Win 64 Bit And Win 32 Bit - What Is Your Experience?

[Beggar]

I just checked the performance in WinXP 32 bit and Win7 32 bit + Win7 64 bit. This with an Intel I7-860. From my experience WinXP and Win7 32 bit have pretty much the same performance. XP a little faster. But if I compare those results with Win7 64 bit I am surprised.

For instance:

WinRAR

in 32 bit OS 6 minutes 34 seconds

in 64 bit OS 7 minutes 56 seconds

Video rendering with Sony Vegas Pro

in 32 bit 2 minutes 40 seconds

in 64 bit 3 minutes 35 seconds

In both cases I used a 32 bit application for the 32 bit OS and a 64 bit application for the 64 bit OS. And of course the same files and the same settings.

Is there any software that is faster in Win 64?

[Benjie]

Sounds like there might have been other background tasks running.

I have Windows 7 64 bit and it's lightening fast for me, It just feels much tighter and snappier than 32 bit.

[FarangBuddha]

Very odd your results. The main benefit of using a 64bit version of Windows is faster performance. Especially with processor and memory intensive apps like audio and video encoding. Doubly unusual as you say the apps you tested were native 64bit versions.

Do you have at least 4gig of RAM?

I use WinVista Ultimate 64 bit with 4gig RAM...for everyday apps like web browsing and wp, I don't noticed any speed diff from XP 32 bit but with audio and video ripping, coversions, etc. there 64bit is definitely much faster.

Will get to test win Win7 in a couple weeks when I receive my free upgrade disks

[chrgrims]

i am using win 7, 64bit. definately not slower than 32bit. i was expecting a lot of compatibility trouble, but i havent had any yet.

[welo]

Checkout this article on Windows 7 32bit vs 64bit:

http://blog.tune-up.com/windows-insights/3...re-performance/

It shows that with 2 GB of Ram some applications run slower on 64bit Windows 7 compared to 32bit Windows 7. With 4GB 64bit usually outperforms 32bit. There is a WinRAR comparison, too.

This is due to the increased memory usage of Windows and Applications under 64bit (I think this is due to the increased minimum size of the smallest data unit (64bit compared to 32bit) but I might be wrong here.), which will cause 'shortages' when run with 2 GB only.

The main advantage of 64bit is not necessarily faster speed per se, but that the OS can address more than 4GB of RAM. With 32bit Windows the usable memory limit is somewhere at 3.5 GB (google for a detailed explanation).

The plain speed will improve only in data intensive applications that are optimized for 64bit since the CPU can work on larger data chunks at once.

Of course encoding tasks or compression tasks should benefit from 64bit.

So I assume that in your case the reason for slow 64bit performance is

• either a RAM shortage (do you have 2GB only?) or
  
• an otherwise not well configured system (driver issue?)
  

The difference in performance is rather extensive, much more than the any switch between 32bit and 64bit should cause.

[Beggar]

I have 4 GB RAM and the latest drivers. Apart from the drivers for my system and the test software it is a clean install - not so Win XP because this is the system I have been using for many years. But Win7 32 and Win7 64 are clean.

I made the tests with the same computer and exactly the same data (same disks). I have no explanation for my results. But if I look at this article that welo mentioned here I do not see much difference between 32 and 64 bit. In addition to this he writes there that 32 bit software runs slower. There is not much 64 bit software out there. Yes, you can use Firefox 64 bit !!! but then you will not have any flash videos anymore as far as I know.

But this test there is done with a pretty outdated CPU - a Core 2 Duo. My I7-860 has 4 cores and this with HT. WinRAR and Sony Vegas use both HT. So it is a completely different setup. The CPU ran when rendering at 32 bit with max 65 % and at 64 bit with max 80 %. The disks ran with a 2 disk RAID 0 setup - Seagate 7200.12 2 x 1 TB, MEM is a fast 7-7-7.

I tested several times with no background jobs running (not so in WinXP - there is too much running to be switched off). But I just can speak for my system and only for the tested software. Perhaps something is wrong with the coding there. If anybody knows settings to turn my results around please let me know. In need speed in Sony Vegas. Otherwise I will have to spend some more years with WinXP. And I was waiting so badly for the next system after Vista (what I skipped)...

Edited December 12, 2009 by Beggar

[trogers]

Reminds me of the early 80s, the 8-bit Apple vs the 16-bit IBM...

