Giving Vista what it deserves...

[DGhost]

I have not heard anything regarding NVidia, but I believe OS control over VRAM is a fundamental requirement in a WDDM class driver.

The biggest sticking points for DX10 being backported would come from how DX10 interacts with WDDM drivers. WDDM is a fair amount more robust than WDM, and it also enables quite a bit of system interaction with the graphics card. I am not "up" on DX10 as much as I would like to be, so I am unsure how much it actually needs these features.

It could be simply that backporting DX10 would involve a performance penalty for some of the more advanced features due to complexities of coding around WDM limitations. It could also be that traditional WDM drivers are not robust enough to handle advanced features in a reliable manner. I dunno.

DWM (Aero) was originally designed around DX9. Somewhere in there they came up with a compelling enough argument to redesign the way graphics drivers interact with Windows and set graphics driver optimization back literally years. Maybe DX10 takes advantage of these changes, maybe not.

[Technoid]

edit: and before it comes up, i would love to hear how my computers, that are painfully outdated and less powerful than your coworkers, have no problems running Vista...

Now we are hitting semantics

Define "problems",
was it having problems running or booting - no,
was it slow as shit consuming all his 2gb ram and swapping like crazy - yes!

And while you may perceive the vista "benefits" as a price well payed for what you think is sufficient speed and responsiveness others may not.

[DGhost]

Seee... that makes absolutely no sense...

First off, Vista will naturally use all available ram. Linux does the same thing. XP does the same thing. This is by design. It makes no sense to have 2GB sitting there with half of it empty and unused when you could be using it to make the OS run faster. So it maintains a cache of recently/frequently accessed files/programs/whatever in ram so it doesn't have to hit disk to load. The difference between XP/Linux and Vista in this case is that Vista maintains that cache between reboots and shutdowns. It can even adjust it based on previous usage scenarios so if you are more likely to run Outlook between 8am and 5pm but run Crysis between 5pm and 9pm, it can actually taylor the cache between those times to make your apps run faster. Another difference between it and XP/Vista is that if you allocate 100% of your memory and force all data out of cache, it actually can repopulate that cache afterwards because it, again, keeps track of it. But the underlying principle of what it is doing is really not fundamentally different than any OS.

This is one of those things that the "prosumers" got horribly wrong. They didn't know how to read the damn task manager. All they do is see "Free: 0" and assume all kinds of shit. For instance, right now I have approx 1.2GB of ram being used for system cache. XP does the same damn thing if you leave it on long enough.

Second up... on a C2D system... even if that WERE swapping... which I *highly* doubt... it would not slow the system down, at all. Why? The exact same reasons why disk access has a lot less of a penalty on any other multiproc NT system for as far back as NT4. Because while one proc is handling the I/O request and waiting for the response, the other proc is handling active threads. Whereas when you have a single processor system you wind up with a thread that effectively hangs the system waiting on that request.

Third... I don't think it was swapping. Indexing the file system, maybe. Repopulating the above mentioned cache, maybe. Lots of different things it could be doing. But, more than likely it would not be adversely affecting system performance unless *gasp* the drivers were immature. Which a lot of them were. Especially if it was running the x64 edition, which you have not actually clarified. x64 Vista was horrid for the first 6 months it was released. The drivers were soooo terrible it was a joke. I've been hearing good things about it recently though and I'm probably going to install it again tonight to see how well it is doing these days.

Again, this goes back to the point I've been making this whole time. Immature drivers have the biggest single performance impact. Second to that, hotfixes released in the last 12 months have done a very good job at fixing a bulk of the problems...

[Nowhere]

I believe Technoid talks more about the little thing that Vista takes much more RAM just to run (nevermind caching)

...
Second up... on a C2D system... even if that WERE swapping... which I *highly* doubt... it would not slow the system down, at all. Why? The exact same reasons why disk access has a lot less of a penalty on any other multiproc NT system for as far back as NT4. Because while one proc is handling the I/O request and waiting for the response, the other proc is handling active threads. Whereas when you have a single processor system you wind up with a thread that effectively hangs the system waiting on that request.
...

