appiah4 wrote:

To put things in perspective, back in 2006 Pentium Northwood and Athlon 64 were king. Pentium D and Athlon 64 X2 were the new hotness and Vista kind of sucked at leveraging multiple cores.

The last Northwood P4s were made in early 2004, and by 2006 they were already dethroned. 2006 was well into Pentium D / Athlon 64 X2 (and even early Core 2 Duos), although undoubtedly many people were running systems 2-3 years old which would still have the single-core P4/Athlons.

appiah4 wrote:

Today, we take 3d acceleration for the desktop for granted.

Even in 2006, when Vista was introduced, we took 3D acceleration, even DX9-level, for granted.
If you didn't have DX9-level hardware, you simply couldn't enable Aero in the first place, so you either had it, and it was accelerated, and it was fast enough, or you were stuck with non-Aero.

If anything, you could argue that Vista didn't accelerate the GDI drawing, and you were more dependent on the CPU there (although as mentioned before, you did get the advantage of being rid of the dirty rectangles approach, so it had to redraw a whole lot less).
But I covered that already, where Windows 7 boasted the return of accelerated GDI and a multithreaded implementation. In practice however, you couldn't tell the difference.

appiah4 wrote:

and Vista kind of sucked at leveraging multiple cores.

It kind of didn't at all. Even XP is quite reasonable with multiple cores, but Vista made another step in the right direction.
But I covered that already, also explaining IO prioritizing and such.
Obvious troll is obvious.
Some facts to back it up: http://www.infoworld.com/article/2630178/micr … ch-faster-.html

These results suggests that when considering Windows 7, performance should be viewed as a reasonable justification for upgrading from Windows XP, but not a driver for migration from Vista.
...
The flat performance results against Vista are reasonable given that, as we noted earlier, Windows 7 is based on the Vista kernel.

appiah4 wrote:

and Intel IGPs were EVERYWHERE and they sucked at EVERYTHING.

They didn't, actually.
The graphics operations that Vista does for the GUI are so simple and basic, that even Intel IGPs handle them just fine.
The slowest machine I bought in the Vista era was a 1.5 GHz Core2 laptop with an Intel 965M chipset. I couldn't tell the difference in GUI performance between that machine and my desktop with a GeForce 8800.

appiah4 wrote:

It also didn't help that Vista's aero did not really set it aside from XP's candybar shell significantly enough aside from a color palette swap

You must be blind, because if to you the only difference from XP and Vista Aero is a color palette swap, you're clearly visually challenged.
Vista and Windows 7 use basically the same 'glass' look for their windows, with semi-transparent borders and such. Which is completely different from XP, which is more of a new skin on the classic Windows UI widgets.
Vista looks far more similar to 7 than it does to XP.
Vista and 7 mainly differ in the default colour scheme, and some changes to the start menu and taskbar. The overall drawing style and underlying technology is identical.

Well, no, we did not take DX9 level acceleration for granted in 2006. In 2006, reasonable DX9 performance was finally affordable, but not even close to widely adopted. Let's check the Steam survey from 2006. This is a survey made on gamers, so it is representative of a minority of the marken that is keeping up to date with technology. On August 2006, prior to the marketshare of DirectX 9 model 2 and 3 code path combined was only around 60%, with 25% on DX8 class hardware and 15% still on DX7 class hardware. An astounding 97% of gamers used single core CPUs in 2006.

https://web.archive.org/web/20060825052346/ht … tus/survey.html

To put things into perspective, Crysis wasn't released yet and the "But can it run Crysis" meme did not exist in 2006, that is what it was like for gamers. This was even worse for the non-gamer crowd. Let me give you an idea of what the average laptop in 2006 looked like by the end of November 2006, post Vista launch: https://web.archive.org/web/20061205090404/ht … TF8&node=565108

We are looking at a plethora of Celeron laptops with Intel IGPs (ew), or if you were luck the most basic ATI Xpress or GeForce Go IGPs that while being DX9 compliant, were only compliant enough in terms of checkbox features, and performed like dogs in any accelerated task. Even the better laptops sucked at dealing with the Aero. As a user I distinctly remember people around me having to switch to Aero Basic to make it workable. Whatever you profess to have experienced - millions of people experienced this and you are the only one carrying Vista's banner as a performance champion, I suppose everyone was wrong.

