For fuck's sake, let this BS die. MinWin was *NEVER* supposed to actually be part of any particular project. It's an internal project, and is certainly still ongoing. Certain offshoots of the MinWin project are and will continue to be used in future versions of Windows, but there was never a promise by anybody actually in a position to make such statements that Win7 would actually be based on MinWin.
Also, MinWin *IS* an evolution of the NT kernel - all that they're doing is removing certain portions which are either obsolete or redundant, and removing dependencies to make the whole thing more modular. It still has much of the standard API and the same codebase. It's not Singularity, it's not Midori, it's not CE or Mobile, it's not 9x, and it's not some completely new thing - it's simply NT slimmed down to the minimum requirements for the functional kernel.
As a side note, I'm not sure what new kernel-level features Win7 is planning to include that justify a bump of a full version number. It feels more like 2000 to XP - a more mature and more user-friendly OS with a few good new features and a lot of little improvements.
The main "must have" UI feature in Vista (for me) is the instant search. I literally can't stand to use XP anymore due to it lacking this - there are various installable search tools but none are as well integrated.
The primary *important* reason to upgrade is the security. Not only has Vista proven far more robust than previous versions of Windows (not immune, but no OS is), but it's finally possible to use NT like the multi-user OS it is, applying the principle of least privilege to user accounts without needing to do all manner of truly stupid shit with RunAs.
The best performance reason to upgrade would have to be SuperFetch - my programs load much faster, with little to no disk thrashing, on Vista then they did on XP.
As for "so-called Microsoft employee" I was a summer intern and am now back at school, but while there I worked in the Win7 group.
64-bit drivers have actually become far more common, to the point where anything on my system and all the peripherals I use even once a week have them - I'm sure there's still some stuff you'd like to use which doesn't, though.
I suppose the driver signing restriction is sort of DRM, but not what I think of when somebody refers to DRM. In any case, I believe you can get around this by importing your own root certificate into the certificate store, then signing all your drivers with that cert (or one derived from it). I can't say I've tried it, though.
Full access to RAM: you need to enable PAE. The reason it is disabled by default (on Windows and on every Linux distro I've tried - over a dozen by now - in the default kernel) is that many 32-bit drivers assume that they're operating in a 32-bit address space, and die in unpleasant ways when there are over 4GB of addressable RAM.
To enable PAE, look up bcdedit (command lone tool for setting Vista's boot parameters, equivalent to editing grub.conf to change the init or kernel parameters on Linux). You don't need a new kernel, though I think the home editions won't allow PAE. Out of curiosity, if you've got so much RAM, why not run 64 bit? PAE is a godawful hack anyhow, and if you need to run 16-bit apps there's DosBox or virtualization...
What DRM have you run into? I can't say I've hit that one at all, and I've used Vista (including listening to and ripping music, watching movies, and playing games) for over 2 years now.
Well, since code signing is part of Trusted Computing, I can assure you that part is still present. As for *media* DRM, I can't say - but I've used Vista for years, as a gamer and as somebody who likes music and movies, and I've had no DRM-releated issues in the least.
Peter Gutmann's article, which I'm guessing you've read and based the above opinion on, was full of crock. It was blatantly obvious when he wrote it that he had never even tried to do his research properly - some of the stuff he described as outright impossible due to DRM worked just fine (unified video drivers for different GPU models, for example), and other things he claimed would happen never did (all audio and video getting downgraded just because you're playing a.mp3 through a non-protected path). He's revised it a few times, removing some of the more patently false BS, but it still reads like BS anyhow.
To reiterate my above point: I've had NO issues stemming from DRM on my system. I don't have Blu-Ray or anything REALLY badly DRMed, but XP won't play those anyhow. The key issue is that everything I tried to do in XP also works in Vista.
Forging the MS code-signing signature would mean determining their secret key, which even if they use the older key length which is being phased out right now would probably take thousands of CPU-years.
Modifying the binary on disk won't work either. The hash would change, which would make the signature invalid. (In fact, forget UAC; Windows will complain bitterly if you modify its system files.)
Exploiting a signed binary would work, but between DEP, ASLR (both of which are always enabled for system binaries), and intense manual and automatic code review, that's been very difficult and rare in Vista.
