Quote:
What he does not mention is that the VMem is limited to 32bits too. So, if the amount of RAM in the computer, plus HW-allocations, does already fill up all the 32bit addressspace, then there simply is no difference! This is because as he mentions, these minimum allocations are made as VIRTUAL MEMORY - the virtual memory is limited to 32bit however. Now, if you have just as much RAM as you have VMem, then you cannot gain more RAM by swapping out unused VMem pages.
Think of it like this: Forget RAM for now. The computer has 32bits (4GB) of conceptual virtual memory. Everything including some hardware stuff needs to fit into that. Parts of this VMem can then either be mapped to disk, or mapped to RAM, but the total amount of VMem cannot exceed 4GB - you cannot have more then 4GB in TOTAL loaded - this limit cannot be exceeded, also not with a swapfile. But if the amount of RAM you have, plus hw-stuff, does already add up to 4GB, then swapping becomes pointless.
Actually, that's not true. On a 32-bit machine, the total amount of physical memory/virtual memory _can_ exceed 4GB. What _cannot_ exceed 4GB is the physical address space of each _process_ on the system.
For example, I am running a
modified kernel on a 32-bit Win7 machine at work. It has 8GB physical memory, and I see total 8GB physical memory in task manager. Also, I can run a couple of 1.5GB virtual machines without any slowdown in speed. Without the mod kernel, I can only see 3GB physical RAM, and can't start more than 1 VM.
In general, I think it's best to leave swapping on. However, if you are only running small apps and not really taxing the memory limits of your machine, then turning off swapping probably wouldn't be a problem. But once you start a memory intensive app like database server or photoshop, then you are going to encounter problems and have to enable swapping anyway.
Also, I think the memory manager on Win7 is much better than WinXP. The sluggish recovery after minimizing an app on WinXP does not happen on Win7 anymore.