And no, I'm not challenged in any way. Windows 7 is obviously an evolution of Windows Vista and I did not deny that (I actually remarked that they are pretty similar and I prefer Vista aero in many ways), but Vista resembles XP in many ways and that has a lot to do with how the whole taskbar area, Start menu layout and several other small things work like XP. To demonstrate, Just switch to Vista Basic and look at what it becomes with Glass turned off. A good many people experienced Vista this way, by the way.

Anyway, I have no idea what gets you so worked up and personal about a piece of software, but for the sake of carrying a sensible conversation I tried to respond civilly. Honestly, I have neither the time nor energy to troll people on a forum, so if you think I am doing that please just block me and do us both a favor. I really don't feel like getting involved in personal bickering.

appiah4 wrote:

On August 2006, prior to the marketshare of DirectX 9 model 2 and 3 code path combined was only around 60%

Hint: 60% is not a minority.
You can also argue that if you're not up-to-date with your hardware, you're not the target audience to run the newly-released Vista either.
So this Steam survey doesn't say all that much.
Only a small minority would upgrade an existing Windows OS, the majority that used Vista would get it preinstalled on a newly bought PC, which would indeed have a DX9 GPU as standard, or better (this was also the time that DX10 was introduced, with the GeForce 8x00 series. In fact, the aforementioned Intel 965M is DX10-capable, and I bought that in 2007).

appiah4 wrote:

To put things into perspective, Crysis wasn't released yet and the "But can it run Crysis" meme did not exist in 2006, that is what it was like for gamers.

But don't forget, Vista and DX10 were what created Crysis. We weren't exactly in early DX9-land anymore.

appiah4 wrote:

To demonstrate, Just switch to Vista Basic and look at what it becomes with Glass turned off. A good many people experienced Vista this way, by the way.

Still mostly the same as Glass, just with some of the shader effects turned off. Quite different from XP still.

appiah4 wrote:

Honestly, I have neither the time nor energy to troll people on a forum, so if you think I am doing that please just block me and do us both a favor.

You're either deliberately spreading all sorts of misinformation and distortions of the truth (as in: trolling), or you're really that deluded.
Also, it's not that I care so much about Vista. I just can't stand people who spread nonsense about whatever topic. The topic just happens to be Vista here.

This is definitely not a fair, apples to apples comparison, but around the same time, Compiz for example was offering similar capabilities with less performance penalty (or less hardware requirement). Thus, the questions arise, why does Aero require more performance capability (assuming the underlying platforms are not wildly different in efficiency - this assumption may be incorrect), could a version of Aero be implemented such that older generation (PS/VS 1.x) or lesser performance hardware could use it, and does Aero offer additional capabilities over other, less demanding compositors?

Ultimately, this gets back to my initial frustration with the steady ramping of Windows' performance footprint. Are consumers really getting value from this, or is it simply Wirth's law at work?

gdjacobs wrote:

why does Aero require more performance capability (assuming the underlying platforms are not wildly different in efficiency - this assumption may be incorrect)

Exactly, Compiz is pretty damn slow and inefficient. Requirements here are quite arbitrary of course. Because what does the software developer consider 'good enough' performance?

gdjacobs wrote:

could a version of Aero be implemented such that older generation (PS/VS 1.x) or lesser performance hardware could use it

In theory they could. In practice, the world before SM2.x was far too haphazard to standardize a GUI on it and expect it to work and appear the same everywhere. So the choice of SM2.x as a bottom line makes a lot of sense if you look deeper into the specifications of the DX9/SM2.x standard. Things like minimum precision for pixel operations, a strict texel:pixel mapping etc. Early hardware just 'rendered whatever sorta looks okay'.
SM2.x isn't all about performance.
In fact, you can find tons of SM1.x hardware that is faster than Intel IGPs that are SM2.x-capable and can run Aero without a problem. The IGPs can run Aero, not because they meet some performance requirements (performance really isn't an issue, as I said, as graphics workloads go, Aero is completely unintimidating for even the slowest of DX9 hardware. You're just rendering a single desktop, not an entire game world with millions of polys), but because they meet minimum requirements for hardware capabilities and precision.

gdjacobs wrote:

and does Aero offer additional capabilities over other, less demanding compositors?

As I tried to explain above, Aero is not just a 'compositor', it is a completely different way of rendering windows, ditching the age-old tree-of-dirty-rectangles approach. Things like 'compositor' and 'Compiz' give me the dirty taste of linux in my mouth, and that does not apply here in the least.

gdjacobs wrote:

Ultimately, this gets back to my initial frustration with the steady ramping of Windows' performance footprint. Are consumers really getting value from this, or is it simply Wirth's law at work?