[dave_boo]

You realise that running the same programme (and doubly so for 64 bit native) on 64 bit requires more memory space right? And that added memory space could push it out of L1/L2/L3 cache and into the main RAM...which we all know would impact performance. Also the tasks you're looking at are heavy floating point operations which Intel's processors have problems with (compared to 32 bit mode) because they're lacking the instructions in 64 bit mode...

[Beggar]

The only advantage would then be more usable memory space if I understand you correctly but not performance. I see the results on my system and so far they don't give me much reason to render videos on a 64 bit system - at least with my system setup and Sony Vegas. For me it is disappointing. I bought a new computer and wanted to get a faster OS too. So in respect of performance there is no reason to leave WinXP 32.

[Beggar]

Reminds me of the early 80s, the 8-bit Apple vs the 16-bit IBM...

I think the winner was pretty clear in the end

[dave_boo]

The only advantage would then be more usable memory space if I understand you correctly but not performance. I see the results on my system and so far they don't give me much reason to render videos on a 64 bit system - at least with my system setup and Sony Vegas. For me it is disappointing. I bought a new computer and wanted to get a faster OS too. So in respect of performance there is no reason to leave WinXP 32.

Well; you don't expect all those new features that Vista/Win7 offers to be negative performance free do you? I would stick with XP whilst running the programmes and use Win7 for everything else because it is a much safer enviroment....

[dave_boo]

I think the winner was pretty clear in the end

Doesn't mean the best product won out<_<

[trogers]

I think the winner was pretty clear in the end

Doesn't mean the best product won out<_<

My 64-bit Acer laptop was first running on 32-bit XP, and recently I did a fresh install of Win 7 64-bit.

The latter is much faster, especially running with FF, Thunderbird, Comodo and Avira, all in 64bit.

[Beggar]

The latter is much faster, especially running with FF, Thunderbird, Comodo and Avira, all in 64bit.

How do you get flash videos in FF 64 bit? YouTube for instance? I could not find a way.

[trogers]

The latter is much faster, especially running with FF, Thunderbird, Comodo and Avira, all in 64bit.

How do you get flash videos in FF 64 bit? YouTube for instance? I could not find a way.

You can't with FF 64-bit. I have also installed FF 32bit for that purpose, but I seldom watch flash or youtube. Only use it for running broadband speedtests.

[minefield]

The latter is much faster, especially running with FF, Thunderbird, Comodo and Avira, all in 64bit.

How do you get flash videos in FF 64 bit? YouTube for instance? I could not find a way.

You can't with FF 64-bit. I have also installed FF 32bit for that purpose, but I seldom watch flash or youtube. Only use it for running broadband speedtests.

I have just upgraded to a Toshiba with 4 gig ram 64 bit and have doubled my speed over all , And no problems with utube, All I have to work out is how to the spell check working on thunderbird

[nygaardjan]

The latter is much faster, especially running with FF, Thunderbird, Comodo and Avira, all in 64bit.

How do you get flash videos in FF 64 bit? YouTube for instance? I could not find a way.

You can't with FF 64-bit. I have also installed FF 32bit for that purpose, but I seldom watch flash or youtube. Only use it for running broadband speedtests.

I have just upgraded to a Toshiba with 4 gig ram 64 bit and have doubled my speed over all , And no problems with utube, All I have to work out is how to the spell check working on thunderbird

I Just bought a ASUS notebook W7 64bits, Ye s, it is faster, but the Skype is not working too well. The webcam is not working. The shared HP Laser colour printer in my office does not have a driver for it, but my home printer also a HP has the 64B driver. All in all i am very satified with the W7 64Bit.

[Beggar]

And no problems with utube

Are you sure that you use a 64 bit browser or do you use the 32 version?

[welo]

It might be that many users who switch to Windows 7 experience a faster system not because of 64bit or the better OS, but because of the optimized performance a fresh install delivers, or because they upgraded the hardware at the same time.

Any speed comparison has to minimize impact of any other component involved but the one you want to test - here the OS - as best as possible. Otherwise the comparision is pretty much useless.

I further doubt that there will be any noticeable difference in applications such as web browsers or Office products. The common opinion (deduced from semi-tech articles on the internet) seems to be that applications might benefit mainly from two things:

• more RAM - which of course is not a direct performance gain, 64bit doesn't increase the efficiency how RAM is used (on the contrary it seems) it just allows the OS and applications to use more than 3GB RAM (efficiently).
  
