PC 4GB RAM limit

[Tim Anderson]

Because it is an unnecessary expense. While there may be uses for
addressing over 4GB to conveniently handle large databases etc.,
there is virtually no real need for that much actual memory.

Well, maybe. Personally I want to run multiple virtual PC machines on a
64-bit OS. I reckon the extra RAM will bring real benefits, as I want to
give each virtual PC 1GB RAM.

Of course what you say is true for everyday apps. On the other hand, I'm
irritated because Intel, in this particular case, give you no warning in the
assembly manual that the 4th GB will be wasted and that you should not
install more than 3GB. This information is in the technical spec (PDF on the
CD) but should be highlighted in the printed guide IMO.

Tim

 

[Tim Anderson]

If you don't install memory in a physical area it isn't wasted.
Virtual memory systems involve remapping memory anyhow, so the
software doesn't see any gaps. There is no real problem.

The "real problem" is a bunch of motherboards that are advertised as 4GB,
accept 4GB, but only make 3GB available (could be more than that, but in my
case a full 1GB is lost).

Tim

 

[Tim Anderson]

With a larger virtual address space (and the larger physical address space
for PCIe), most of these problems are removed.

That's what I figured. With PCIe and AMD64 extensions (and a 64-bit OS) I
was surprised to run into this problem.

Tim

 

[Last Boy Scout]

The cheaper motherboards (ie. most of them) for P4 or Athlon 64 support a
maximum of 4GB RAM, even if the motherboards and processors support the
AMD64 extensions that can address sqillions of GB. On the other hand, more
upmarket boards support more RAM - up to 24GB or maybe more - for Xeon and
Opteron.

If you buy one of these 4GB boards and install 4GB RAM, you don't get the
use of all of it. The top of the 4GB address space gets shadowed by system
functions such as PCI Express addressing. This is not just a small detail
- typically you lose 1GB of your 4GB. See:

http://www.itwriting.com/blog/?postid=152

I can't at the moment find a clear explanation of this. I understand about
the shadowing, but the question of course is why a modern board can't use
a higher range of addresses to make the full 4GB available to the OS. The
manufacturers mutter about "PC Architecture", but then again they also
make boards that *do* overcome this limit.

With PAE, PC processors have been able to address more than 4GB for years.
So why are we still running into this limit?

Tim

If you need more than 4 GB of RAM then you need a specialized
motherboard and operating system to be able to take advantage of it.
Windows XP was not really designed for a high-end environment, it was
designed for big business and the consumer market. Maybe a 64 bit
operating system can do a lot better. The OS plays an integral part of
the total package. Even if you can address more RAM on the hardware
side, your OS has to be able to use it. There are a lot of processes
that are running in the background for an OS to operate properly. All
of these take RAM. If you throw 2 processors into the mix and are using
the porper type of OS to take advantage of the situation it will take
more memory and cache space to manage the extra processor. Then you may
want to look at things like Massive Parallel Processing MPP, Symmetrical
Multi-Processing SMP, and non-standard memory allocation between the
processors NUMAQ. This is why the software licensing on a mainframe is
around $100,000 annualy.

 

[Kadaitcha Man]

Last Boy Scout, <[email protected]>, the out of place, fanatical Jehovah's
witness, and minor, worthless author of usenet posts, prattled:

Windows XP was not really designed for a high-end environment

BWAHAHAHAHAHAHAHAHAHAHAHAHAHAHA! You ****ing idiot. Windows Enterprise
Server 2003 is nothing but a glorified XP, you ****ing brain-dead fannyfart.

 

[Al Dykes]

Well, maybe. Personally I want to run multiple virtual PC machines on a
64-bit OS. I reckon the extra RAM will bring real benefits, as I want to
give each virtual PC 1GB RAM.

Of course what you say is true for everyday apps. On the other hand, I'm
irritated because Intel, in this particular case, give you no warning in the
assembly manual that the 4th GB will be wasted and that you should not
install more than 3GB. This information is in the technical spec (PDF on the
CD) but should be highlighted in the printed guide IMO.

Tim

The decision to limit user's physical memory is a cost and marketing
decision on the part of the mobo manufacturer. There may be an AMD64
mobo that allows more than 4GB memory but I doubt it. The number of
home users that need this much memory noy is about zero, and business
users can be hit up for a more expensive mobo.

If you get an dual-Opteron Mobo you can stuff 8GB into it and I'm
pretty sure that a single CPU system will access all 8GB and you'll
have the option of adding a second CPU later.

(The second memory bank will be slightly slower than the one closest
to the CPU.)

 

[Tim Anderson]

The decision to limit user's physical memory is a cost and marketing
decision on the part of the mobo manufacturer.

I have no problem with that. I do have an issue with marketing a board as
"4GB" when it is in effect 3GB.

Tim

 

[Tim Anderson]

If you need more than 4 GB of RAM

Actually is was just the 4GB I wanted, not "more than".

Windows XP was not really designed for a high-end environment, it was
designed for big business and the consumer market. Maybe a 64 bit
operating system can do a lot better.

As I also mentioned, this is not an OS issue; and in any case I'm using
64-bit operating systems.

Tim

 

[David Maynard]

I have no problem with that. I do have an issue with marketing a board as
"4GB" when it is in effect 3GB.

Tim

Unless there's something odd about the motherboard it *is* 4GB.

What happens is the BIOS should remap the top 1G above the I/O area so you
end up with 3GB RAM - 1GB I/O - 1GB RAM.

XP/2000 Pro won't see it, however, because they're limited to a 4 gig space
and remapping the top 1G above 4 puts it 'out of range'. Win2k Advanced
Server and the Win2k3 variants can use it, though, as they support 8 and
16GB address spaces (with PAE switch).