As I already said before, at the very least they get nicely synced, tear-free, flicker-free and 'garbage'-free updates of their display.
I think it was well worth it.
Also, as already said, Aero is actually *faster* than classic mode, if you use capable hardware. That's what hardware acceleration is about.
Of course that requires some people to break through their cognitive dissonance first, and reject the idea that 'more eyecandy==slower', or whatever falsehood it is that they are clinging to.

Scali wrote:

That only makes it slower if your machine doesn't have the disk space and memory to cope with that. As long as you're in the 'sa […]
Show full quote

95DosBox wrote:

I think what made it slow was all that extra bloat. It was about 10 times the size of XP installed.

That only makes it slower if your machine doesn't have the disk space and memory to cope with that. As long as you're in the 'safe zone', you don't notice.
The same goes for XP... It was much slower than its predecessors on machines with less than say 512 MB. But once you had 512 MB or more, it ran fine.

95DosBox wrote:

Sounds like I should spend some time using Aero but the Classic interface is basically the snappiest (2nd) in XP and Vista.

Not sure what you mean by 'snappiest', but Aero is simply faster with things like rendering text and drawing controls etc.
If you mean that the fancy animations in Aero feel like you're being slowed down, yea I suppose. But I guess like with XP, you can tweak those to run faster or just disable them altogether.
What I like best about Aero is that everything is v-synced and double-buffered: You get no 'tearing' or other garbage when moving windows around. Everything moves around smoothly and without drawing artifacts.
In classic mode you actually erase the screen if you drag a window over it, and you can see it redrawing.

The visual tearing issue might be true compared to earlier versions. I haven't used Vista as a constant OS to evaluate that aspect. But if you used Vista in its default look/theme with Aero which I assume that is your preferred setting and compare it to XP reverted to Windows Classic Mode with Quick Launch enabled with Clear Desktop icon to the far left next to the Start Button, Group Tasks removed so all windows have their own space rather than lumped together adding more delays and have all Visual effects set to best performance meaning removed it beats them from my experience. And I am referring to using XP, Vista, or W7 on a quad core not a P4 or earlier era which is more noticeable. I might see if I can tweak Vista the way I want and add Aero on top if it doesn't remove the Quick Launch and all my preferences. I think Aero was basically added to mimic a Mac OS at the time look to draw in more users from that crowd. Graphics usage was more taxing compared to today's 2GB-4GB graphic cards which probably barely make them wink at Vista today but then it might have been 32MB on the passive low end cards and 512MB on the high end enthusiast cards. I think another issue that steered me away from Vista aside from no Intel USB 3.0 xHCI driver support was the Blu-ray software player I had wouldn't function in it but it would in XP and W7. When Vista and W7 are similar enough and W7 had more functionality it was hard to consider Vista as a secondary OS but anymore except for experimentation. That might change as soon as the USB xHCI drivers get ported over and DX12 gets patched in it will breath new life into Vista.

Another key issue that made me disliked the later Windows was I noticed extra lag time dealing with 1000 or more video files in the same folder. If you rename one file it takes like a minute to refresh the entire folder after. Do you encounter this issue on your Vista/W7/W10? This folder refresh lag doesn't occur in XP at all and you can continue renaming files right away and don't lose your position of where you were in the folder view. Even renaming a file it does not include the entire filename plus extension highlighted so you have to add another step in highlighting the filename extension portion.

I prefer using a 128GB or less primary hard drive to partition the OS as things are much easier to setup and format for legacy support an no needs for any software patches for large capacity drives or MBR / GPT OS boot issues. So to keep things lean is a bonus and not deal with patching things that work fine if you know the limits. A few years it would have been XP 32-bit and Vista 64-bit but after tasting how fast USB 3.0s are when dealing with editing HD Videos I now changed my mind and today I would probably just use XP and W7 on the main drive. Vista and W10 could be placed on a secondary hard drive for experimentation on a 1TB. There are also issues with disk cloning in DOS when you exceed 1TB but everything else just works great keeping to this configuration. If I need really large capacity drives they are hooked up externally via USB up to 16TB each. If I need more simultaneous space I can add some USB cards with 7 ports and get another 16x7TB = 112TB per USB card.

dr_st wrote:

Windows 7 did bring some optimizations for increased UI responsiveness. At least Microsoft talked at length about it. […]
Show full quote

Scali wrote:

Windows 7 was no better, but nobody complained about that. And even today, as we see in this thread for example, people complain about Vista for being resource-hungry, but Windows 7 gets a free pass

Windows 7 did bring some optimizations for increased UI responsiveness. At least Microsoft talked at length about it.

