A8N-sli memory - beware it won't take 4Gb

B

Blonks

Even though the board claims to support 4Gb of RAM is won't take 4 x
1GB sticks. This seems to be cuased by the ram not being buffered and
hence the board cannot support the power requirements. There is a
compatibility table on the Asus page that shows which dimms are
supported and 4 x 1Gb shows non are supported.

Has anybody managed to get such a setup working i.e. 4Gb of RAM, with
this or any other modern non-server board? I wanted to use all the
memory for running multiple Vmware machines.

Thanks

Ian
 
3

3200+

Blonks said:
Even though the board claims to support 4Gb of RAM is won't take 4 x
1GB sticks. This seems to be cuased by the ram not being buffered and
hence the board cannot support the power requirements. There is a
compatibility table on the Asus page that shows which dimms are
supported and 4 x 1Gb shows non are supported.

Has anybody managed to get such a setup working i.e. 4Gb of RAM, with
this or any other modern non-server board? I wanted to use all the
memory for running multiple Vmware machines.

Thanks

Ian

It does. We do it quite a lot.

Jon
 
P

Paul

Blonks said:
Even though the board claims to support 4Gb of RAM is won't take 4 x
1GB sticks. This seems to be cuased by the ram not being buffered and
hence the board cannot support the power requirements. There is a
compatibility table on the Asus page that shows which dimms are
supported and 4 x 1Gb shows non are supported.

Has anybody managed to get such a setup working i.e. 4Gb of RAM, with
this or any other modern non-server board? I wanted to use all the
memory for running multiple Vmware machines.

Thanks

Ian

Maybe you put registered sticks in the board, and the board beeped ?
It takes unbuffered DIMMs.

The OP here tried 4x1GB. With two PCI-E video cards, he got 2.25GB
available, and with one PCI-E video card, he got 2.75GB. This implies
that the A8N-SLI is just about the worst desktop board you could
select for a memory hungry application. The lost memory makes
way for necessary I/O space (PCI-E video cards, PCI-E plugin cards,
PCI bus cards).

http://groups.google.ca/[email protected]

This is an example of a desktop board that claims to support
8GB.

"SK8N Specification Summary
Socket 940 for AMD Opteron processor
NVIDIA nForce 3 Pro150
Dual-Channel memory architecture
Supports PC2700/PC2100/PC1600 ECC DDR SDRAM
Registered DIMMs
4 x 184-pin DDR DIMM sockets for up to 8GB memory"

The SK8N does support memory hoisting, as noted here. If you have
8GB of memory, what memory hoisting does, is map it as 3GB down
low, leaves a gap between 3GB and 4GB, then puts the remaining
5GB above the 4GB mark. So, one chunk of 3GB, one chunk of 5GB,
and via some mapping, it looks like a contiguous virtual 8GB
chunk to the system. The hole is then used for the I/O space.
Apparently, with the right version of BIOS, the SK8N supports
this function, which tells you support for this was an
afterthought at Asus.

http://groups.google.ca/[email protected]

What is strange, is if I look at the 26094 document at AMD, I
get the impression that any athlon64/opteron should support
the memory hoisting function. So, maybe this is purely a
BIOS limitation, in which case complaining to Asus might
eventually get it fixed. (Due to AMD's idea of modular
documentation, when a feature is seen in their documentation,
you have no way of knowing whether the feature works on
S754, S939, S940, Opteron etc. Truly a pathetic way to
document hardware.)

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF

The memory hoisting function is not without side effects. The
AMD document states that certain memory optimizations are
disabled, so the memory bandwidth might drop a bit. Also, you
need OS and application support for this kind of thing, so
perhaps a server style OS, plus a version of VMware intended to
run on server type machines, would be required. I don't know
anything about that stuff.

Chances are, if a motherboard supports Opteron processors, it
will do a good job in the memory arena. Unless you can find a
server motherboard manufacturer that dabbles in desktops, and
does a good job on all their BIOS.

Another tidbit:
http://groups.google.ca/[email protected]

I tried downloading a Tyan desktop (S2865) manual, hoping that
maybe they implement memory hoisting, but the manual is so bad,
I cannot tell if this function is implemented or not. They have
an entry called "Bottom of 32-bit[31:24] IO[E0]" and another
called "S/W memory hole Romapping", and there is zero info on
what they do.