• better performance for some applications that run intensive calculations on data (e.g. encryption software) and/or load really large files (>3GB) and which are optimized for 64bit architecture.
  

I think the wikipedia article really gives a good overview on a 32bit vs 64bit comparision (of course language is a bit 'techy'):

http://en.wikipedia.org/wiki/64bit#32_vs_64_bit

While 64-bit architectures indisputably make working with large data sets in applications such as digital video, scientific computing, and large databases easier, there has been considerable debate as to whether they or their 32-bit compatibility modes will be faster than comparably-priced 32-bit systems for other tasks. In x86-64 architecture (AMD64), the majority of the 32-bit operating systems and applications are able to run smoothly on the 64-bit hardware.

[...]

Speed is not the only factor to consider in a comparison of 32-bit and 64-bit processors. Applications such as multi-tasking, stress testing, and clustering [..] may be more suited to a 64-bit architecture when deployed appropriately. 64-bit clusters have been widely deployed in large organizations such as IBM, HP and Microsoft, for this reason.

[...]

The main disadvantage of 64-bit architectures is that relative to 32-bit architectures, the same data occupies more space in memory (due to swollen pointers and possibly other types and alignment padding). This increases the memory requirements of a given process and can have implications for efficient processor cache utilization.

So besides all this tech talk for most users it comes down to the point whether we 'ordinary users' benefit more from 64bit or 32bit Windows at the moment (and in the next few months).

I find your (Beggar's) results pretty irritating and disappointing - I just recently installed 64bit Windows in the belief that, if it doesn't give me any speed advantage (yet), it will not give me any disadvantage either. This seems to be wrong, and if even a rather data/calculation intensive test such as compression and encoding does not benefit from 64bit, what does??

A few more thoughts:

• I assume that harddrive performance is an important factor in your benchmark, can you run a benchmark on that alone and see if it compares on both systems? What is your setup in regard to harddrives?
  
• Can you monitor RAM usage during the test, maybe the increased memory usage on 64bit outweighs the advantage in your scenario
  
• Can you run the test on smaller files (for the same reason - to eliminate a 'problem' due to limited memory
  
• Can the result be possibly due to a difference in quad core usage on 64bit? Not that this would be less disturbing, but it would be at least an explanation...
  

welo

[Beggar]

I will check this out further with smaller files. Just give me a bit of time. But with smaller files the timeframe might be to short. The system always does something in the background and I might end up with random results then. But I will check and post here.

All disks are Seagate 7200.12. 1 TB. Rendering was done from two RAID 0 disks (source files) to a single Seagate in AHCI. All disk are on the Intel Matrix controller. The WinRAR tests were done just with source and target at RAID 0. To avoid defragmentation problems as far as possible created OUTPUT test files were always deleted before starting the next test. The disks were the same and not the system disk. If you know a good software to test without disks let me know. Drivers might have a very big impact on the results.

Perhaps I will check Avira someone mentioned here too. But I doubt this is a real 64 bit application. I think it is just a version that will work with 64 bit. I guess it is not 64 since it is listed in the folder Program Files (x86) and not in the Program Files folder where the 64 bit applications are. I installed the version that is called 64 bit and it ended up in C:\Program Files (x86)\Avira

At all my tests antivirus software was switched off of course - at this time not installed at all on the Win7 systems. Only on XP.

For me and MY system it will not make much difference if smaller files might be faster since I have large file quantities to render. The video in Sony Vegas was only 13 minutes. It was rendered from PAL DV-AVI stereo to PAL DVD format with MainConcept MPEG-2 (settings see picture). Normally I use space wasting constant bit-rate. I tested this too. But it does not make a difference in the gap between 32 and 64 bit.

Edited December 13, 2009 by Beggar

[Benjie]

Why don't you simply use a proper speed benchmarking program rather than trying to come up with your own test with so many factors out of your control.

[Beggar]

Why don't you simply use a proper speed benchmarking program rather than trying to come up with your own test with so many factors out of your control.

I do not want to benchmark my system. I want to find out for MY applications and MY system if it makes sense to switch to 64 bit. I need to know if I win or lose with exactly the time intensive applications I use.

