theTroll :
mikrev007 - As a software engineer for the last 20 years let me explain this to ya.
When you write an application you have to understand what memory model you will be using. Is it going to be a 32 bit or 36, or 64 bit application because it changes how to have to handle the memory.
So PAE will only help if the application is written to make use of this. This costs lots of extra time and money due to the code and the testing of that code. Microsoft SQL server and a few other of it's severs applications (Exchange) are written to do this. I don't know of ANY non-server application that is written to use PAE addressing.
So PAE does not fix the 4 GB limit.
Now you know.
theTroll
Well I'm not a programmer I have to work for a living
but even I understand the basics. If you write a 32-bit application it uses VIRTUAL memory and thus PAE and PHYSICAL memory are transparent to it. Perhaps you could tell us exactly what software application(s) you have written in the last 20 years.
Let's try to simply explain how things work.
Hardware is normally mapped from 4GB downwards so if you have 1GB of hardware addresses then it will reside in the 3GB to 4GB address range. Now if you have 4GB of RAM the first 3GB can be mapped to the 0 to 3GB addresses but the last 1GB needs to be remapped and most modern BIOSes will do this by using the 4GB to 5GB address range. Now even though PAE may be enabled by default to allow hardware DEP and hence the 32-bit Windows OS is using 64-bit page tables, Microsoft decided to IGNORE memory addresses above 4GB so even though that memory is there to be used, the client Windows OSes from XP SP2 onwards ignore it.
When an application is written for windows it uses RVA virtual addresses. We may have one process that runs from address 0x400000 and a second or more processes that also run from address 0x400000. How is it possible that they can run from the same address, because it's a virtual address and each process normally maps to different physical addresses. This is very useful because as our physical memory addresses become fragmented we can still map, for instance 4K blocks, as contiguous virtual memory.
Some points that may be relevant.
PAE has been around since 1995, long before EMT64 and if you check the CPU's that support PAE you will see they have at least 36 pins or balls for address lines which means being able to physically address 2^36 or 64GB.
Some drivers may behave badly with PAE but with certified drivers I would hope not.
If the North Bridge does not support more than a 32-bit interface then even a 64-bit system will not allow the use of a full 4GB or more of RAM.
32-bit Windows processes will generally be limited to using 2GB RAM under both a 32-bit and 64-bit OS. With large address aware processes this may be expanded to 3GB for 32-bit with 4GT and 4GB for 64-bit.
Some 32-bit desktop / client linux distro's such as Fedora 10/11 support more than 4GB of RAM by default (not 32-bit Ubuntu).
It is possible for Windows 32-bit Vista / 7 to support more than 4GB as can be tested via a fairly simple kernel modification.
And finally getting back to the OP's question IMHO it probably is a waste for most people.