Uhm, no, that's not the biggest problem with swapping. The biggest problem is that active, interactive thread doesn't have the data it needs and interactiveness of this thread vanishes. Or, alternatively, swapping competes with HDD i/o of what you're doing.

Again, this goes back to the point I've been making this whole time. Immature drivers have the biggest single performance impact. Second to that, hotfixes released in the last 12 months have done a very good job at fixing a bulk of the problems...

Hmm...you might be onto something here. Just after the realese of Vista friends of mine left me their new laptop to set it up for them. In the process I also played with Vista a bit. But I haven't installed any 3rd party drivers. Everything was perfectly fine on 512 MB (minus what went to integrated GFX), also when playing UT99 (but testing stopped there, I didn't install anything else). OTOH few people told me that Vista tends to accumulate cruft that slows it down after few months...

[DGhost]

I believe Technoid talks more about the little thing that Vista takes much more RAM just to run (nevermind caching)

Vista, out of box, does not consume 2GB of ram.

Uhm, no, that's not the biggest problem with swapping. The biggest problem is that active, interactive thread doesn't have the data it needs and interactiveness of this thread vanishes. Or, alternatively, swapping competes with HDD i/o of what you're doing.

generally speaking, the active interactive thread is loaded in memory and does not need to swap. The system pushes inactive pages of memory into swap first so as to provide that free ram. with one exception - if you have a single thread demanding more memory than the system has available. this is slowly becoming less and less the case as system memory sizes increase. in that situation, then yes, you are correct.

background swapping affects foreground apps because it literally stalls the system while waiting for the i/o requests to complete on a single processor system. on a multiproc system, it can do this on a seperate proc and still leave a single proc address interactive applications.

swapping and disk i/o at the same time is a rough mix anyways. thread interactivity for whatever is doing the disk i/o is already shot unless it's multithreaded, and then you're gonna have a very large impact on the interactivity because again, swap requests have priority and stall the system. again, this can be alieviated by a multiproc system.

[Elie]

Vista, out of box, does not consume 2GB of ram.

But it does consume 1GB of ram on idle, on my t60 thinkpad!

I don't think an OS should consume that much. The problem here is Vista (the 32bit version) supports up to 4GB of ram the last gig will be reserved by windows so you are left with three, so if an OS hogs 1GB for internal processes etc. that really leave you with 2GB to spare (providing you do have 4 GB's in the system) enter applications like photoshop, Premiere pro, After Effects and 3D Max etc. and now you are in a bind with your HD swapping like mad.

I can make XP idle at 250MB's, perfect because I have more ram available to me, I think Micrososf's thinking is for home users that do nothing but browse the internet and use Office or play games. Vista should not be used for serious video/graphics or high end applications that demand tons of memory. Vista 64bit may be an option in this case, again you have to worry about driver and application support even more.

Cheers,
Elie

[Nowhere]

Vista, out of box, does not consume 2GB of ram.

I'm perfectly aware of that, as evidenced by my story about Vista running fine on less than 512MB. However, due to bigger memory consumption of Vista, it's entirelly possible that memory hungry app that was running fine on 2GB & XP, swaps on Vista... (plus I seem to remember some articles discussing bigger memory consumption of games under Vista, don't remember speculated reasons)

generally speaking, the active interactive thread is loaded in memory and does not need to swap. The system pushes inactive pages of memory into swap first so as to provide that free ram. with one exception - if you have a single thread demanding more memory than the system has available. this is slowly becoming less and less the case as system memory sizes increase. in that situation, then yes, you are correct.
...

I meant specifically this situation.

...
background swapping affects foreground apps because it literally stalls the system while waiting for the i/o requests to complete on a single processor system. on a multiproc system, it can do this on a seperate proc and still leave a single proc address interactive applications.
...

(I'm not sure that's the case. It stalls the app making the request, competes with other disk acceses, but doesn't stall the whole system. )

...
swapping and disk i/o at the same time is a rough mix anyways. thread interactivity for whatever is doing the disk i/o is already shot unless it's multithreaded, and then you're gonna have a very large impact on the interactivity because again, swap requests have priority and stall the system. again, this can be alieviated by a multiproc system.

And this situation.