I was a summer intern. I was implying that XP is also basically just a refined 2000; aside from the look and feel, it's a remarkably similar OS overall. In particular, the biggest differences that come to mind at XP's release time were the fast user switching and system restore (there were others, of course, but it's hard to remember much else that was very new and exciting).
We (the team I was on) were running Win7 on most of our machines, including production boxes, by the end of my internship. I won't claim it's ready to ship yet, but it's easily within a year. It certainly may change in several significant ways before release - there was a substantial (if behind-the-scenes) feature cut while I was there - but for the most part it's already usable and entering the heavy bug-fixing stage, rather than still in the feature development stage.
I wish I could say I know of a really good reason. My best guess is that it's a SKU differentiator, which isn't a particularly good reason from a consumer standpoint but does mean that MS can justify the high cost (from a home user standpoint) of their "premium" editions (enterprise/ultimate) by pointing at the significantly cheaper home editions. I'd ask the same thing about BitLocker, which is also only available in the premium SKUs.
It's also worth pointing out that Interix doesn't include a X server. You can install one, either in Interix (though it's not free) or in Win32 (I use xming, a native port using mingw). It includes the client libraries and headers, so you can compile and run graphical apps, but you'd need to have (and run) X. On the other hand, as of Vista you can write binaries which use both.so and.dll libraries, so you could probably use the native windowing tools.
On the other hand, it would make Linux/Windows incompatibility easier. I've written code for a Linux device (not even running on x86) that I developed in Interix, and there were only a couple of lines that needed #ifdef for which platform. The makefile and build process were identical.
Hell, the outcry is probably attracting new players. There are probably t least a few people who wouldn't really have looked at the game, certainly not at its pre-release screenshots, and now have - and it's a sure thing that at elast some of them (myself included) liked what they saw.
It's also worth noting that turning down the texture filtering, lowering the gamma on your monitor, and playing at 800x600 (bonus points if it's on a CRT that stratches and pixelates things) would go a long way toward re-creating that circa 2000 graphical feel...
One thing oddly absent from the screenshots: the Windows build number, visible in the lower right corner of the desktop (for pre-release builds) or the winver program.
It's definitely a Windows 7 build no more than a couple months old, though.
Just a note: XP Professional and Vista Enterprise or Ultimate can run a NT subsystem for POSIX, including a fairly complete Unix-like OS called Interix. On XP, look for the "Services For Unix" (SFU) downloads, on Vista it's called "Subsystem for Unix Applications" (SUA).
Although bash isn't included in Interix by default, it's downloadable for free, either manually or via command-line package manager, from http://www.suacommunity.com/ (along with many other tools, including perl, ssh/sshd, svn, and the full GNU build toolchain, to name the ones I use most often). You can run Win32 programs from within an Interix shell as well, so I actually use bash as my primary Windows CLI shell these days.
Having worked on the Win7 team, I'd say Vista to Win7 felt more like the difference between 2000 and XP. There are a couple new big features (Win7 has multitouch support, BitLocker has been dramatically improved, etc.), a variety of UI tweaks and tricks (the new theme picker, the modified system tray, and more of that sort), and some mostly-behind-the-scenes changes (faster bootup and hibernation on multicore machines, UAC by default now elevates without prompting for Microsoft-signed executables, and a few others).
It *is* an improvement, but could arguably be described as a refined and matured version of Vista, with a couple new features. It's a bigger change, especially from the user perspective, than XP RTM to XP SP2, but much smaller than XP SP2 to Vista.
A) Vista uses, on a clean install, about 3 to 4 times the base resources of XP (more CPU while indexing, but I've had no trouble with it doing that when I'm trying to do something else). Considering it's had 3 iterations of Moore's Law to play with, that's not bad overall. Personally, the only time I've had bad experiences with Vista systems was trying to run an OEM copy, or trying to run it with 512MB or less (non-shared) system RAM. The OEMs install so much crap as to make the system WAY slower than it really is, and while Vista will run on 512MB RAM (without Aero), it runs about like XP on its "minimum requirement" of 128MB.
B) UAC prompts much less by default (though it's highly configurable - look for the Local Security Policy settings in Vista). This is the equivalent of going through and making a bunch of stuff in the OS setuid root - convenient but dangerous. External programs (installers, or anything which must run as Admin) will still generate prompts.