But at this occasion I want to clarify - my post was not about 64 32 bit. It was about MY system, MY applications and 64 32 bit. I don't want to create the impression here that I can speak for all 64 bit systems and all applications!!! Just take all the different CPUs from AMD and INTEL. I tried to find out if I can improve my system settings to achieve better performance with 64 bit. And I guess if you do not use performance and time intensive applications then it will not make much difference if you use 64 or 32 bit - apart from software and driver problems because 32 has here an advantage. If something takes 0.001 seconds or 0.010 seconds is impossible to realize.

Edited December 13, 2009 by Beggar

[dave_boo]

I do not want to benchmark my system. I want to find out for MY applications and MY system if it makes sense to switch to 64 bit. I need to know if I win or lose with exactly the time intensive applications I use.

But at this occasion I want to clarify - my post was not about 64 32 bit. It was about MY system, MY applications and 64 32 bit. I don't want to create the impression here that I can speak for all 64 bit systems and all applications!!! Just take all the different CPUs from AMD and INTEL. I tried to find out if I can improve my system settings to achieve better performance with 64 bit. And I guess if you do not use performance and time intensive applications then it will not make much difference if you use 64 or 32 bit - apart from software and driver problems because 32 has here an advantage. If something takes 0.001 seconds or 0.010 seconds is impossible to realize.

And you're doing the correct thing. Benchmarks are just e-penis showings. If your systems is optimised for what you want then you've accomplished what you set out to do.

For instance I've compared my transcodes using handbrake on Linux against those done on Windows (both running nearly the same x264 svn) and Windows destroyed it. There was no reason for it. However I started investigating the matter and come to find out that it was the scheduler--and the x264 developer already knew and had submitted that bug to the kernel guys. Since 2.6.32 performance has improved some 50% on FPS. So I'm happy now...

[Beggar]

This with CPU cache etc. what you dave_boo said makes sense to me. But one thing does not get out of my head. The CPU cores never ran with 100 percent during my tests. There must be a bottleneck in my system. So I am thinking if I should change my 2 disk RAID 0 setup at least to a 3 disk RAID 0 setup. But I am not sure if this is the real problem. I have no empty disk to check and 3 TB in RAID would be too much anyway. What do you think where the problem could be?

Edited December 13, 2009 by Beggar

[welo]

This with CPU cache etc. what you dave_boo said makes sense to me.

Indeed, I thought the same. However, isn't that what a special 64bit version of a program should account for, optimize data chunks and algorithms to not have such effects on a 64bit system? For sure there a several layers of optimization, but if WinRAR performs THAT bad on 64bit compared to 32bit in general the programmers didn't do a good job. (Or the compiler, since I guess most of those optimizations are done automatically by the compiler).

I am still trying to find benchmarks and articles on the web to compare your results to.

But one thing does not get out of my head. The CPU cores never ran with 100 percent during my tests. There must be a bottleneck in my system. So I am thinking if I should change my 2 disk RAID 0 setup at least to a 3 disk RAID 0 setup. But I am not sure if this is the real problem. I have no empty disk to check and 3 TB in RAID would be too much anyway. What do you think where the problem could be?

From what I remember a process usually runs on one core only, and needs optimization to run on multiple cores. The speed increase on multiple core systems comes from distributing (different) processes on available cores. E.g. running system processes on one, and user applications on another, hence the performance increase.

A single process will therefore never have 100% CPU usage. I am not completely sure on this and a bit lazy to google about it Maybe dave_boo knows more on that topic?

However, there should not be a difference in regard to core management in Windows 7 between 32bit and 64bit.

My guess is still that your 64bit system is somehow crippled. And the best explanation I can find is that some essential component driver is causing the significantly worse results.

You could run any system benchmark suite on both Windows 7 versions and compare the result. Maybe this could give you a hint on what is the bottleneck.

I didn't use any benchmark suites since my teenager days when I overclocker my celeron 300Mhz to 450Mhz (yeah, I'm getting old ), but I'm sure there should be a free one out there. And you could start with the Windows 7 builtin performance test just to get an idea.

welo

[dave_boo]

This with CPU cache etc. what you dave_boo said makes sense to me. But one thing does not get out of my head. The CPU cores never ran with 100 percent during my tests. There must be a bottleneck in my system. So I am thinking if I should change my 2 disk RAID 0 setup at least to a 3 disk RAID 0 setup. But I am not sure if this is the real problem. I have no empty disk to check and 3 TB in RAID would be too much anyway. What do you think where the problem could be?

