A8N-SLI Deluxe RAM question

G

Guest

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

I have 2x1GB DIMMs of Crucial Ballistix memory installed on my A8N-SLI
Deluxe, and I'm looking for a bit more to ease some Photoshop work I'm
doing. I'd like to add 2 more 1GB DIMMS, but my understanding is that I'll
only be able to realize ~2.75GB total due to some BIOS issue. In fact, the
Crucial web site even makes the disclaimer" "Important information about
your system. Although the memory can be installed one module at a time, the
best performance comes from using matched pairs of modules. Although the
nForce 4 chipset supports 4GB of memory, Windows and the BIOS will usually
see 2.75-3GB when 4GB is installed. This is due to limitations in memory
addressing by the BIOS."

Is this destined to be a perpetual issue, or can add 2 more 1GB DIMMs and
hope that the issue is resolved in future BIOS releases? I'm currently
running 1006.
 

Paul

Splendid
Mar 30, 2004
5,267
0
25,780
Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

In article <pJRhe.15305$iU.12499@lakeread05>, "J&SB" <jandsb@cox.net> wrote:

> I have 2x1GB DIMMs of Crucial Ballistix memory installed on my A8N-SLI
> Deluxe, and I'm looking for a bit more to ease some Photoshop work I'm
> doing. I'd like to add 2 more 1GB DIMMS, but my understanding is that I'll
> only be able to realize ~2.75GB total due to some BIOS issue. In fact, the
> Crucial web site even makes the disclaimer" "Important information about
> your system. Although the memory can be installed one module at a time, the
> best performance comes from using matched pairs of modules. Although the
> nForce 4 chipset supports 4GB of memory, Windows and the BIOS will usually
> see 2.75-3GB when 4GB is installed. This is due to limitations in memory
> addressing by the BIOS."
>
> Is this destined to be a perpetual issue, or can add 2 more 1GB DIMMs and
> hope that the issue is resolved in future BIOS releases? I'm currently
> running 1006.

These are the release notes for the 1006 BIOS:

[ 1006 ]
1. Fixed after resuming from S3, the CPU multiplier and CPU clock
will be restored to default value if user overclock.
2. Fixed show wrong device (ZIP) on summary screen when PATA
controller was disabled. It should show 'None'.
3. Change the CHIP fan fail threshold RPM to fix sometimes POST
will wrong show CHIP Fan fail Msg.
4. Update A8N-SLI Deluxe Logo.
5. Add Memory Re-map function SETUP Item.

The last item, item #5, notes that a new setting has been added
to the BIOS interface. The BIOS was missing the necessary setting
to handle configurations with large memory installed.

While some of the info in the following thread is mixed up, at
least the poster in this particular post, notes that Windows shows
4GB available when "Memory Hole for PCI MMIO" is enabled. Note that
"joeWI" is using "Windows Server 2003 Enterprise Edition with
Service Pack 1" as his OS, and the experience with WinXP may not
be so straightforward.

http://www.bleedinedge.com/forum/showpost.php?p=78759&postcount=11
(thread) http://www.bleedinedge.com/forum/showthread.php?p=78759

It is unclear to me, whether with, say, WinXP, you would need to
add /PAE or /3GB to the boot.ini file or not. When "memory hoisting"
is involved, to lift the memory in the memory mapped I/O hole, above
the 4GB mark, then more than 32bit addressing is needed. One way of
getting this is with PAE. Some of this is explained in this posting.

http://groups.google.ca/group/alt.msdos.batch.nt/msg/cac585f0671e2a2f?dmode=source&hl=en

If you do decide to go with 4x1GB, please post back your
experiences. I'm curious if Memtest86+ can handle this situation
or not. I'm also curious if there is any performance penalty,
over and above having to run command rate 2T at DDR400. To
figure that out, run a memory benchmark with two sticks, with
command rate 2T and memory set at DDR400. Then, insert the two
additional sticks and retest. Finally, enable the MMIO BIOS
setting and test again. Memtest86+ has a read bandwidth indicator
in its interface, and you can also use programs like Sisoft Sandra
and probably a few others, to benchmark the effects of these
options.