With Linux you have to make sure the kernel (2.4.x or higher) is compiled
with the PAE option enabled or else you have the same 4GB situation as XP.

 

[Al Dykes]

Unless there's something odd about the motherboard it *is* 4GB.

What happens is the BIOS should remap the top 1G above the I/O area so you
end up with 3GB RAM - 1GB I/O - 1GB RAM.

XP/2000 Pro won't see it, however, because they're limited to a 4 gig space
and remapping the top 1G above 4 puts it 'out of range'. Win2k Advanced
Server and the Win2k3 variants can use it, though, as they support 8 and
16GB address spaces (with PAE switch).

With Linux you have to make sure the kernel (2.4.x or higher) is compiled
with the PAE option enabled or else you have the same 4GB situation as XP.

This sounds like a 32 bit system. I assume that PAE doesn't exist in
a 64 bit system (or for backward compatibility the OS simply maps PSE
segemnts into normal chucks of 64 bit address space. )

PAE is slow becasue of the context switches. A package like Oracle
can make efficient use of it but it's no replacemnt for a flat 64 bit
address space.

 

[Tim Anderson]

Unless there's something odd about the motherboard it *is* 4GB.

What happens is the BIOS should remap the top 1G above the I/O area so you
end up with 3GB RAM - 1GB I/O - 1GB RAM.

This it is *not* doing. I tend to agree with you though.

XP/2000 Pro won't see it, however, because they're limited to a 4 gig
space and remapping the top 1G above 4 puts it 'out of range'.

I'm running 64-bit Operating systems (Windows x64 and Linux) that have no
problem addressing this range.

Tim

 

[Adam Webb]

what motherboard is it? my Asus A8V has some option somewhere in the bios to
"enable" 4gig or disable (ie "steal" memory for itselfs) i think.

 

[David Maynard]

This sounds like a 32 bit system.
Yes

I assume that PAE doesn't exist in
a 64 bit system (or for backward compatibility the OS simply maps PSE
segemnts into normal chucks of 64 bit address space. )

PAE is slow becasue of the context switches. A package like Oracle
can make efficient use of it but it's no replacemnt for a flat 64 bit
address space.

 

[David Maynard]

This it is *not* doing.

How did you determine that?

I tend to agree with you though.




I'm running 64-bit Operating systems

I see.

(Windows x64

I don't know what capabilities and bugs might exist in a pre beta O.S.

and Linux) that have no
problem addressing this range.

I'm not sure but what you may be jumping to premature conclusions. By that
I mean there are still configuration issues and 'limits' in addition to the
'theoretical maximums'. For example, it may be that certain assumptions the
O.S. is making about memory allocation are different than the ones the
motherboard is doing.

I'm afraid I can't debug your situation as I don't have that motherboard
nor 4 gig of ram but I've seen various problems mentioned about 4gig and
'64-bit' linux. One person, for example, complained that 4 gig was fine
till he 'upgraded' to the 64 bit kernel and, in that case, it seemed he
ended up with a mix of 64 bit addressing and PAE which, of course, are two
different things.

Not saying that's your problem but it's an example of how things are
sometimes not as they seem nor as simple as the 'theory'.

You sure something isn't causing a 32 bit 'compatibility' mode?

 

[DevilsPGD]

In message <[email protected]> David Maynard

I don't know what capabilities and bugs might exist in a pre beta O.S.

You must not get much news up whereever it is you live, Windows x64 has
been in beta for over a year and went RTM several weeks ago.

In short, it hasn't been pre-beta for many many months.

 

[Tim Anderson]

How did you determine that?

Because the BIOS reports "1048 MB consumed by system" on boot up, before the
OS loads. The problems is also documented in the tech. spec.

I don't know what capabilities and bugs might exist in a pre beta O.S.

Full release build.

Suse Linux 64-bit can't see the RAM either - not surprising, as it it lost
before the OS boots.

Tim

 

[Tim Anderson]

How did you determine that?

Because the BIOS reports "1048 MB consumed by system" on boot up, before the
OS loads. The problems is also documented in the tech. spec.

I don't know what capabilities and bugs might exist in a pre beta O.S.

Full release build.

Suse Linux 64-bit can't see the RAM either - not surprising, as it it lost
before the OS boots.

Tim

 

[Tim Anderson]

what motherboard is it? my Asus A8V has some option somewhere in the bios
to
"enable" 4gig or disable (ie "steal" memory for itselfs) i think.

This is an Intel board unfortunately.

But yes, it sounds like that would fix it.

Tim

 

[David Maynard]

In message <[email protected]> David Maynard



You must not get much news up whereever it is you live, Windows x64 has
been in beta for over a year and went RTM several weeks ago.

In short, it hasn't been pre-beta for many many months.

Right.

The problem was me doing a half dozen things at the same time and thinking
of something other than XP/64.

 

[David Maynard]

Because the BIOS reports "1048 MB consumed by system" on boot up, before the
OS loads. The problems is also documented in the tech. spec.

I see. Well, since you hadn't mentioned it again I threaded back to see if
you said what board it was and from the docs it appears to not be a '64
bit' board. By that I mean, it apparently doesn't have anything to support
64 bit specific features and, in particular, no more than the traditional 4
GB address space, which could explain why it doesn't remap HiMem.

The literature targets it at 'desktops' and, specifically, XP/2000 as the
supported O.S. and neither of them would be able to use remapped memory
even if the board did it, which might also explain why they didn't bother.

I was thinking in terms of a board that actually supports (part of) the 64
bit address space, in some form and if, for example, you look at the
documentation for the Intel server board that was posted earlier in this
thread you'll see it has a BIOS option for mapping, or not, that space to
above installed memory. (It also supports 8 GB.)