[Technoid]

The problem here is Vista (the 32bit version) supports up to 4GB of ram the last gig will be reserved by windows so you are left with three,

That is NOT true!

since vista 32 only supports 4gb memspace, any memory on your vidcard, raid card, soundcard etc et will effect your total reported system ram!

if you have 4gb ram and a vidcard with 512mb windows will report 3.5gb ram.

Notice that I'll keep ram'ing this point through untill it sticks!

[Elie]

Interesting, is that only on Nvidia video cards, since they use memory sharing architecture or will that effect everything?

[Technoid]

every kind of board that has memory.

[DGhost]

There are actually two different issues at play there...

first is the application independant Virtual Address Space, which is 4GB on a 32bit OS. Windows actually reserves 2GB of that for the system and for kernel mode. So specifically, no application can address more than 2GB of ram at a time UNLESS you enable the /3GB switch, which basically adjusts that so the each process can address 3GB of the 4GB limit, and 1GB is reserved for the system. This limitation is in place even if you enable PAE. In order to work around it, you have to use the AWE API in order to access memory instead of the normal system calls. AWE and PAE are two seperate things. You can use one without the other. If you enable PAE, it automatically enables 3GB addressing.

Second issue is then the fact that, for whatever reason, Large Memory video cards in Vista eat up address space, as described by Technoid. Current belief is that because of the way that WDDM operates it forces Video Memory to be mapped inside the 4GB Virtual Address Space inside the kernel reserved space, and hence counts against total available memory.

OK. This leads into the demand for 64bit memory addressing. Just running a 32bit app on a 64bit OS does not remove these barriers. You need to run a 64bit clean application on a 64bit os to work around this. Because of the technical limitations involved, it is generally recommended that if you are using an application that requires this much memory that you go with a 64bit version. Period.

OOOOOKkkkk.

Nowhere - yes, paging to disk stalls the system on a uniproc PC. Why, you might ask? because the system thread responsible for paging memory cannot 1) write to cache, 2) return control to the system until the data is written. In a multiproc PC, this would simply result in a delay for the thread that is requesting it. In a uniproc PC, it results in a tiny delay and a loss in interactivity as the system is unable to execute any other thread until the memory is moved. When you are doing large transfers of memory, it results in a cumulative effect over a period of time, as it is a multitasking OS and does not move all the memory at once. it moves it in pages or series of pages and lets other applications execute in between them.

the reason for games under vista consuming more memory comes from the fact that they are designed around DX9 and running under DX10. As DX9 did not have any sort of video memory management features, you had to maintain your assests locally in memory and copy them to video ram as nessicary. Under DX10 you can give everything to DX10 and let it figure out how and when it needs to be where. The situation that resulted is that assets were getting duplicated in system memory as a result. There is a hotfix that specifically addresses this.

As far as applications pushing over 2GB in space... what you don't really understand is that Windows itself will start pushing data out into the page file as the active process starts allocating it. On a multiprocessor PC, this can happen in the background with very little hit to interactivity on the part of the process doing . On a uniproc PC, you are simply screwed.

Now, if an application is trying to read from disk and windows is trying to page to disk, then you will see a slowdown on the part of that... however, if it is a multithreaded app (as described) and you are running on a multiproc system, you will not see a loss of interactivity, just an increase in how long it takes to load files.

GETTING BACK TO THE ORIGINAL POINT.

A Core 2 Duo (2 CORES!!!!) with 2GB has no real reason to either 1) page to disk, or 2) suffer a horrid loss of interactivity while paging to disk UNLESS you are using specific applications or usage scenarios. That is a drastically different situation than "Vista is SOOOO SLOOOOOOWWW".

that was my whole point.

[Umfriend]

Still, Vista is way way slower when running heavy apps than XP is when running a single smaal app. Gotcha!

[Technoid]

Well, the vidcard memory thing is not because of vista drivers 'cause the same effect happens in xp.

[DGhost]

did it? a lot of the reports i heard was this issue was specific to Vista. I mean, it has to map into the VAS anyways and counts against ram either way, but the specifically reporting that the memory isn't there at all..

[Technoid]

Probbaly because few feels the need to use 4gb with xp, but vista on the otherhand

I have personnally seen the problem on xp, since we have a "memmory crazy" customer