I seriously doubt that it's the speed that your disks are capable of feeding the processor. I'd imagine that you're looking at some 150 MB/s out of that array; that's some 20x the speed of the top bit rate for Blu-Ray. Think your processor could process 20 Blu-Ray streams at the same time; much less transcode them?

I have a feeling that more likely the problem is software based. While a 64 bit O/S can be a bit slower than the 32 bit version I'd look more towards the actual transcoding programme. Most likely it's the issue. I'm not hip to much on Windows, but I'd imagine there'd be some sort of programme that would have both 64 bit binaries and 32 bit for you. Unfortunately transcode is F/OSS only and Handbrake seems to only have 32 bit for Windows...otherwise I'd recommend you using one of those two.

I am a bit peturbed though that both the programmes you 'benchmarked' are giving you screwy results....and that actually makes me want to ask if you could disable the Turbo Boost in BIOS and do the tests on both O/S. You may also want to disable Hyper-threading (used to cause slowdowns!) and do a second run...

Indeed, I thought the same. However, isn't that what a special 64bit version of a program should account for, optimize data chunks and algorithms to not have such effects on a 64bit system? For sure there a several layers of optimization, but if WinRAR performs THAT bad on 64bit compared to 32bit in general the programmers didn't do a good job. (Or the compiler, since I guess most of those optimizations are done automatically by the compiler).

I am still trying to find benchmarks and articles on the web to compare your results to.

The problems you'll have with finding a benchmark is that few people look at transcoding....as I mentioned earlier you'd need to find a floating point heavy benchmark that probably wouldn't apply because the programming is different.

From what I remember a process usually runs on one core only, and needs optimization to run on multiple cores. The speed increase on multiple core systems comes from distributing (different) processes on available cores. E.g. running system processes on one, and user applications on another, hence the performance increase.

A single process will therefore never have 100% CPU usage. I am not completely sure on this and a bit lazy to google about it Maybe dave_boo knows more on that topic?

A process can spawn as many threads as it feels capable of handling; but otherwise you're fairly correct. Also you can assign (at least you used to be able to) processor affinity, which core gets which task, even in Windows. Without assigning the affinity the O/S should try and balance the load across the cores. In Linux running htop will even tell you how many threads the process has running.

However, there should not be a difference in regard to core management in Windows 7 between 32bit and 64bit.

No there shouldn't....but you are talking about Windows

My guess is still that your 64bit system is somehow crippled. And the best explanation I can find is that some essential component driver is causing the significantly worse results.

I had assumed that anyone who was smart enough to run Vegas Pro was smart enough to make sure they had the updated drivers, but good point....

You could run any system benchmark suite on both Windows 7 versions and compare the result. Maybe this could give you a hint on what is the bottleneck.

I didn't use any benchmark suites since my teenager days when I overclocker my celeron 300Mhz to 450Mhz (yeah, I'm getting old ), but I'm sure there should be a free one out there. And you could start with the Windows 7 builtin performance test just to get an idea.

welo

[welo]

I've been doing more reading on the topic (well, just the last 45 minutes) and found some interesting readings:

http://www.roylongbottom.org.uk/vista64.htm#anchorFP

Roy Longbottom's extensive benchmark website. He writes at a level that makes it hard for me to follow, since I'm application programmer and never did any scientific or caluclation intensive programming.

From what I understand the topic 32bit vs 64bit is much more complex than I thought when it comes to floating point arithmetics. dave_boo, you mentioned before that Intel's CPUs are not strong on floating point calculations, and this is repeated by Longbottom (64 bit compilations 'slow on Core 2 Duo').

It seems there are different aspects that will affect floating point performance in a specific benchmark/program, the main difference being which instruction set is being used (SSE vs SSE2 vs classic x86 FPU). Intel seemed to have had a major weakness on 64bit SSE2, which can be avoided by special compiler adjustments (?). Longbottom added an update to his benchmarks just this September eliminating Intel's major floating point bottleneck on 64bit.

Linpack Benchmark

The major surprise was that Core 2 Duo demonstrated particularly slow performance on some 64 bit compilations that produce SSE2 instructions, where the Athlon 64 could be up to twice as fast. These slow results were from the original 2006 versions but these were corrected using a later compiler in 2009. Below are results for the Linpack and Livermore Loops benchmarks

 <pre><b>Linpack Benchmark - Results in MFLOPS

							  64 Bit	   32 Bit	  Original</b>

  Core 2 Duo 2400 MHz, Vista	   823		 1480		 1315
  Core 2 Duo 2009 compilation	 1602

  Athlon 64 2210 MHz, XP x64	  1044		 1014		  838
  Athlon 64 2009 compilation	  1091
</pre>

SSE3DNow and MemSpeed

A later search found Intel Documentation, confirming that there are complications concerning use of the same register following movlpd, which can cause pipeline stalls. This was also corrected in the 2009 recompilation using the movsdx instruction.

