Edit: I've been logged out twice from this site while I looked more closely at the relevant links. Of course, the following does not take into account the rapidity with which you guys respond to each other.
I'm with
Grump642: you will likely never need more than 10 GB total active memory at any one time. You probably could get by with quite a bit less.
You are correct in that the pagefile(s) do not need to be that big,
nandersen. The pagefile is not mapping RAM, it's supporting virtual memory needs.
Note that if you have one physical drive (with one or more logical-drive partitions), you should still have only one pagefile. And that size-of-RAM rule of thumb was more valid in early Silicolithic times when max addressable RAM was far less than today. But even then, I viewed it like a broken 12-hour analog clock that's right twice a day -- that banal "truth" is not relevant to its intended use.
I question the article's conclusions. It references MS fellow Mark Russinovich (formerly of SysInternals, which seems to be now a part of Microsoft) by linking
his Windows Vista Kernel article. Unlike Russinovich articles, the TweakHound article is relatively bereft of
reasons supported by measured data.
A couple of the articles cited on the pagefile section (page 6) of that article give you rules of thumb with caveats:
"If there is no other information available, the normal recommendation of 1.5 times the amount of RAM in the computer is a good place to start." That's the
total pagefile size.
But there is other info available to us these days. Let's consign "rule of thumb" to the wastebin that already contains "wild stab at it", "shooting blindly", and "blue sky". I propose a rule of actual metrics.
Your pagefile is overflow protection for your virtual memory support. It isn't a mirror of your actual RAM*, so doubling it can waste disk space. It should not be tied to the RAM available, rather it should be related to the memory needed by OS and apps.
An application's available memory map is considered to be the size of RAM + pagefile, up to a point.
For 32-bit Windows, applications can use up to 2 GB (32-bit OS application process space default) each. The 32-bit OS needs another 2GB (again, default). Large pagefiles on 32-bit OSes can waste space that will never be used. I think a 3 GB pagefile with 2-4 GB RAM is plenty on 32-bit XP and 32-bit Vista (SP1).
A 64-bit OS can make use of more RAM + pagefile, although practical addressing limitations still exist; it's not a power-of-two thing. I think the new limit is now 128 GB, although OSes might limit that. And I believe that apps themselves still have a 4 GB limit.
My opinion: The pagefile size should be the size of the maximum usage of memory expected in the operation of your PC with its 64-bit OS, less the actual addressable RAM, plus a relatively small safety margin. Consider that it will contain parts of the OS and currently-loaded application
s (plural) that are not active, and it can contain parts of the application that has the focus -- parts that are not in use at the moment.
This means that with 64-bit systems, you might be able to get by without any pagefile at all, if you have enough RAM. (Of course, this doesn't account for OS idiosyncrasies. Older Windows OSes are bad about early use of pagefile when lots of RAM is still available.)
Edit: Looks like
EggChen has just added support to this statement.
On the other hand, more apps these days require much more memory. If you use graphics or sound tools along with game SDKs that need 4 GB each, you'll need more pagefile.
So: You can use the default pagefile while you gather those metrics about how much memory you are using. Then you can decide on optimization if you need it.
If 'twere I: I'd create a contiguous block for it not bigger than 6 GB on a single drive on a system with 4 GB RAM. I'm not sure that merely specifying a custom size in the Virtual Memory dialog makes it contiguous, but perhaps defragging will.
I'd say that with larger hard drives, you can allocate a large block of disk space that won't be missed, but it is not practical to allocate space that will never be used.
__________
*Note that this does not account for flush-all-to-disk hibernation; I don't know if that uses a separate file on the drive or not. I would hope so.
"I see your e-peen with my ball-peen, and I call. Oops. Them's the breaks."