Scali wrote:

I've always found it quite ironic that people make it sound like Vista is one of the worst versions of Windows ever, and Windows 7 is one of the best.
In reality, Vista and Windows 7 are very close together, and you're not too far from the truth if you say that Windows 7 is 'Vista SP3'.

People in general have no patience/skill/interest to research and understand things. They stick by their first impressions (or more likely, collective first impressions of others). Read my Daikatana vs Quake II write-up if you want another example.

Scali wrote:

Until recently, there was very little that Vista couldn't do, which Windows 7 can. Today, things are skewed somewhat because Vista is EOL, and most vendors no longer bother to support their software and hardware on Vista anymore. Windows 7 will reach that point in the near future as well

Not quite so. Win7 is supported until early 2020, and given its popularity it is more likely that vendors will try to support it for as long as possible, versus as little as possible as was with Vista.

Furthermore, because of the technology it got, Win7 will forever be able to run IE11, Office 2013/2016, Visual Studio 2012/2015, and everything derived from it. That's half a decade of SW stack support more than Vista has (latest versions that run on Vista are 2010).

I think this is has become more evident recently for myself. At one point in the early days I dissed Vista due to the sluggish performance and the preactivated UAC nag window that wanted you to authorize each little thing I normally did which wasn't questioned in XP. Later I turned UAC off permanently. But until these HD Videos and a need to help build a W10 system for a friend I started playing around with Vista and W7 again to see their actual performance with video editing and USB 3.0 speeds to a Ramdrive. Boy was there a huge difference. I used to think from actual testing that USB 3.0 speeds were only double that of USB 2.0 so I basically scoffed at it as being unnecessary. Now after a month of video editing HD videos the wait time is tremendously cut down that sometimes it feels like all that extra wait time I used to feel in XP and Vista was unnecessary. If you aren't dealing with massive amounts of data XP and Vista are great the way they are and solid systems. But Vista has become Windows 2000, the unsupported OS. This is what made me move over from Windows 2000 back in the day because programs keep insisting XP requirement and later SP1, SP2, and SP3 became a requirement so it was inevitable. This is where Vista 64-bit now sits as a Windows 2000 32-bit equivalent now gathering more dust. Like I said earlier until USB 3.0 xHCI for the Intel USB is possible, throw in W7 SP1 compatibility, add DX12 there is no chance in hell Vista can survive any longer or resurrect itself from being current and useful. The thing that made me flip back to Vista after giving it a bad rap earlier was the entire Windows Classic Quick Launch user interface. But given how much time I now save editing HD videos recently I can take missing out on that. I just shift the entire taskbar to the far left and Show Desktop is in the bottom left corner mimicking its earlier position close enough. Quick Launch I'll have to rehack back later because the Pin to Taskbar alternative is just inferior to me.

Scali wrote:

appiah4 wrote:

On August 2006, prior to the marketshare of DirectX 9 model 2 and 3 code path combined was only around 60%

appiah4 wrote:

To put things into perspective, Crysis wasn't released yet and the "But can it run Crysis" meme did not exist in 2006, that is what it was like for gamers.

But don't forget, Vista and DX10 were what created Crysis. We weren't exactly in early DX9-land anymore.

I had a friend who had an old laptop that came preinstalled with Vista 64-bit. Had 4GB of Ram (2x2GB). He complained it was constantly overheating and I did some research and found a slightly higher performing CPU with more cache and used slightly less wattage. Go figure but it was hard locating compatible laptop CPUs so everything is a gamble finding the right compatible socket type. Well since I liked to tinker with things and you know how laptops usually came preinstalled with bloatedware on top of the OS? I decided to create an XP, Vista, W7 MultiOS clean installation setup of each with their own partition just to do a performance comparison of Crysis 1 on all three OSs.

Here's what I found.

