Archived from groups: alt.comp.periphs.mainboard.asus (
More info?)
On Fri, 04 Feb 2005 01:24:54 -0500, nospam@needed.com (Paul) wrote:
>In article <ljv501546kfp6p50bidckeklshrjspameo@4ax.com>,
>npowroz.remove.this.part@rogers.com.and.this.part wrote:
>
>> I just finished building a new system for a customer based on a
>> P5AD2-E Premium (a really nice board!).
>>
>> Since the system will be used as a web and database server, I have
>> maxed it out in memory and disc configuration, but for some reason,
>> all of the memory isn't visible. I have 4 1GB Kingston DDR2 modules
>> installed, which theoretically takes the board to its rated 4GB
>> maximum limit.
>>
>> However, the BIOS can only see 3.25GB of memory. I know that each
>> stick of memory is good. If I only install a pair of DIMMs, the full
>> 2GB in the pair is accessible. Each pair combinaton works perfectly,
>> but with all 4, we only get 3.25GB.
>>
>> Is this a known limitation that I will simply have to live with, or
>> have I overlooked something in the BIOS? I think I am running the
>> latest BIOS flash, but I haven't gone into that area as yet.
>>
>> Any advice appreciated.
>>
>> Cheers
>> Norm
>
>Page 159 of the 925X Express Chipset document, shows an example
>of what is going on. The address space available is 32 bits,
>and some of the address space must be made available for PCI or
>PCI Express devices (memory mapped I/O).
>
>ftp://download.intel.com/design/chipsets/datashts/30146402.pdf
>
>What this means is, the Northbridge "throws away" some of the
>DRAM memory, and addresses falling into the top most area of
>memory, go out on the PCI or the PCI Express bus.
>
>There are some computing products, that support the remapping
>of DRAM memory. What they do, is they leave a space of, say
>1GB or so (from 3GB to 4GB), for access to the various busses
>on the motherboard, and then, they lift the GB of displaced
>RAM and map it from 4GB to 5GB. (To get up there, of course
>the processor has to have more address bits available, and
>some mechanism in the OS to use them. Google on PAE, PSE,
>36 bit, address space, as search terms.)
>
>Section 4.4 pg 128 of this document, illustrates a chipset
>that supports the concept. I just happened to have this
>doc in my collection, and there are more capable chipsets
>than this one.
>
>http://www.intel.com/design/chipsets/e7205/datashts/25193702.pdf
>
>There could well be a similar function supported
>on Athlon64 motherboards that accept an Opteron processor.
>Page 80 of this document discusses how it works.
>
>http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF
>
>At least one Asus AMD motherboard doesn't seem to support that
>"hoisting" function, and I cannot tell you whether the
>Northbridge has anything to say in the matter or not,
>on Athlon64. (The memory controller is in the processor,
>so all that should be needed is a compliant BIOS.)
>
>There is some info here, on OS support for large memory.
>The main message of this posting (the second half of the posting),
>is a lot of things have to be working right, before you get to
>use a large memory.
>
>http://groups.google.ca/groups?threadm=2mfmc8Fmff71U1%40uni-berlin.de
>
>HTH,
> Paul
Whoa!! I have to say that I never expected an answer as well
researched or presented as the one you just gave. Quite honestly, I
feel overwhelmed. The amount of information, and the obvious depth of
your knowledge and research, are impressive to say the least.
You have provided me with everything I need to know about the problem,
as well as highlighting the fact that there is nothing I can do about
it, since it's a hardware imposition. It kind of harks back to the old
8088 days in which memory between the 640KB and 1MB boundaries was
needed for hardware communication areas, such as video or other
direct-memory and memory-mapped devices. And of course, everyone used
memory managers such as QEMM in order to maximize the "high memory"
space.
Thanks again, Paul.
Cheers
Norm