C) What security do you want increased? They spend years fixing security bugs in the NT codebase (specifically XP and early Longhorn/Vista) and you (Slashdot in general) complain about how long it's taking to get the new OS. They provide an easy way to run as standard user without the pain that doing so in XP caused, and you call it "annoying". They add ASLR, and you complain that performance has decreased by a few percent.
It sounds like they're tapping into the signals that would normally be sent to the muscles (not to the motor nerves themselves, but the last stage prior to them). In a computer analogy, this would be like reading signals between the filesystem driver and the physical device driver - all the "filtering" of what you would actually say has probably already been done. Similarly, this wouldn't catch fleeting thoughts which you would never vocalize. On the other hand, it quite possibly *would* catch thoughts which you would normally say only under your breath or when the mic is off. There's still plenty of potential for embarassment...
MicroB s a fantastic browser, but the lack of XUL has a few major downsides: the context menu isn't extension-customizable, most extensions need significant re-writing to work at all, and customization in general is worse than it should be.
Additionally, I'm pretty sure it uses the older version of Gecko that can't even pass Acid2. I don't know how hard it would be to fix that, though, and for all I know they did in Diablo, but I don't have a n800 at hand to test with.
On the flip side - it's a browser that runs on a 400MHz ARM Linux system, runs Flash 9 (pay attention, Jobs), can play Windows Media and Real streams (probably Quicktime as well), and has a variety of extensions including AdBlock Plus (fantastic for smaller screens), Flashblock (very nice on a low RAM/low CPU handheld), and others. It's not really tabbed, but given the screen real-estate needed for a tab bar and the hardware button for switching between windows, it doesn't need to be.
500x the absorption does seem quite impossible. However, it's possible that what was actually meant is "500 times less energy is not absorbed" which is a number that theoretically could be as large as you like.
If you assume a cell is 10% efficient (they're actually rather above that now) then its loss percentage is 90%. Reducing that by a factor of 500 would mean losses of only 0.18%, or an absorption efficiency of 99.82% (which is vastly in excess of anything we have today).
While it's still an astoundingly large number, it at least avoids breaking the laws of math and physics, and requires only a single, simple misinterpretation by a journalist who quite possibly never took any college-level math or physics. I know people who make that kind of mistake all the time. Hell, I've done it myself on occasion.
Um WTF? I agreed with you RIGHT up until you started talking about things like NT4. NT 4 may have been good in its day, for hardware that could run it reasonably, but by modern standards it is neither fixed nor stable. It also is lacking a lot of APIs and system calls which means that modern software often just won't run on it.
Out of curiosity, have you tried Allegiance? It's primarily an online cockpit-view (though there are other camera views) space dogfighting game. It also has some fun strategic elements. Joystick is supported and by far the best way (I used a Force Feedback Pro, which was fantastic).
The game was originally a commercial product out of Microsoft Research, but didn't really go anywhere. MSR open-sourced the game, and it is now community supported, with the client available for free download.
As a suggestion for a game where ramming matters (and which in some ways is similar to how Jumpgate looks, but with real cockpit view) try Allegiance. It was a commercial game out of MS Research that didn't really go anywhere, so they open sourced it (no joke) and it is now community supported. Great joystick support, fun gameplay, an interesting mix of dogfighting and strategic play. 2000-era graphics will run fine on any system, though they look a little boxy.
I used Photoshop since it's a common example, but in my personal case the issue is gaming. The premium graphics EVE client has over 1GB of data files plus active session data, and while that's OK with me (2GB of physical RAM) alt-tabbing between EVE (after a lot of non-stop play) and a long-open browser session can take quite a while indeed - 5 seconds easily, and I have my pagefile split across 2 physical drives.
In short, there are times when it is OK to consume all available RAM (gaming, where performance is king) and times when I really wish I could tell the browser to trim down its damn working set (just because it will get swapped out doesn't mean it doesn't have an effect - and I've got a fast connection, so caching is less of a benefit to me than it might be for others).
I'll look for that setting in Opera. I doubt it'll be enough for me to actually switch my primary browser, but it's certainly interesting.
In short, it's a problem because the browser has no way of knowing whether you want it to use that 1GB for caching, or you'd rather it release all that so you can alt-tab between it and Photoshop without spending 10 seconds thrashing the page file every time. In other words, while the software *might* be able to tell how much physical RAM is available, it has no way of knowing how much of that it is OK to take.
