gamerk316 :
First off, NEVER turn off Virtual Memory; all Windows Addressing is done by the virtual memory subsystem, all disabling Virtual Memory does is force a 1:1 Physical/Virtual RAM address matching. Causes far more issues then its worth, for no benefit.
My main question here, given you are using a notebook: What's the GPU? I'm betting its a crappy Intel iGPU, which reserves somewhere on the order of 2GB of RAM for the GPU.
1 GB overhead + 2GB reserved for Intel GPU = 3 GB reserved by the system.
That's my best guess.
I would be inclined to agree with you if it weren't for the fact that the OP reported that all 4 GB of memory was available to the 64-bit Windows environment.
So, even if the Intel graphics processor does make high demands on the memory, Windows 7 64-bit said 'No' and Windows 7 32-bit said 'Sure, take as much as you like'.
What's the difference? Same hardware, different results; how can this problem not be due to the 32-bit environment?
It seems to me that it is Windows 7 that is deciding how to allocate RAM, not the GPU adapter.
And the only way I can see this being down to the graphics side of thing at all is if the 32-bit drivers on this system are inefficiently coded. Perhaps they don't take the space above 4 GB into account, you know, the memory that is mapped to above 4 gig for the private use of drivers, the memory that I believe is exempted from page swapping?
Perhaps the driver fails to register its demands correctly and forces Windows to allocate twice the RAM even though the GPU will only ever use half of it.
This is easily checked by uninstalling the graphics adapter and then look at the situation when the GPU is making no demands at all. If it's a graphics driver problem then this should show up as more available memory in Resource Manager.
Taking you up on your Virtual Memory point: Firstly, as far as I'm concerned, turning of the VMM resulted in the first direct evidence that there was a problem in accessing real memory. So, never say never.
As you are probably aware, the VMM argument is a long-running one and has two sides. I think that it is not a coincidence that the pro-VMM side are also pro-Microsoft and the not-so-pro-VMM side are more critical of Microsoft.
We say things like, 'If Microsoft are so good then why do we need virus protection that can protect their software more efficiently than Window Defender can?' or 'Why do third party defragmenters performs so much better than Windows' one does' or, one of my favourites, 'If Microsoft are so good then why is there a constant flood of security and compatibility updates flowing from them?'
Look at Cubase, Sound Forge, A/V converters or re-coding tools... all of these do a huge amount of processing and what they all have in common is that they handle the system resources much more efficiently than Windows does.
This is not news though. Don't you remember the old joke about 'Microsoft Works' being an oxymoron?
In Windows 98, having over 768 MB of RAM enabled me to switch off VMM about fifteen years ago. This gave me a 40% increase in disk performance. I know, Windows 98, and XP wasn't much better and they do say that they had cracked it by Vista and its successors, but that was fifteen years ago and I haven't had it on since.
Don't get me wrong, I'm not saying that I didn't have problem - VMM being switched of was the reason that my Office Helper wouldn't start. It was funny how the software didn't rightly know why it wouldn't start, it just kept telling me to 'Reinstall Microsoft Office'.
My Encarta DVD wouldn't install either and again, the Microsoft software failed to realise why.
(In fact, funny aside, every time I ran Encarta, I would get an error and a box would pop up asking if I wanted to send a report to Microsoft. I kept pressing 'Don't Send' until out of desperation, I tried 'Send'.
Well, a message came up saying 'finding IP', then 'IP found', then 'sending report' and finally a message came up that said 'Report Sent' and the box closed.
The funny thing was, I had no modem on my desktop for political purposes and in any case, I had no access to any internet connection whatsoever.
Microsoft, huh?)
Okay, well it's true that serious benchmarking has shown that on average, one only achieves a 1.2% increase in performance as a maximum if the VMM is turned off but this doesn't tell the whole story.
We are thinking about it the wrong way round; it's not that 'switching off the VMM causes an increase in performance', it's that switching it on reduces the computer's performance. And that 'one percent reduction' is solely due to the fact that the VMM is enabled.
When it comes to benchmarking, identical scenarios are processed with one set of processes being performed with VMM enabled and another set with it disabled. Repeat a few times and look at the average.
The problem with this approach is that VMM doesn't have to work at all, the processes running while it is enabled can be run without it. If VMM is idling then it is not surprising that at doesn't load the system.
But just look at how the frame-count is decreased when VMM does have to do some work.
A better test would be to remove half the memory to ensure that a game like Crysis will run out of memory with VMM disabled and then benchmark the difference. The game will work with VMM enabled but look at the frame-count with just 2 GB of memory - it drops like a rock, doesn't it?
Not only that, while VMM is reading or writing the disk or memory, the application has to wait for VMM to complete.
It is ludicrous to deny that the VMM is a poor substitute for extra memory.
To put it another way, if two busy processes are sharing the same RAM which is managed by VMM are being alternately suspended/resumed, then how can there
not be a hit on performance.
And something else that amused me about one the benchmark reports was that although the author had conceded that there was a slight, in his opinion insignificant, performance hit by VMM, the 'peak limit readings' were about 80% with VMM off and 40% with it on and he stressed this.
However, what he seemed to have failed to realise is that when VMM is enabled, the amount of available memory is effectively doubled to include the swap file. This means that the peak limit is also doubled and 80% of 4 GB is about the same as 40% of 8 GB so no real story there after all. The VMM is doing nothing. Even the available memory figures stay roughly similar.
It turn out that if you have loads of memory, you simply do not need VMM. The apps and the technology can take care of it and Windows can keep out of it.
Unless you really need to reduce your performance slightly.
The tramp living at the bottom of my garden isn't doing any harm but I'd still like him to go away. He keeps eating all the bird-seed.
And to quickly finish up - in fifteen years I have never come across a BSoD that can be directly attributed to a disabled VMM. Like I said, I have had problems with software but these have been caused by bad coding of the software which is often from Microsoft and you would think that MS software coders would know that Encarta needed VMM enabled to run rather than just crashing and lying to me about going onto the internet.
It is the same with blue screens. They are the result of bad coding by programmers who have to rely on VMM to compensate for the fact that they can't manage their index tables properly.
It's a bad day for programming when we have to rely on VMM to ensure that drivers use the stack correctly. In other words, VMM sort of behaves as an error checker as well. And that take CPU cycles of time.
In the end, if an app causes a blue screen when VMM is off, blame the app.
Sorry if I went on.