Don't expect four sticks of memory to run command rate 1T at DDR400.
I would not expect, even with Venice and San Diego versions of
processors, that this would be possible. Command rate 2T is used
to compensate for the memory loading of four sticks (32 loads on
the address/control signals).

Paul
 
G

Guest

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

Paul wrote:
> In article <pJRhe.15305$iU.12499@lakeread05>, "J&SB" <jandsb@cox.net> wrote:
> It is unclear to me, whether with, say, WinXP, you would need to
> add /PAE or /3GB to the boot.ini file or not. When "memory hoisting"
> is involved, to lift the memory in the memory mapped I/O hole, above
> the 4GB mark, then more than 32bit addressing is needed. One way of
> getting this is with PAE. Some of this is explained in this posting.
No matter what you do, Windows XP Home/Professional simply will not use
ram which is mapped above 2^32, you can enable /PAE /3GB or whatever and
it simply won't work. This is an artificial limitation of the PAE
implementation in Windows XP. Windows Server 2003 Standard will support
the full 4GB, for even more ram (with 4 unbuffered 2GB sticks composed
of 16 128Mx8 chips this should be possible on that board, though these
modules aren't quite mainstream yet...) you need even more expensive
windows versions (enterprise, datacenter, up to 32/64GB)

Alternatively, the Windows XP Professional x64 version supports the full
4GB too (in fact, up to 128GB) and should be a bit easier on your wallet...
Alternative OS (no, DOS doesn't count...) usually support more than 4GB
ram just fine too. That said, Asus is known to be sloppy with
implementing features like memory remapping and sometimes doing it a bit
wrong, so it might not work with all OS. Windows should be fine though
(that's Asus standard answer with broken bios features "but it works
with windows, so it obviously must be correct!" - even though Windows
might complain too but just happens to ignore the error)

Roland
 

Paul

Splendid
Mar 30, 2004
5,267
0
25,780
Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

In article <42891562$0$1163$5402220f@news.sunrise.ch>, Roland Scheidegger
<rscheidegger@gmx.ch> wrote:

> Paul wrote:
> > In article <pJRhe.15305$iU.12499@lakeread05>, "J&SB" <jandsb@cox.net> wrote:
> > It is unclear to me, whether with, say, WinXP, you would need to
> > add /PAE or /3GB to the boot.ini file or not. When "memory hoisting"
> > is involved, to lift the memory in the memory mapped I/O hole, above
> > the 4GB mark, then more than 32bit addressing is needed. One way of
> > getting this is with PAE. Some of this is explained in this posting.
> No matter what you do, Windows XP Home/Professional simply will not use
> ram which is mapped above 2^32, you can enable /PAE /3GB or whatever and
> it simply won't work. This is an artificial limitation of the PAE
> implementation in Windows XP. Windows Server 2003 Standard will support
> the full 4GB, for even more ram (with 4 unbuffered 2GB sticks composed
> of 16 128Mx8 chips this should be possible on that board, though these
> modules aren't quite mainstream yet...) you need even more expensive
> windows versions (enterprise, datacenter, up to 32/64GB)
>
> Alternatively, the Windows XP Professional x64 version supports the full
> 4GB too (in fact, up to 128GB) and should be a bit easier on your wallet...
> Alternative OS (no, DOS doesn't count...) usually support more than 4GB
> ram just fine too. That said, Asus is known to be sloppy with
> implementing features like memory remapping and sometimes doing it a bit
> wrong, so it might not work with all OS. Windows should be fine though
> (that's Asus standard answer with broken bios features "but it works
> with windows, so it obviously must be correct!" - even though Windows
> might complain too but just happens to ignore the error)
>
> Roland

I was looking for a Microsoft web page that would address the
issue of needing 5GB of address space, but only having 4GB of
physical memory. The KB articles that attempt to address memory
limits, aren't too precise about exactly what is supported.