Some programs have the ability to configure how much memory you want the program to allow itself. This sounds like a pretty cool idea to me, but I'm not aware of any way to do it in any mainstream browser.
IE8b2 will hit 21/100 if you give it a sec to get past the really long test at around 12 or so. It doesn't actually matter whether you allow MSXML or not, either.
Chrome is definitely faster at Javascript and better at Acid3, but IE isn't *quite* as bad as you're saying. It is also improving - 21 isn't much but it's better than the previous version.
Actually, since fork(2) must clone the application memory space, I beleive it can under some circumstances take longer than CreateProcess. That said, the fundamental point is correct; provided the per-process allocation is kept small, only one copy of the static data need reside in physical memory.
32-bit Vista (or XP) does not, by default, have PAE enabled. This is due to the "drivers sometimes flake out" issue you described (and inaccurately ascribed to x64 machines - x64 drivers shouldn't care at all whether their memory space is within the first 4GB or not).
However, with PAE enabled (it's fairly easy to do; on XP edit boot.ini and on Vista use bcdedit.exe from an Admin command prompt, then reboot) it is indeed possible for a 32-bit copy of Windows to see more than 4GB of RAM. That is, after all, the entire point of PAE... although you are correct that the individual applications will still only have 2 (or 3) GB of mappable memory space (note that the space isn't actually allocated; most of it will never be requested at all).
PAE has been supported since, IIRC, the Pentium Pro of roughly 10 years ago. This doesn't mean it is good to run, merely that the hardware is capable. Whiel some cheap boards may have lacked that support, it should be available in any modern board. Of course, it won't be long now before high-end machines need more than the 36-bit address space (64GB) PAE provides -which won't be a problem, because everything other than ultra-low-power and legacy systems will be 64-bit by then.
Run, not install (MSXML is already installed on the system, but default settings in IE7 and up don't run most ActiveX without permission). It will actually continue just fine if you don't allow it to run MSXML, though there's a test which occurs at about that time that takes 4 or 5 seconds to get through (on my machine) - which gives the impression that the page has finished. Wait a bit and it should reach 21, with or without the ActiveX control.
Slashdot Mirror
For fuck's sake, let this BS die. MinWin was *NEVER* supposed to actually be part of any particular project. It's an internal project, and is certainly still ongoing. Certain offshoots of the MinWin project are and will continue to be used in future versions of Windows, but there was never a promise by anybody actually in a position to make such statements that Win7 would actually be based on MinWin.
Also, MinWin *IS* an evolution of the NT kernel - all that they're doing is removing certain portions which are either obsolete or redundant, and removing dependencies to make the whole thing more modular. It still has much of the standard API and the same codebase. It's not Singularity, it's not Midori, it's not CE or Mobile, it's not 9x, and it's not some completely new thing - it's simply NT slimmed down to the minimum requirements for the functional kernel.
As a side note, I'm not sure what new kernel-level features Win7 is planning to include that justify a bump of a full version number. It feels more like 2000 to XP - a more mature and more user-friendly OS with a few good new features and a lot of little improvements.
The main "must have" UI feature in Vista (for me) is the instant search. I literally can't stand to use XP anymore due to it lacking this - there are various installable search tools but none are as well integrated.
The primary *important* reason to upgrade is the security. Not only has Vista proven far more robust than previous versions of Windows (not immune, but no OS is), but it's finally possible to use NT like the multi-user OS it is, applying the principle of least privilege to user accounts without needing to do all manner of truly stupid shit with RunAs.
The best performance reason to upgrade would have to be SuperFetch - my programs load much faster, with little to no disk thrashing, on Vista then they did on XP.
As for "so-called Microsoft employee" I was a summer intern and am now back at school, but while there I worked in the Win7 group.
Ah!
64-bit drivers have actually become far more common, to the point where anything on my system and all the peripherals I use even once a week have them - I'm sure there's still some stuff you'd like to use which doesn't, though.
I suppose the driver signing restriction is sort of DRM, but not what I think of when somebody refers to DRM. In any case, I believe you can get around this by importing your own root certificate into the certificate store, then signing all your drivers with that cert (or one derived from it). I can't say I've tried it, though.
Full access to RAM: you need to enable PAE. The reason it is disabled by default (on Windows and on every Linux distro I've tried - over a dozen by now - in the default kernel) is that many 32-bit drivers assume that they're operating in a 32-bit address space, and die in unpleasant ways when there are over 4GB of addressable RAM.