XP only used DX9 despite fake DX10 hacks.
Vista was DX10
W7 was DX10

I noticed for XP the 4GB was sufficient enough for the game and the FPS was higher than Vista and W7.
Vista on 4GB was a bit sluggish but I used SP2 and DX11 Patch to make it equate W7 as close as possible.
W7 was DX10 only because the nVidia graphics integrated into the laptop didn't support DX11. The FPS were lower than the Vista SP2 DX11 patched.

Later I did another test when I managed to buy expensive 4GB DDR2 memory for it. I maxed it out to 8GB and wanted to see if having more than 4GB made a difference for Crysis 1.

It indeed had a huge significance in the load time of the levels and probably leveled out some in game lag.

I would say from the results in FPS just from memory it benchmarked liked this on this particular laptop.
XP 32-bit SP3, Vista SP2 DX11, W7 SP1.

I'll probably have to do another true benchmark to gather and jot down the FPS and all those numbers but that would only pertain to that laptop. Given it's limited memory 512MB video memory and if I were to redo such a test I would do it on my quad core on a GTX 750 or a 960 with 1GB/2Gb video ram configurations and compare those three OS and their corresponding frame rates at the same resolution.

The only thing I would alter for two different experiments would be downgrading Crysis 64-bit for Vista and W7 to only use DX9 mode so it should be faster and more balanced when comparing to XP. I will also be running the game off a pure Ramdrive just to eliminate any chance of HDD or SSD interfering with lag.

But all in all during my tests I got really good at the 1st level of the game it was hilarious as this wasn't usually my type of game but when you repeatedly playing the same level over and over and over and finding news ways to kill the enemy in interesting ways that probably haven't ever been done as most people moved onto to the next level to finish the game. After all that testing I starting enjoying the game and customizing my keyboard / mouse config for optimal kill efficiency.

appiah4 wrote:

And no, I'm not challenged in any way. Windows 7 is obviously an evolution of Windows Vista and I did not deny that (I actually remarked that they are pretty similar and I prefer Vista aero in many ways), but Vista resembles XP in many ways and that has a lot to do with how the whole taskbar area, Start menu layout and several other small things work like XP. To demonstrate, Just switch to Vista Basic and look at what it becomes with Glass turned off. A good many people experienced Vista this way, by the way.

Any time they did a change in user interface, DOS to Win 1-3X, 9X/ME, NT4+. Microsoft always tried to make it different. Adding different names to the same locations as previous versions making it more confusing. Adding more steps to get to where it was previously done in less steps. But altering the main user interface desktop look they should not have done from Vista+ or at least kept it as an alternative user interface theme like previous versions. If Windows 7 had nothing special going for it, no USB 3.0 xHCI support, and the same level of gaming compatibility as Vista requirements Windows 7 wouldn't have really had a huge advantage it does today. If all programs made for W7 worked on Vista including DX11.1 support it would have been a case of which had the best user interface. Vista 64-bit would have won in my book if that happened. Maybe if they had called Windows 7 - Windows You or Windows U because a lot of their advertising claim was they had listened to all the Windows users and added the best suggestions into it on top of Vista. I think if they had a blind study Vista vs Windows 7 most might have picked Vista today. The Win You or U name would probably have sunk them worse than Win ME. Win 7 had a nice ring to it and with great marketing the rest is history.

Well, I certainly can't blame you about the X Window System. it's certainly being used in places where it's not really appropriate anymore. In many cases, I think most of the original Xlib stuff is being skipped for either XRender or straight GL. So, once another form of modesetting, surface rendering, and IPC has been standardized, X11 (at least for desktop use) will be able to RIP.

I personally never noticed any performance issues with Compiz, even with my PS 1.1/VS 1.4 Radeon 9100 on an Athlon XP. It does have issues such as releasing the GPU fully when backgrounded, but this is most likely due to it's experimental and non mainstream nature. It was essentially a demonstration of what could be done but by no means a finished product, kind of like Looking Glass from Sun. Pretty much tear free, though. 😀

Thanks for clearing up some of the advantages of SM2. If I can summarize my understanding, it offered a more robust and regular platform for MS to develop Aero for. Is that correct?

I certainly agree that (higher and lower complexity) eyecandy will be faster if the graphical manipulations can be performed on the GPU vs the CPU. However, considering everything performed on a GPU, are there not costs in terms of performance and responsiveness with more desktop graphical effects or features vs less? Theoretically, is there a way to implement a tear free display on a lesser performance budget than was done with Aero? Is it necessarily harder to do?

gdjacobs wrote:

Thanks for clearing up some of the advantages of SM2. If I can summarize my understanding, it offered a more robust and regular platform for MS to develop Aero for. Is that correct?

