 megahertz
Mobile XP vcore 1.60 Multiplier 11.5
RADEON 9800 PRO Driver 6.14.10.6497
Using Windows 2000 Version 5.0.2195 with latest patches.
I have been running stable with Mushkin DDRRAM 756MB (3 each 256MB
PC3500 - Silver DIMM) = 3DMark05 = 2507 specmark rating.
However just tried installing Mushkin DDRRAM 2GB (2 each 1GBB PC3200 -
Black DIMM) in Dual Channel configuration - slots A1 & B1
When re-running 3DMark05 the system stops dead after approxametly 2
minutes of testing.
There is a new BIOS file at the ASUS download website of
C18E1013.zip Version 1013 Date 2004/11/19
Notes say:
Patch new Samsung DDR400 DRAM to fix enter WinXP problem
Fixed Speech IC detect abnormal for CPU overclocking
Support new CPUs. Please refer to our website at:
http://www.asus.com.tw/support/cpusupport/cpusupport.aspx
When I tried using Asusupdt60201.zip to automate flash of BIOS, but
the latest version of BIOS the application allows is
Version 1010 (03/31/04)
What BIOS should I be using to work with 2GB dual-channel memory?
Thanks in advance,
John O
What memory works with Nforce2 is unpredictable. Reading posts on
nforcershq.com doesn't suggest any common theme. I had luck using
a low latency RAM (CAS2), but that doesn't mean that CAS2 is an
essential part of a RAM choice.
Mushkin at one time, sold "stacked" RAM, where two silicon die are
used to build a larger capacity RAM chip. To Mushkin's credit,
they prominently displayed this info in their ad copy, and have
compatibility info on some of their web pages. I would not recommend
"stacked" RAM, as it would appear that the Nforce2 chipset doesn't
have an overabundance of drive capability. (Stacked RAM presents
twice the capacitive load on the address bus.)
Checking now, I see Mushkin has some products that use 64Mx8
(512 megabit) chips, and those can be used to build an unstacked
normal module. Those should stand a better chance of working.
Only one product listed has no info whatsoever, as to how it is
constructed.
Basically, as near as I can guess, the problem is with the
Nforce2 chipset. It seems to lack a robust design when driving
dual channel at 200MHz, and users need to experiment to find
a RAM to work with it. I've read a couple of posts from people
who tried PC4000 memory and that didn't work, so it is not a
simple clock rate issue. I personally had better luck with
CAS2 memory, but I'm not sure if anybody makes a 1GB (non-stacked)
unbuffered DIMM that runs at a CAS of less than 3.
Any workaround will reduce memory bandwidth. The things you can
try are:
1) Reduce FSB. Dropping from 200MHz to 180MHz might improve
operation for you. You could also try setting the memory
clock to a reduced setting, and leaving the FSB alone.
Generally, operating FSB and memory in sync is the best
setting, from an error rate perspective.
2) Put the two DIMMS on the same channel. Leaving Slot3 blank
means there is no electrical noise from the second channel,
on the Nforce2 Northbridge chip. It cuts the switching
noise in half. That seems to help.
3) There is a Northbridge option for Command Rate. On nforcershq
this is referred to as CPC or "command per clock". A Command
Rate of 2, means the address is presented for two cycles, and
the info is strobed into the chip on the second cycle. This
wastes memory bandwidth due to the wasteful first cycle of
address. A command rate of 2 is the same as saying "command
per clock disabled". A command rate of 1 (address and strobe
presented for only one cycle) is "command per clock enabled".
CPC is not an option in an Asus BIOS. I use a hacked BIOS by
"Trat", from Nforcershq, and there are two different BIOS, one
with CPC on and the other with CPC off. I am currently using
CPC on, and the Asus BIOS always default to CPC on, as it is
the highest performance option. The "Trat" BIOS allows you to
select the lower performing CPC off option, by flashing that
version of BIOS. You have to alternate between BIOS files, to
change the CPC setting (implicit with hacked BIOS version, not
a BIOS setup screen option).
"CPC off" loses bandwidth equivalent to a 30MHz drop in
memory frequency (ie. drop from 200MHz to 170MHz).
All of the above workarounds erode memory bandwidth, so no matter
which of them you use, you lose precious memory bandwidth. If
you can return the Mushkin DIMMs, I'd try to find some CAS2
512MB DIMMs instead, and live with 1GB. I've had three sticks of
CAS2 512MB memory on my A7N8X-E and they ran rock solid at 200MHz
clock. (I've done an experiment while running that unbalanced
channel config, and part of the memory space has reduced memory
bandwidth. So, again, if you run 3x512, you will probably end up
with the same level of performance hit as running 2x1024 at some
workaround setting. What happens is, when one channel has 1024MB
on it, and the other channel has 512MB, the channels cannot
alternate for the last 512MB, so 512MB of memory is slower than
the rest of the memory. As a test program strides through memory,
the measured bandwidth changes as a function of memory address
tested.)
I did a quick check, and I cannot find any 1GB unbuffered sticks
that offer CAS lower than 3 at DDR400. Samsung doesn't have any
chips listed for that, and Corsair's DIMM is CAS3 as well. You
can try buying another 1GB module based on 64Mx8 chips from
Crucial, Kingston, Corsairmicro.com, or a generic Samsung, but
as long as all of them offer only CAS3 at DDR400 rates, there is
no guarantee they will work any better.
I test memory with memtest86 (from memtest.org) and Prime95
(from
http://mersenne.org/freesoft.htm ) . Memtest86 runs as a
standalone test floppy, while Prime95 runs from Windows.
Memtest86 is good for testing that the memory does not have
a very obvious stuck-at fault, while Prime95 is good for testing
"speed faults", where a portion of memory will not run fast
enough internally to keep Nforce2 happy. I run multiple copies
of Prime95 from a Knoppix boot CD, and in that case I use a linux
version of Prime95. The Prime95 "torture test" mixed option
should run error free in an overnight test, before you consider
running 3DMark again. Once memtest86 and Prime95 have proved the
CPU and memory work, continued problems in 3DMark might means
the video card or Northbridge AGP slot have a problem.
The last thing I would expect, is a BIOS change to help. Even
though I have downloaded and used a hacked BIOS during my testing,
it did not seem to be an element of finding a solution. If you
are running a reasonably recent BIOS, I'd concentrate on finding
just the right DIMM.
Good luck (you'll need it).
Since this phenomenon is demonstrated on other brands of
motherboards, this is not just an Asus problem.
If you do find a memory that works better, post your results,
as user reports are the only way to find working solutions.
I used Ballistix 512MB CAS2 PC3200 memory for my 2x512 or 3x512
successfully working configurations. Running at 2-2-2-6
DDR400 (CPC on) in both cases. Processor was 200x11.
A happy but annoyed A7N8X-E user,
Paul