To enable PAE, look up bcdedit (command lone tool for setting Vista's boot parameters, equivalent to editing grub.conf to change the init or kernel parameters on Linux). You don't need a new kernel, though I think the home editions won't allow PAE. Out of curiosity, if you've got so much RAM, why not run 64 bit? PAE is a godawful hack anyhow, and if you need to run 16-bit apps there's DosBox or virtualization...
What DRM have you run into? I can't say I've hit that one at all, and I've used Vista (including listening to and ripping music, watching movies, and playing games) for over 2 years now.
Well, since code signing is part of Trusted Computing, I can assure you that part is still present. As for *media* DRM, I can't say - but I've used Vista for years, as a gamer and as somebody who likes music and movies, and I've had no DRM-releated issues in the least.
Peter Gutmann's article, which I'm guessing you've read and based the above opinion on, was full of crock. It was blatantly obvious when he wrote it that he had never even tried to do his research properly - some of the stuff he described as outright impossible due to DRM worked just fine (unified video drivers for different GPU models, for example), and other things he claimed would happen never did (all audio and video getting downgraded just because you're playing a .mp3 through a non-protected path). He's revised it a few times, removing some of the more patently false BS, but it still reads like BS anyhow.
To reiterate my above point: I've had NO issues stemming from DRM on my system. I don't have Blu-Ray or anything REALLY badly DRMed, but XP won't play those anyhow. The key issue is that everything I tried to do in XP also works in Vista.
Forging the MS code-signing signature would mean determining their secret key, which even if they use the older key length which is being phased out right now would probably take thousands of CPU-years.
Modifying the binary on disk won't work either. The hash would change, which would make the signature invalid. (In fact, forget UAC; Windows will complain bitterly if you modify its system files.)
Exploiting a signed binary would work, but between DEP, ASLR (both of which are always enabled for system binaries), and intense manual and automatic code review, that's been very difficult and rare in Vista.
I was a summer intern. I was implying that XP is also basically just a refined 2000; aside from the look and feel, it's a remarkably similar OS overall. In particular, the biggest differences that come to mind at XP's release time were the fast user switching and system restore (there were others, of course, but it's hard to remember much else that was very new and exciting).
We (the team I was on) were running Win7 on most of our machines, including production boxes, by the end of my internship. I won't claim it's ready to ship yet, but it's easily within a year. It certainly may change in several significant ways before release - there was a substantial (if behind-the-scenes) feature cut while I was there - but for the most part it's already usable and entering the heavy bug-fixing stage, rather than still in the feature development stage.
I wish I could say I know of a really good reason. My best guess is that it's a SKU differentiator, which isn't a particularly good reason from a consumer standpoint but does mean that MS can justify the high cost (from a home user standpoint) of their "premium" editions (enterprise/ultimate) by pointing at the significantly cheaper home editions. I'd ask the same thing about BitLocker, which is also only available in the premium SKUs.
It's also worth pointing out that Interix doesn't include a X server. You can install one, either in Interix (though it's not free) or in Win32 (I use xming, a native port using mingw). It includes the client libraries and headers, so you can compile and run graphical apps, but you'd need to have (and run) X. On the other hand, as of Vista you can write binaries which use both .so and .dll libraries, so you could probably use the native windowing tools.
On the other hand, it would make Linux/Windows incompatibility easier. I've written code for a Linux device (not even running on x86) that I developed in Interix, and there were only a couple of lines that needed #ifdef for which platform. The makefile and build process were identical.
Hell, the outcry is probably attracting new players. There are probably t least a few people who wouldn't really have looked at the game, certainly not at its pre-release screenshots, and now have - and it's a sure thing that at elast some of them (myself included) liked what they saw.
It's also worth noting that turning down the texture filtering, lowering the gamma on your monitor, and playing at 800x600 (bonus points if it's on a CRT that stratches and pixelates things) would go a long way toward re-creating that circa 2000 graphical feel...
One thing oddly absent from the screenshots: the Windows build number, visible in the lower right corner of the desktop (for pre-release builds) or the winver program.
It's definitely a Windows 7 build no more than a couple months old, though.