<pre><b>Maximum Speeds in MFLOPS

			   Core 2 Duo 2400 MHz, Vista		Athlon 64 2210 MHz, XP x64

			 s=s+x[m]*y[m]   x[m]=x[m]+y[m]	 s=s+x[m]*y[m]   x[m]=x[m]+y[m] 
			   Dble   Sngl	 Dble   Sngl		Dble   Sngl	 Dble   Sngl
</b>
 <b>Assembled SIMD</b>   3166   6347	 2340   4692		1999   3998	 1011   2140

 <b>32 bit SISD</b>	  1053   1059	 1173   1147		 673	727	  970	878

 <b>64 bit SISD</b>	   761   1275	  398	943		 891	918	  808	735
 <b>2009 compile</b>	 1260   1270	 1172   1188		 976	978	  976	979

 <b>x87 FPU</b>		  1591   1593	 1180   1177		1047   1100	  853   1085</pre>

But Longbottom also states:

Other benchmarks using floating point generally show superior Core 2 Duo performance

Based on those findings I see two explanations for Beggar's benchmark results:

• A major software issue on his 64bit Windows setup (drivers)
  
• The tested 64bit applications rely on SSE2 instructions and don't incorporate a workaround for Intel's 64bit weakness
  

If I only had the equipment to re-run Beggar's benchmarks then we could eliminate the first point, but unfortunately it's difficult to setup something similiar on my Laptop.

welo

[2008bangkok]

If you have 4 gb ram then you should use 64 bit software simple, you get 1gb more memory to use as 32bit can only handle max 3gb ram even with the updates (vista sp) on 32bit then windows only shows that the system has 4gb ram but still is only uses 3gb ram..

So for that reason alone if you have 4gb ram then use 64bit

Edited December 14, 2009 by 2008bangkok

[Beggar]

First off all a lot of thanks to all of you for your explanations. I started with the idea the 64 bit might be always faster than 32 bit - at least if you use true 64 bit applications. In the meantime I have learned that I was somehow naive.

Of course all drivers and the BIOS are the latest for XP, Win7 32 and Win 64. But this does not mean much. New versions and new problems...

I will make the test that you mentioned with no HT and no Turbo Boost and others. Just give me time until Thursday since I have a visit here.

I will set up a ram disk and make tests from there even if then there is not much memory left. Swapfile I will block (set to 0) - it was blocked anyway during the other tests to avoid false data because of this.

If anybody has a free test I should run let me know - I still have here WinXP SP3, Win 7 32 and Win 7 64. But again be aware that this will benchmark my system and might be no guidance at all for others.

So welo for instance if you have a test that I should run let me know.

In respect of the RAID 0 output that dave_boo mentioned. It is higher than 150. See pictures. The Intel RAID 0 Volume is of course the RAID. The other disk is exactly the same Seagate with the same firmware without RAID. But the RAID picture is confusing too. The performance does not fall down much and so I am afraid I might be at the limit of the interface already and an additional disk might not bring much more. But I am not sure.

[Beggar]

If you have 4 gb ram then you should use 64 bit software simple, you get 1gb more memory to use as 32bit can only handle max 3gb ram even with the updates (vista sp) on 32bit then windows only shows that the system has 4gb ram but still is only uses 3gb ram..

So for that reason alone if you have 4gb ram then use 64bit

If you need a lot of memory you are completely right. There is no other option than 64 bit.

But 4 GB is a lot already for most applications - at least for mine. The difference between the 3.5 GB with 32 and 4 GB with 64 is not so big if you take into account that Win 64 will eat up more anyway for itself and for the 64 bit software.

I always run the swap file at 0 and so far no application complained. Not true - there is a game that asks for the swap file at startup even if it has enough memory available.

But just look at the graph in my picture. The first small upside is when Sony Vegas with the video project was started and the next one when Vegas started to render (this in WinXP and MANY other applications running). I does not use much memory - the total for everything what is running is not even 50 percent. So some GB more will not bring more I am afraid at least in my case.