This article mentions that WinXP will access above 2^32 when using
the PAE switch. Now, AMD64 has the MMIO feature, and I don't know
if that means the whole computer will work properly or not, with
this 5GB address space, 4GB physical configuration. A question
that has come up in my mind about this particular configuration,
is the layout of the kernel memory versus user space - if the kernel
lived in the upper GB of memory, perhaps PCI I/O operations would
never go near 2^32, further improving the odds that it might work.

http://blogs.msdn.com/carmencr/archive/2004/08/06/210093.aspx

Another disturbing thing in that thread, was the posting at the
bottom of the page. A user upgraded to SP2 and found the system
reported only 3.18GB when before SP2 it reports 4GB.

On the one hand, I could suggest to the OP that the best thing
would just be to try it and test it, but I'm not even confident
I could suggest a good test strategy that would exercise the
"corner cases" well enough to bless the resulting system. Maybe
simply installing the 4GB of memory, and enjoying the use of
3.1GB of it (the non PAE option), is as good as it'll get, without
a lot of unnecessary effort and expense.

One of the reasons I suggested going with the 4GB config, is based
on a few experiences people have had mixing pairs of DIMMs on
these boards. There was one person, who mixed a pair of CAS2
256MB DIMMs with a pair of CAS3 DIMMs, and one pair of DIMMs was
ignored by the BIOS. That is now making me nervous about recommending
mixed sets of DIMMs to people, like 2x1GB + 2x512MB. If would be
pretty embarrassing to suggest such a config, and get a posting
later lamenting the fact that the new DIMMs were ignored. So,
whether the OP goes with a 3GB configuration or a 4GB configuration,
both of those configurations will require finding working
examples in the private forums somewhere.

Paul
 
G

Guest

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

Paul wrote:
> I was looking for a Microsoft web page that would address the issue
> of needing 5GB of address space, but only having 4GB of physical
> memory. The KB articles that attempt to address memory limits, aren't
> too precise about exactly what is supported.
I didn't find anything really helpful neither :-(.
There is a KB article in german, http://support.microsoft.com/kb/875352
which claims to be a translation of
http://support.microsoft.com/kb/875352 but contains a lot more
information (the english version is much newer, looks like MS shortened
it considerably). Specifically, it has a (very confusing) paragragh
about driver compatibility, and at the end of this paragraph it says the
HAL and the kernel memory management has been changed and ignores all
addresses above 4GB to avoid driver compatibility problems (this
statement at the end more or less contradicts what was said previously
in the same paragraph). There are other reports this indeed worked in
SP1, but not SP2
(http://www.windowsforumz.com/Configuration-Manage-XP-SP1a-versus-SP2-Memory-Limitations-ftopict370706.html).
If I'd have to guess I'd say MS changed the behaviour because with SP1
noone except people who knew what they were doing and what problems to
expect enabled PAE, but with SP2 /NoExecute is the default (and this
more or less implies /PAE) so far more people would encounter driver
compatibility problems. Still, a pity you can't get the old behaviour
back :-(.


> This article mentions that WinXP will access above 2^32 when using
> the PAE switch. Now, AMD64 has the MMIO feature, and I don't know if
> that means the whole computer will work properly or not, with this
> 5GB address space, 4GB physical configuration. A question that has
> come up in my mind about this particular configuration, is the layout
> of the kernel memory versus user space - if the kernel lived in the
> upper GB of memory, perhaps PCI I/O operations would never go near
> 2^32, further improving the odds that it might work.
>
> http://blogs.msdn.com/carmencr/archive/2004/08/06/210093.aspx
>
> Another disturbing thing in that thread, was the posting at the
> bottom of the page. A user upgraded to SP2 and found the system
> reported only 3.18GB when before SP2 it reports 4GB.
Sounds like if that would happen if you'd use /NoExecute, but according
to this article with /PAE instead that should indeed work.
I got that information about windows xp just not supporting memory
mapped above 2^32 from c't (not sure which edition that was), which is
usually considered a reliable source.

Sounds to me like it's time to upgrade to XP64 if you really want to use
4GB of ram. MS offers quite a reasonable upgrade from XP Pro if you
bought it not too long ago, which imho has only one serious drawback,
your license for the old XP version you had is no longer valid, thus you
cannot legally do a dual-boot (if some apps you use don't run) unless
you buy another copy.

Roland