Just a note: XP Professional and Vista Enterprise or Ultimate can run a NT subsystem for POSIX, including a fairly complete Unix-like OS called Interix. On XP, look for the "Services For Unix" (SFU) downloads, on Vista it's called "Subsystem for Unix Applications" (SUA).
Although bash isn't included in Interix by default, it's downloadable for free, either manually or via command-line package manager, from http://www.suacommunity.com/ (along with many other tools, including perl, ssh/sshd, svn, and the full GNU build toolchain, to name the ones I use most often). You can run Win32 programs from within an Interix shell as well, so I actually use bash as my primary Windows CLI shell these days.
Having worked on the Win7 team, I'd say Vista to Win7 felt more like the difference between 2000 and XP. There are a couple new big features (Win7 has multitouch support, BitLocker has been dramatically improved, etc.), a variety of UI tweaks and tricks (the new theme picker, the modified system tray, and more of that sort), and some mostly-behind-the-scenes changes (faster bootup and hibernation on multicore machines, UAC by default now elevates without prompting for Microsoft-signed executables, and a few others).
It *is* an improvement, but could arguably be described as a refined and matured version of Vista, with a couple new features. It's a bigger change, especially from the user perspective, than XP RTM to XP SP2, but much smaller than XP SP2 to Vista.
A) Vista uses, on a clean install, about 3 to 4 times the base resources of XP (more CPU while indexing, but I've had no trouble with it doing that when I'm trying to do something else). Considering it's had 3 iterations of Moore's Law to play with, that's not bad overall. Personally, the only time I've had bad experiences with Vista systems was trying to run an OEM copy, or trying to run it with 512MB or less (non-shared) system RAM. The OEMs install so much crap as to make the system WAY slower than it really is, and while Vista will run on 512MB RAM (without Aero), it runs about like XP on its "minimum requirement" of 128MB.
B) UAC prompts much less by default (though it's highly configurable - look for the Local Security Policy settings in Vista). This is the equivalent of going through and making a bunch of stuff in the OS setuid root - convenient but dangerous. External programs (installers, or anything which must run as Admin) will still generate prompts.
C) What security do you want increased? They spend years fixing security bugs in the NT codebase (specifically XP and early Longhorn/Vista) and you (Slashdot in general) complain about how long it's taking to get the new OS. They provide an easy way to run as standard user without the pain that doing so in XP caused, and you call it "annoying". They add ASLR, and you complain that performance has decreased by a few percent.
It sounds like they're tapping into the signals that would normally be sent to the muscles (not to the motor nerves themselves, but the last stage prior to them). In a computer analogy, this would be like reading signals between the filesystem driver and the physical device driver - all the "filtering" of what you would actually say has probably already been done. Similarly, this wouldn't catch fleeting thoughts which you would never vocalize. On the other hand, it quite possibly *would* catch thoughts which you would normally say only under your breath or when the mic is off. There's still plenty of potential for embarassment...
MicroB s a fantastic browser, but the lack of XUL has a few major downsides: the context menu isn't extension-customizable, most extensions need significant re-writing to work at all, and customization in general is worse than it should be.
Additionally, I'm pretty sure it uses the older version of Gecko that can't even pass Acid2. I don't know how hard it would be to fix that, though, and for all I know they did in Diablo, but I don't have a n800 at hand to test with.
On the flip side - it's a browser that runs on a 400MHz ARM Linux system, runs Flash 9 (pay attention, Jobs), can play Windows Media and Real streams (probably Quicktime as well), and has a variety of extensions including AdBlock Plus (fantastic for smaller screens), Flashblock (very nice on a low RAM/low CPU handheld), and others. It's not really tabbed, but given the screen real-estate needed for a tab bar and the hardware button for switching between windows, it doesn't need to be.
500x the absorption does seem quite impossible. However, it's possible that what was actually meant is "500 times less energy is not absorbed" which is a number that theoretically could be as large as you like.
If you assume a cell is 10% efficient (they're actually rather above that now) then its loss percentage is 90%. Reducing that by a factor of 500 would mean losses of only 0.18%, or an absorption efficiency of 99.82% (which is vastly in excess of anything we have today).
While it's still an astoundingly large number, it at least avoids breaking the laws of math and physics, and requires only a single, simple misinterpretation by a journalist who quite possibly never took any college-level math or physics. I know people who make that kind of mistake all the time. Hell, I've done it myself on occasion.