Good luck in your quest,

Paul
 
M

Mercury

have you tried xp 64bit?


Paul said:
Blonks said:
Even though the board claims to support 4Gb of RAM is won't take 4 x
1GB sticks. This seems to be cuased by the ram not being buffered and
hence the board cannot support the power requirements. There is a
compatibility table on the Asus page that shows which dimms are
supported and 4 x 1Gb shows non are supported.

Has anybody managed to get such a setup working i.e. 4Gb of RAM, with
this or any other modern non-server board? I wanted to use all the
memory for running multiple Vmware machines.

Thanks

Ian

Maybe you put registered sticks in the board, and the board beeped ?
It takes unbuffered DIMMs.

The OP here tried 4x1GB. With two PCI-E video cards, he got 2.25GB
available, and with one PCI-E video card, he got 2.75GB. This implies
that the A8N-SLI is just about the worst desktop board you could
select for a memory hungry application. The lost memory makes
way for necessary I/O space (PCI-E video cards, PCI-E plugin cards,
PCI bus cards).

http://groups.google.ca/[email protected]

This is an example of a desktop board that claims to support
8GB.

"SK8N Specification Summary
Socket 940 for AMD Opteron processor
NVIDIA nForce 3 Pro150
Dual-Channel memory architecture
Supports PC2700/PC2100/PC1600 ECC DDR SDRAM
Registered DIMMs
4 x 184-pin DDR DIMM sockets for up to 8GB memory"

The SK8N does support memory hoisting, as noted here. If you have
8GB of memory, what memory hoisting does, is map it as 3GB down
low, leaves a gap between 3GB and 4GB, then puts the remaining
5GB above the 4GB mark. So, one chunk of 3GB, one chunk of 5GB,
and via some mapping, it looks like a contiguous virtual 8GB
chunk to the system. The hole is then used for the I/O space.
Apparently, with the right version of BIOS, the SK8N supports
this function, which tells you support for this was an
afterthought at Asus.

http://groups.google.ca/[email protected]

What is strange, is if I look at the 26094 document at AMD, I
get the impression that any athlon64/opteron should support
the memory hoisting function. So, maybe this is purely a
BIOS limitation, in which case complaining to Asus might
eventually get it fixed. (Due to AMD's idea of modular
documentation, when a feature is seen in their documentation,
you have no way of knowing whether the feature works on
S754, S939, S940, Opteron etc. Truly a pathetic way to
document hardware.)

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF

The memory hoisting function is not without side effects. The
AMD document states that certain memory optimizations are
disabled, so the memory bandwidth might drop a bit. Also, you
need OS and application support for this kind of thing, so
perhaps a server style OS, plus a version of VMware intended to
run on server type machines, would be required. I don't know
anything about that stuff.

Chances are, if a motherboard supports Opteron processors, it
will do a good job in the memory arena. Unless you can find a
server motherboard manufacturer that dabbles in desktops, and
does a good job on all their BIOS.

Another tidbit:
http://groups.google.ca/[email protected]

I tried downloading a Tyan desktop (S2865) manual, hoping that
maybe they implement memory hoisting, but the manual is so bad,
I cannot tell if this function is implemented or not. They have
an entry called "Bottom of 32-bit[31:24] IO[E0]" and another
called "S/W memory hole Romapping", and there is zero info on
what they do.

Good luck in your quest,

Paul
 
R

Roland Scheidegger

Blonks said:
Even though the board claims to support 4Gb of RAM is won't take 4 x
1GB sticks. This seems to be cuased by the ram not being buffered and
hence the board cannot support the power requirements. There is a
compatibility table on the Asus page that shows which dimms are
supported and 4 x 1Gb shows non are supported.

Has anybody managed to get such a setup working i.e. 4Gb of RAM, with
this or any other modern non-server board? I wanted to use all the
memory for running multiple Vmware machines.

