I give up - or - help? - or meh I buy a PS3

the answers are from a mod (slowtarget):
Quote Originally Posted by _Pax_ View Post
You do realise, it is the OPERATING SYSTEM, not the individual program itself, which addresses and allocates memory ... right?

Yes, I'm well aware of memory allocation routines. Somewhere in my eleven years of professional software development, including a year of windows kernel-mode debugging, five years of multiplatform application development and two years of work patching the Linux kernel and third party applications, I ran across descriptions of how memory allocation works.

I'm well aware that the OS allocates memory. I'm also aware that each process has its own memory address space, a space which, in the i686 ISA, is 4GB, regardless of the amount of memory visible or addressable by the operating system. This allows me to understand that when, for example, a computer is running an operating system based off of 64 bit addresses and capable of addressing far more memory, any processes running in a compatible 32 bit mode will still only be capable of seeing those same 4GB of space. Since I have experience coding Win32 system libraries, I'm also aware that the OS reserves the upper half of that 4GB address space limiting effective address space of a Win32 process to just 2GB, unless the binary is compiled with the /LARGEADDRESSAWARE flag which will allow the process to make effective use of another 1GB of reserved space.

Since I know that the Civ V binary was compiled with this flag, I know that Civ V is allowed to request memory allocation of up to 3GB for its own use. While it technically true (and pedantic) to declare that the OS allocates memory, anyone who has actually spent time coding does not refer to a malloc() call as "a request to have the OS allocate memory", but "a memory allocation". No one who has any business writing a malloc() call is under the illusion that the allocation is done in user mode instead of kernel mode.

Quote Originally Posted by _Pax_ View Post
After all, my graphics card alone will take up to 4GB of memory, all by itself, if allowed and needed (869MB dedicated, 3199 shared - see snippet of DxDiag, below).

No... what you have is a memory block allocated by Vista/7 which is dedicated to the video subsystem by default. It's called (boringly) Shared System Memory. Vista/7 earmarks the addresses for use but only uses them if the hardware does not supply sufficient memory. The memory is not allocated to the video subsystem, simply reserved. If the system runs low on other memory and requests another allocation, it will reclaim unused memory from the Shared System Memory and allow processes to use it.

So: No, your graphics card will not take up to 4GB of memory unless you're running an 6 monitor system. It is not currently taking that memory and any process requesting that memory will be allowed to allocate it. Wait... no, that should be: ...any process requesting that memory will receive notification from the operating system via the return value of the system call that the requested amount of memory has been marked for the inclusion in the requesting processes memory address space.

Quote Originally Posted by _Pax_ View Post
You don't strike me as being particularly knowledgeable about hardware and software issues yourself, not with a comment like that one. So perhaps you're not the best person to be trying to take others to task for a lack of their own knowledge in such areas ...?

Do you still feel this way? Did I say anything incorrect in my previous comment?

Quote Originally Posted by _Pax_ View Post
Re-read that. BEcause when I read it, I see that poster claiming that with one city and two military units, and while sitting there doing nothing, with the game waiting for the player to click "next turn" ... the game was constantly allocated 4kb of system memory per second.

Agreed. I misread.

However, I doubt their methods. There are plenty of reasons for that to occur and I've seen Civ V's memory climb... but not when it was in stasis. There is a lot of pre-loading/lazy-loading and background calculations which happen in the first few minutes after loading a game. After an hour, the memory is pretty stable.

A number of people have profiled this. So far, no one has shown a persistent memory leak. I doubt that this report is the first. I could be wrong. If you or the OP has memory profile data over a long period of time that shows a leak, then we'll make a new thread and I'll personally see that Greg or the Techs take a look at it.

Quote Originally Posted by _Pax_ View Post
That shared memory is coming out of the 8GB of System RAM my machine has. A memory leak - which involves System Ram - could eventually begin to impair graphical performance.

No. That's not the way it works. You've got it wrong in a couple places.

1: Memory leaks happen in process memory address space, not system memory address space. A memory leak is, by definition, portions of the process address space which are no longer directly addressed or managed by the process, making them, effectively, lost and difficult or impossible to correctly free. It must be process space. As such, a 32 bit, large address aware, Windows binary can only leak, at most 3GB of RAM. On a system with 8GB of RAM, this will never become a danger to the system as the system will always have 5GB of RAM not allocated to that process, and that is almost assuredly more than what is needed to keep the system functioning long enough to recover from the inevitable crash of the leaking process.

2: The shared memory, as I've explained, is not allocated memory and does not "steal" usable memory from processes. The first reason is that you've got a video card with dedicated RAM and so the shared memory is not used at all by the video subsystem. The second reason is that, even if you had a video card without dedicated memory, the shared memory is reserved at a lower priority than incoming allocation requests from processes. In the case that the video subsystem actually needed the Shared System Memory and found insufficient memory available for its current process, the video subsystem itself would shut down, terminating the current process (almost certainly the one providing the greatest load).