Um WTF? I agreed with you RIGHT up until you started talking about things like NT4. NT 4 may have been good in its day, for hardware that could run it reasonably, but by modern standards it is neither fixed nor stable. It also is lacking a lot of APIs and system calls which means that modern software often just won't run on it.
Out of curiosity, have you tried Allegiance? It's primarily an online cockpit-view (though there are other camera views) space dogfighting game. It also has some fun strategic elements. Joystick is supported and by far the best way (I used a Force Feedback Pro, which was fantastic).
The game was originally a commercial product out of Microsoft Research, but didn't really go anywhere. MSR open-sourced the game, and it is now community supported, with the client available for free download.
http://freeallegiance.org/
As a suggestion for a game where ramming matters (and which in some ways is similar to how Jumpgate looks, but with real cockpit view) try Allegiance. It was a commercial game out of MS Research that didn't really go anywhere, so they open sourced it (no joke) and it is now community supported. Great joystick support, fun gameplay, an interesting mix of dogfighting and strategic play. 2000-era graphics will run fine on any system, though they look a little boxy.
http://freeallegiance.org/
Also, ramming is a big part of EVE. It doesn't do damage but it sure knocks stuff around...
I used Photoshop since it's a common example, but in my personal case the issue is gaming. The premium graphics EVE client has over 1GB of data files plus active session data, and while that's OK with me (2GB of physical RAM) alt-tabbing between EVE (after a lot of non-stop play) and a long-open browser session can take quite a while indeed - 5 seconds easily, and I have my pagefile split across 2 physical drives.
In short, there are times when it is OK to consume all available RAM (gaming, where performance is king) and times when I really wish I could tell the browser to trim down its damn working set (just because it will get swapped out doesn't mean it doesn't have an effect - and I've got a fast connection, so caching is less of a benefit to me than it might be for others).
I'll look for that setting in Opera. I doubt it'll be enough for me to actually switch my primary browser, but it's certainly interesting.
In short, it's a problem because the browser has no way of knowing whether you want it to use that 1GB for caching, or you'd rather it release all that so you can alt-tab between it and Photoshop without spending 10 seconds thrashing the page file every time. In other words, while the software *might* be able to tell how much physical RAM is available, it has no way of knowing how much of that it is OK to take.
Some programs have the ability to configure how much memory you want the program to allow itself. This sounds like a pretty cool idea to me, but I'm not aware of any way to do it in any mainstream browser.
IE8b2 will hit 21/100 if you give it a sec to get past the really long test at around 12 or so. It doesn't actually matter whether you allow MSXML or not, either.
Chrome is definitely faster at Javascript and better at Acid3, but IE isn't *quite* as bad as you're saying. It is also improving - 21 isn't much but it's better than the previous version.
Actually, since fork(2) must clone the application memory space, I beleive it can under some circumstances take longer than CreateProcess. That said, the fundamental point is correct; provided the per-process allocation is kept small, only one copy of the static data need reside in physical memory.
32-bit Vista (or XP) does not, by default, have PAE enabled. This is due to the "drivers sometimes flake out" issue you described (and inaccurately ascribed to x64 machines - x64 drivers shouldn't care at all whether their memory space is within the first 4GB or not).
However, with PAE enabled (it's fairly easy to do; on XP edit boot.ini and on Vista use bcdedit.exe from an Admin command prompt, then reboot) it is indeed possible for a 32-bit copy of Windows to see more than 4GB of RAM. That is, after all, the entire point of PAE... although you are correct that the individual applications will still only have 2 (or 3) GB of mappable memory space (note that the space isn't actually allocated; most of it will never be requested at all).
PAE has been supported since, IIRC, the Pentium Pro of roughly 10 years ago. This doesn't mean it is good to run, merely that the hardware is capable. Whiel some cheap boards may have lacked that support, it should be available in any modern board. Of course, it won't be long now before high-end machines need more than the 36-bit address space (64GB) PAE provides -which won't be a problem, because everything other than ultra-low-power and legacy systems will be 64-bit by then.
Run, not install (MSXML is already installed on the system, but default settings in IE7 and up don't run most ActiveX without permission). It will actually continue just fine if you don't allow it to run MSXML, though there's a test which occurs at about that time that takes 4 or 5 seconds to get through (on my machine) - which gives the impression that the page has finished. Wait a bit and it should reach 21, with or without the ActiveX control.