Yes. You have to draw the line somewhere. And you have to think lowest common denominator. I mean, sure, if you pick some of the best-of-breed cards from the DX8 or even DX7-era, you'll find that they already fulfill many, if not all the requirements you'd need for consistent rendering of a 2D GUI.
But what about the worst? NV and ATi were miles ahead of the lesser gods in those days. Intel had notoriously bad precision, and then there were other small players like ImgTech, SGI/XGI/Trident/VIA etc.
Some of these cards were absolutely horrible. Not to mention other legacy stuff like textures that can only be power-of-two sizes.

Rendering 2D is actually more 'difficult' than 3D in a way... That is, if you are doing realtime 3D, then being a pixel off here or there in some frames, it isn't that big of a deal. But if you have such issues when rendering (antialiased) text, or when you want to do the actual compositing, things can get quite annoying. You may get all sorts of artifacts.
When you want proper 2D rendering, you want to be sure that every pixel you render is in exactly the right place. With DX9/SM2.x, the specs simply demand that the hardware rasterizes and textures points, lines and triangles in a specific way: https://msdn.microsoft.com/en-us/librar ... s.85).aspx
So by definition, all hardware will render the same.
With older hardware, it's a tossup. NV and ATi generally already implemented these particular rasterization rules. Other vendors may or may not (there was no formal requirement for DX rasterization prior to version 9).

Remember when ClearType was introduced, and you had the issue that some monitors had pixels ordered RGB, others BGR, which never mattered, until then?
Well, Aero is a pretty similar thing.

gdjacobs wrote:

I certainly agree that (higher and lower complexity) eyecandy will be faster if the graphical manipulations can be performed on the GPU vs the CPU. However, considering everything performed on a GPU, are there not costs in terms of performance and responsiveness with more desktop graphical effects or features vs less?

That depends. GPUs are deeply pipelined architectures, and a lot of operations are 'free' because you are bottlenecked by the memory bandwidth.
You have to understand that GPUs are designed to render complex 3d scenes, which means tons of polygons, on which you would normally apply multiple textures (colormap, normalmap, perhaps some precalced lightsourcing/reflection etc), which you combine together with some shader function.
So you generally have architectures that are optimized to fetch from a lot of textures simulataneously, and render to a single target buffer.

A 2D GUI is far simpler: each window can be constructed of just 2 triangles, so the entire screen is only a handful of polygons at most.
Likewise, there's only one texture being applied when 'blitting' to screen: the window contents, which have been previously rendered to a texture (which doesn't have to be updated all the time, only when something actually changes inside that window).

So that basically means that you will usually be bottlenecked by the raster operator units that store the pixels in the target buffer, and you are only using a single texture fetch unit.
Adding the 'glass' effect simply fetches from a second texture, and applies a simple shader for the 'glass' effect. You can do this 'for free', because you have tons of idle units on the GPU anyway, even on a simple Intel IGP.

gdjacobs wrote:

Theoretically, is there a way to implement a tear free display on a lesser performance budget than was done with Aero? Is it necessarily harder to do?

I'm not sure if that question is relevant, because I don't really know what the 'performance budget' for Aero is anyway. My laptop with the Intel 965M chip is probably one of the slowest possible chips with the SM2.x support for the full Aero Glass experience, and as I said, it handles it with ease. I can't tell the difference with my GeForce 8800.
And the 965M is pretty useless. You can barely even play a game like Half Life 2 on it acceptably.
So I don't know how low you could go before Aero Glass would actually have performance issues.

Hmm... then it would be interesting to know where the bulk of inefficiency comes from with Vista and later.

95DosBox wrote:

I would say from the results in FPS just from memory it benchmarked liked this on this particular laptop.
XP 32-bit SP3, Vista SP2 DX11, W7 SP1.

You can't compare the DX9 and DX10 versions of Crysis though. The DX10 version has far more detail and much higher quality, even on the 'same' detail settings. That's because it uses new DX10 shaders and texture formats (eg the shadows look better/sharper in DX10 because of this). As a result, it will also slower at the same settings on most hardware.
You can however run the game in DX9 mode on Vista and Windows 7 with a commandline switch.
If you do that, the performance differences between the three OSes should be negligible, unless you hit some kind of bottleneck somewhere (as in not having enough memory to keep enough of the game resident).

gdjacobs wrote:

Hmm... then it would be interesting to know where the bulk of inefficiency comes from with Vista and later.

