Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

Hello.

I bought one of the A8V Deluxe/AMD64 3400+ deals from TigerDirect a few
weeks ago, and assembled a new system around it. I have an Enermax 300W
PSU, ATI Radeon 9800 Pro, WD 160GB SATA on the Via controller and two
Corsair CMX512-3200LL memory modules in the system. Whenever I run the
memory at 2:1 (DDR400), I have stability issues; if I downgrade to 5:3
(DDR333) it runs very stable. This memory is supposed to be *very* fast
(2-3-2-6 timings); any thoughts as to what might be going wrong?

I've searched the message threads on several boards and I've seen
mention of problems with incorrect SPD values, but the defaults seem
right (as reported by Everest Home Edition) and I haven't been able to
get the system to fail under memtst86. I'm not sure what to try next to
get it to run at 400MHz.

Thanks,

- Leo


--
Leomania

Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

In Nachricht news:Leomania.1gqfwa@news.computerbanter.com,
Leomania <Leomania.1gqfwa@news.computerbanter.com> schrieb:

> I'm
> not sure what to try next to get it to run at 400MHz.

Try to "enable" 2T Command-Rate in Bios' Memory Settings. This should
solve Your stability problem.

--
Viele Grüße,
Stephan Collet

Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

In article <31dfm0F3aejnmU1@individual.net>, Stephan Collet
<news.stephan.collet@arcor.de> wrote:

> In Nachricht news:Leomania.1gqfwa@news.computerbanter.com,
> Leomania <Leomania.1gqfwa@news.computerbanter.com> schrieb:
>
> > I'm
> > not sure what to try next to get it to run at 400MHz.
>
> Try to "enable" 2T Command-Rate in Bios' Memory Settings. This should
> solve Your stability problem.

Have you increased Vdimm ? High performance memory needs
more voltage to work properly. You should be using memtest86
from memtest.org, to determine whether the memory is sound.
Prime95 from mersenne.org (run from Windows) is also a good
test for the memory (use mixed size torture test). Do not use
2.9V or higher, as that is the limit of the processor (most BIOS
will not do that anyway).

Also, check the slots you are using. If you were to use
A1,A2 then that would be single channel, and two loads on
that channel. If a channel has two double sided DIMMs in it,
the processor's memory bus drive is probably not sufficient
to make the timing work, and that is when 2T command rate is
necessary.

Configs likely to need 2T command rate at DDR400 (double sided
DIMMs)

A1,A2 (Single channel, two DIMMs on one channel)
B1,B2 (Single channel, two DIMMs on one channel)
A1,A2,B1,B2 (Dual channel, two DIMMs on each channel)

Configs that should not need 2T command rate at DDR400 (double
sided DIMMs). These are dual channel configurations.

A1,B1 (Dual channel, one DIMM on each channel)
A2,B2 (Dual channel, one DIMM on each channel)

As for the memory speed itself, AMD is waffling on exactly
what valid configurations might be. Originally, AMD published
a table of memory configurations for the S754 socket. When
the first S754 boards came out, Asus copied the table exactly
as AMD had published it, while other manufacturers embellished
the table, and changed some of the values. Maybe the motherboard
manufacturers thought they knew better than AMD, as to what the
drive capabilities of an AMD processor were. In any case, when it
comes to the other sockets, like S939 or S940, information on
what is supposed to work or not work, is much harder to come by.
AMD punts on the issue, and the datasheet says valid configs are
controlled by the motherboard manufacturer (which is bullshit,
because AMD knows exactly what current drive it put in the
memory address buffers).

Now we come to the case of the A8V, and its S939 socket, and
unbuffered DIMM interface. The first thing that is strange,
is that the "DDR400 Qualified Vendor List" is perfect! Every
DIMM listed, is stable in 1 stick, 2 stick, and 4 stick
configurations. Other dual channel, dual slot boards, do not
enjoy this distinction, and you will normally find some brands
of RAM that cannot be run in a 4 stick config at DDR400. When I
first saw this table, I didn't think too much of it, until I
saw this post:

http://groups.google.com/groups?th [...] google.com

In that thread, the poster replies to himself, with the info
he got from Asus Taiwan. Basically, they say the motherboard
falls back to DDR333 rates, when using four DIMMs. That is
why the "DDR400 Qualified" list is able to so easily run the
four DIMM configuration. It is running at DDR333! What a great
fudge job on the Qualified list!

What this means is, running two DIMMs _per channel_ is a dicey
proposition. Your choices are, to run DDR333 Command Rate 1T,
or DDR400 Command Rate 2T. (Maybe someone can benchmark these
two options, and see which is better. They should have about
the same penalty for memory bandwidth.) Since the memory bus
drive is a function of pins on the processor itself, it is
possible that a different die stepping could improve the
situation, but as AMD feels there is no need to document how
their processors work, we'll never know. It is not the motherboard
makers responsibility to put multiple tables in the manual,
stating if you are using a CG, then this config will work, and
if you use some other die stepping, something else will work.

To see what AMD has to say, see page 15 of this doc:

http://www.amd.com/us-en/assets/co [...] /31411.pdf

"The use of 2T timing allows support of many DIMM combinations
at the maximum DDR speeds listed in Table 1. The 2T timing
feature causes commands and addresses to be driven for two clock
cycles and qualified with an associated chip select on the
second clock cycle, allowing an extra clock of setup to
accommodate heavy DIMM loading (such as double-rank DIMMs).
Refer to the BIOS and Kernel Developer¹s Guide for the AMD
Athlon 64 and AMD Opteron Processors, order# 26094, for the
DIMM combinations that require 2T timing to operate at the
full DRAM speed."

Well, the info is not in 26094. (The "en2t" bit controls command
rate setting is there, but no info on when to use it.) It is
the old runaround.

http://www.amd.com/us-en/assets/co [...] /26094.PDF

AMD should step up to the plate, and document exactly how
their product works. After all, they wanted to put the
memory controller inside the processor, so they should
provide the necessary information. (It is likely the people
in marketing prevent this info from being published.)

This doesn't affect you, running two DIMMs, as long as they are
in a dual channel configuration, like A1,B1 or A2,B2. In that
case, there is only one load per channel, and it better be able
to run DDR400 Command Rate 1T, or Asus is really telling a lie
in their Qualified Vendor List.

Hmm...

http://www.abxzone.com/forums/show [...] hlight=a8v
http://www.insanetek.com/index.php?page=overclocka64
http://xtremesystems.org/forums/sh [...] ht=divider

And this thread is particularly flavorful. It has info on
revisions, and some worked examples of the clock relationships.

http://www.ocforums.com/showthread [...] ost2748995

Based on the above four links, it would be a good idea to get
a copy of http://www.cpuid.com/cpuz.php and verify what the
hardware is doing. In particular, check the real memory
frequency, in case it is a lot higher than you think. A high
real memory clock frequency could account for your problems,
and make using 2T at some point, more likely.

HTH,
Paul