Re: A7N8X-E deluxe : new ram problem

In article <u%mqg.20892$ap3.12120@xxxxxxxxxxxxxxxxxxxxxxxxxx>, "Venom"
<Venom.au@xxxxxxxxxxxxx> wrote:

I've been running it with a pair of Kingston KVR DDR400 512Kit (2x256)
rock stable for 2 years in Dual channel 200mhz mode and since I've
added a stick of 1G Kingston KVR (KVR400X64C3A), Windows starded to
make BSOD after booting up:


ASUS seems to make their own rules on each different motherboard. My A7V600
has also been rock steady on 2x256 of single sided ddr400 for over a couple
of years. I felt the need to raise the ram and I have had nothing but
trouble since. The manual says when using ddr400 you can only use two sticks
yet I assume when using ddr333 you can use all the slots. How else would you
be able to run the 3Gb that they say the board is capable of using?
If I use one double sided 1GB stick in any of the slots the computer works
but is unstable. Any mixture of double sided ram and Windows will not load.
Because it works so well with just the two identical sticks of single sided
ram I have run the budget dry buying assorted sticks of ram and the final
test will come later this week when the (last attempt) twin pack of single
sided ddr400 2x512MB arrive and I try them. The best the guys down at my
favourite computer spare parts store can tell me that Intel boards do not
have these problems. I have no experience with Intel stuff. This may or may
not help you but at least you know now that you are not the only one being
driven around the bend by bloody ram.

The Nforce2 chipset has three memory slots, but is a dual channel
design. Two slots fit on one data bus. One slot on the other data
bus. Each slot has its own private address bus, which should make
a big difference to the quality of operation.

The address bus tends to be the sticking point, because when a
module has 16 chips on it, and all the address inputs receive a
copy of the same info, there are 16 loads on the bus. If we put
three sticks on a single channel motherboard, and there was only
one address bus for the whole thing, there are 48 loads on the bus.
This severely limits the clock rate that can be used - a lot of
time is needed for the bus to settle, from one transaction to the
next, and using three double sided sticks is bad, no matter whose
chipset is used.

On AMD socket 754 processors, they use two address busses and one
data bus. And you'll notice, that the bus performance tends to be
turned down by the BIOS, when trying to use a significant part of
the three available DIMM slots. In that case, the processor does
not have a lot of electrical energy to waste on driving RAM, as
the chip has a lot of other noisy stuff going on. That is one
advantage of using a separate Northbridge memory controller - it
does memory and video card slot and the memory controller can get
a better quality design, as there is less noise to deal with.

The Nforce2 should, technically speaking, be virtually bulletproof.
It certainly looked that way at introduction, with Anandtech memory
stress testing showing three DDR400 sticks running at CAS2 at full
speed. In my experience, the Northbridge seems to be performance
limited (almost like a noise problem), as it did not like the crap
CAS3 memory I fed it. Using three sticks of more expensive CAS2
fixed that. (I can, in fact, run 3x512MB double sided Ballistix
CAS2 memory, with not a hint of a problem.) I could also have got
relief from a hacked Command Rate 2T BIOS, where the address bus
is run at half the normal rate, but similar to Athlon64, that
would reduce the already pretty low memory bandwidth available
to the processor.

On the A7V600, I don't know how many address busses would be present.
It seems like only one, based on seeing no difference in behavior,
as memory is moved from one slot to another. Like all single channel
three slot designs, the clock more or less has to drop, to get
stable operation when all slots are in use.

With 1GB sticks, you have to be very careful what you buy. The
cheapest 1GB sticks, have no major brand name sticker on the RAM.
That is because a major name supplier, would not be caught making
non JEDEC compliant memory configurations. A favorite trick on
the cheapest 1GB RAM, is the use of 128Mx4 chips. To make a full
bank of such RAM, takes 16 chips to make the bank, meaning some
control signals have 16 loads, whereas in a JEDEC compliant
configuration, there would only be 8 loads on those signals.
Since this config is not approved, I have _never_ been able
to get documentation for one of these DIMMs, to prove that is
how they are constructed.

A 1GB module should be designed with 64Mx8 chips. Two eight chip
banks are the result. Since the address depth is only 64M per
chip, there will be more chipsets that can properly address the
whole stick. (For example, I believe Nforce2 cannot use 128Mx4
chips, and could only address half the chip.) The typical price
difference between crap and usable (16x64Mx8) type memory, is
maybe $15 per stick.

If using those 1GB modules, if you wanted to use three of them
on an Nforce2, I'd want to buy CAS2 memory, to compensate for the
marginal Northbridge chip design. On an A7V600, I doubt that
would help, and due to the total single channel load, the clock
would have to be turned down, to get joy. It might have to drop
to DDR266, for three double sided sticks to run happy.

What buying 512MB single sided DIMMs does for you, is guarantee
you are getting 64Mx8 chips. You cannot use 128Mx4 chips, because
there would not be enough chips to make a 64 bit wide bank. So,
demanding single sided sticks does solve that problem, of getting
128Mx4's. But using two single sided 512MB would not be appreciably
different than using one quality double sided 1GB with the right
16x64Mx8 config.

In the case of Intel chipsets and RAM, there is a suspicion that
they use Command Rate 2T all the time. But Command Rate is never
mentioned in Northbridge design docs, and neither is it a
parameter observable in a chipset register. I only have the guess
of someone at Xbitlabs, as a clue to how Intel chipsets differ.
I think you would find, for some of the older Intel chipsets,
like maybe an 845, that they are just as loading sensitive as
other chipsets.

Using DDR memory is never easy, with the main weakness being
a multidrop transmission line with fixed termination at the
end. DDR2 has demonstrated better characteristics, although
I don't think you will find any DDR2 designs with a single
channel three slot design. At most, a DDR2 channel uses two
slots, at least the ones I've looked at. And memory stress
testing on DDR2 systems, with a different termination method,
has shown little difference between having one or two sticks
on the bus. There is still the possibility of design quality
differences between VIA, SIS, ULI, Nvidia, ATI, but it should
have less to do with the basic characteristics of the memory
bus, and more to do with how good the Northbridge chip
designers were.

For Asus motherboards, not all the manuals have contained an
appropriate warning, when if comes to using three double sided
sticks. If your A7V600 manual does not contain a warning, then
read the A7V8X manual, or the A7V333 manual, etc, and see how
other chipsets in the same family are limited to lower clock
rates with more sticks. VIA chipsets tend to reuse the same
sinful designs over and over again.

Paul
.