Not the GUI at least.
I think what slows modern OSes down is things like built-in virus scanners, all sorts of other special security layers and containers, and all the 'whizz-bang' stuff like having thumbnails/previews in Explorer (which means you probably have a ton of plugins installed in Explorer, which scan every directory as you open it, and try to see if there are any files they need to generate a preview for. Which you can disable/uninstall I believe).

They aren't ineffecient in itself, they just do all sorts of things that older OSes didn't do, and that you do not necessarily want to do.
In terms of raw performance, as shown above with the XP/Vista/7 multithreading benchmarks, if it's just about crunching numbers and things, these modern OSes easily outperform XP and earlier.
You have to separate 'user experience' from 'extracting performance from hardware'. The former is highly personal and depends a lot on how you use your computer. The latter can easily be measured.
My experience as a software developer is that Windows only got faster over the years. Thread scheduling has improved, synchronization objects and other parts of the OS have adopted new CPU features to make them more efficient, IO prioritization has been introduced, things like disk caching and memory management just keep getting smarter, support for things like TRIM on SSD drives is implemented etc.
When I measure some of my high-performance code on a new version of Windows, I am never disappointed. It generally runs as fast or faster than before.
The only thing you must never EVER do, is to run your system with less memory than it needs. But memory is cheap these days, so just buy some.

Scali wrote:

95DosBox wrote:

I would say from the results in FPS just from memory it benchmarked liked this on this particular laptop. XP 32-bit SP3, Vista S […]
Show full quote

I would say from the results in FPS just from memory it benchmarked liked this on this particular laptop.
XP 32-bit SP3, Vista SP2 DX11, W7 SP1.[/qoute]

You can't compare the DX9 and DX10 versions of Crysis though. The DX10 version has far more detail and much higher quality, even on the 'same' detail settings. That's because it uses new DX10 shaders and texture formats (eg the shadows look better/sharper in DX10 because of this). As a result, it will also slower at the same settings on most hardware.
You can however run the game in DX9 mode on Vista and Windows 7 with a commandline switch.
If you do that, the performance differences between the three OSes should be negligible, unless you hit some kind of bottleneck somewhere (as in not having enough memory to keep enough of the game resident).

That's what I mentioned for a future test to force all three to use DX9 to balance it all out on the same playing field. One day this will happen if I get enough time. But between the two 64-bit on DX10 the Vista SP2 DX11 beat the W7 SP1. A future DX9 tri test would probably still go to XP the way I have it set up. I can't see with the extra load/bloat from Vista and W7 to edge XP out. Facts will prevail.

gdjacobs wrote:

Hmm... then it would be interesting to know where the bulk of inefficiency comes from with Vista and later.

I think the memory usage for the game and the video memory will play a significant part.

The laptop had only 512MB video memory and a dual core with 4GB.

I think around 8GB you will see a definite change as it helps Vista and W7.
16GB will probably notice all three evening out pretty much or the other two catching up to XP or possibly edging XP out.

But a lot of hardcore overclockers purposely install XP to max out benchmark performance. If XP really was worse I don't see why they go through all the trouble to get it to install and run it.

My laptop with the Intel 965M chip is probably one of the slowest possible chips with the SM2.x support for the full Aero Glass experience, and as I said, it handles it with ease.

Actually it was one of the faster Intel chips with Aero support 😀 . At the release time of Vista more Intel notebooks used the 940/945M chipset with GMA 950. GMA 950 was SM2 capable but lacked HW TnL and vertex shaders. They could be found on many business class laptops like HPQ NC6xxx series with the classic 4:3 aspect ratio but 1400x1050 resolution. They were very good notebooks with very good keyboards and a bunch of expansion ports (USB, CardBus, ExpressCard) and they were only about one year old. But Aero was very slow with them.
On the AMD side there were the Turion64 based HPQ NX6xxx series with the same case and screen. The situation seemed to be better and under classic 3D work they were actually better (they used ATI xpress 1100/1150). They actually reached the magical '3' Aero WEI score unreachable to GMA 950 😀. But under Aero they were actually slower during normal desktop usage. This was due to a special behavior of Turion64 (and all Athlon64) CPU's. Namely the memory clock was linked to the CPU core clock. So e.g. at 1600MHz that used CPU/5 divider the memory clock was 320MHz (DDR2-640). But under normal desktop usage scenarios the CPU switched to 800MHz power saving mode and so to 160Mhz (DDR2-320) memory clock. Without dedicated VRam the CPU and GPU had to shared this bandwidth. The reduced memory bandwidth under 1400x1500 (but even under 1024x768) made the experience awful under Aero. Both the Intel and AMD version of these series gave much better desktop experience under XP.
So it can be concluded that on the notebook front many popular chipsets at that time were not quite Aero ready.