The A8N-sli likely supports 4GB with bios 1006, from the bios changelog
"5.Add Memory Re-map function SETUP Item", this should do the trick.
What has been said about needing an OS which supports this is still true
though - windows xp professional (32bit) won't this support at all, even
with PAE the address range is limited to 2^32 - thus the remapped memory
is out of reach of the OS. Windows XP Server (standard edition) also
"only" supports 4GB of ram, but this time for real, i.e. does not limit
the address range to 2^32, thus you will get the full 4GB. With even
more memory (I'm not 100% certain, but believe the socket 939 boards
should actually be able to take 4 2GB dimms, though these would need to
be built with 16 (unbuffered) 1gbit chips, and I've not yet seen such
memory modules, I have no idea if they will ever appear in the market as
memory manufacturers are unlikely to invest in ddr1 memory manufacturing
at this point) you'd need the advanced server or enterprise edition.
Or use linux or windows xp 64bit...

Roland
 
P

Paul

Roland Scheidegger said:
The A8N-sli likely supports 4GB with bios 1006, from the bios changelog
"5.Add Memory Re-map function SETUP Item", this should do the trick.
What has been said about needing an OS which supports this is still true
though - windows xp professional (32bit) won't this support at all, even
with PAE the address range is limited to 2^32 - thus the remapped memory
is out of reach of the OS. Windows XP Server (standard edition) also
"only" supports 4GB of ram, but this time for real, i.e. does not limit
the address range to 2^32, thus you will get the full 4GB. With even
more memory (I'm not 100% certain, but believe the socket 939 boards
should actually be able to take 4 2GB dimms, though these would need to
be built with 16 (unbuffered) 1gbit chips, and I've not yet seen such
memory modules, I have no idea if they will ever appear in the market as
memory manufacturers are unlikely to invest in ddr1 memory manufacturing
at this point) you'd need the advanced server or enterprise edition.
Or use linux or windows xp 64bit...

Roland

There are some 2GB unbuffered modules at the top of this page,
but they are only DDR333.

http://www.samsung.com/Products/Semiconductor/common/product_list.aspx?family_cd=DRM030202

The PC2100 version is even on the AMD Athlon64 qualified memory
list (near the bottom of the doc):

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Athlon_64_Unbuffered_DIMM_AVL.pdf

I cannot find any for sale though.

Fun times ahead :)

Paul
 
R

Roland Scheidegger

Paul said:
There are some 2GB unbuffered modules at the top of this page,
but they are only DDR333.

http://www.samsung.com/Products/Semiconductor/common/product_list.aspx?family_cd=DRM030202

The PC2100 version is even on the AMD Athlon64 qualified memory
list (near the bottom of the doc):

http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Athlon_64_Unbuffered_DIMM_AVL.pdf

I cannot find any for sale though.

Fun times ahead :)
Indeed. These modules use multi-die memory chips though. Maybe that's
the reason they can't reach DDR-400 speed (?), and I doubt pricing is
ever going to be mass-market compatible with such devices.

Roland
btw forgot to add previously, just because a board has the memory remap
option doesn't mean it's actually going to work in all OS, that's
especially true for Asus which have a habit implementing some things not
quite spec-compliant. The remapping was known to be broken on the SK8N,
it happened to work with some luck on W2k3 server, but didn't in linux
and may very well not have worked in w2k xp64 neither. No idea if asus
have fixed that and are implementing it how they should now in the a8n...
 
P

Paul

Roland Scheidegger said:
Indeed. These modules use multi-die memory chips though. Maybe that's
the reason they can't reach DDR-400 speed (?), and I doubt pricing is
ever going to be mass-market compatible with such devices.

Roland
btw forgot to add previously, just because a board has the memory remap
option doesn't mean it's actually going to work in all OS, that's
especially true for Asus which have a habit implementing some things not
quite spec-compliant. The remapping was known to be broken on the SK8N,
it happened to work with some luck on W2k3 server, but didn't in linux
and may very well not have worked in w2k xp64 neither. No idea if asus
have fixed that and are implementing it how they should now in the a8n...

On the Samsung web page, it says the module uses 128Mx8 chips.
i don't think they are stacked. When I compared the input
capacitance on some 1GB and some 2GB modules, they are the
same. It is academic anyway - Micron had some 4GB modules for
$6000 or $7000, so I doubt anyone would be able to afford a module
like that. Still, it is fun to dream...

Paul
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Top