95DosBox wrote:

But between the two 64-bit on DX10 the Vista SP2 DX11 beat the W7 SP1.

Yup, because there is little difference between the two technologically, as I said before.
Things can go either way in these benchmarks, the differences are very small, sometimes Vista is a smidge faster, other times, Windows 7 is.

95DosBox wrote:

A future DX9 tri test would probably still go to XP the way I have it set up. I can't see with the extra load/bloat from Vista and W7 to edge XP out.

It's not so much bloat as it is a completely different design.
In Vista, Microsoft completely redesigned the graphics driver, and moved a lot of processing out of the kernel-mode portion of the driver, and into user-mode.
As such, while Vista and later OSes still support DX9, the implementation is totally different, and is more of an 'emulation layer' on top of the new driver model that is aimed towards the DX10 API.
There may be slightly more overhead in DX9 in certain tests. In practice it is normally negligible though.
There are also other advantages: the drivers are now 'hogging' the system less in kernel-mode, meaning you get better cooperation between multiple threads/processes that use the GPU.
Of course, you will need specific types of benchmarks to see these differences. When doing benchmarks it is always important to understand what you want to measure, and how you can measure that.

95DosBox wrote:

But a lot of hardcore overclockers purposely install XP to max out benchmark performance. If XP really was worse I don't see why they go through all the trouble to get it to install and run it.

See above.
Perhaps in the overclockers-world, they value certain benchmarks that just happen to work best on XP for some reason. That doesn't say much about XP as a whole.
Or perhaps they just use XP because that is 'what they know'. It could well be that Windows 10 is actually faster, but nobody ever bothered to check, because they always benchmark with XP, and that's how they compare.

Falcosoft wrote:

Actually it was one of the faster Intel chips with Aero support 😀 . At the release time of Vista more Intel notebooks used the 940/945M chipset with GMA 950. GMA 950 was SM2 capable but lacked HW TnL and vertex shaders.

That conclusion is flawed 😀
Having hardware TnL and vertex shaders doesn't necessarily make your code run faster.
In fact, in some of my D3D code the 965 was actually slower than the Q35 (GMA3100) *because* it tried to use hw vertex shaders. Not just a bit slower, but 3 times as slow!
The reason? It only has 8 unified shader units. If you use software emulation for vertex shaders, all 8 units can be used as pixel shaders, and the CPU will take over the vertex processing.
If you use hardware vertex shaders, then you effectively have less pixel shaders, and the whole thing starts to bottleneck. The Q35 always has 8 dedicated pixel shaders, and performance-wise they seem about the same.
I actually blogged about that, back in the day: https://scalibq.wordpress.com/2010/02/05/dire … -how-very-nice/

This Q35 got a framerate of about 360 fps. My laptop with Intel GM965 (a DX10/SM4.0 part) only scored about 120 fps at best. Whi […]
Show full quote

This Q35 got a framerate of about 360 fps. My laptop with Intel GM965 (a DX10/SM4.0 part) only scored about 120 fps at best.
While debugging the software vertexprocessing fallback path in my D3D9 engine, I figured out why my GM965 was so much slower: The Q35 doesn’t have hardware vertexprocessing, so it will always default to software. When forcing my X3100 to software processing, I got about 315 fps out of it.
So it seems that while my X3100 has REAL vertexprocessing in hardware (some hardware used to report hardware vp for compatibility reasons, but would still use a CPU path internally), the hardware is so low-end that it’s actually slower than software vp.
Not too surprising in retrospect. After all, the X3100 has only 8 unified shaders. With software vp, that means it has 8 dedicated pixel pipelines. With hardware vp, it has to share the 8 shader pipes between vertex and pixel processing. The concept of unified shaders is a good one, but it goes from the assumption that a significant portion of the pipes are idle, so that they can be re-used for other tasks. Works fine when you have dozens of pipelines, but when you have only 8, they aren’t ever idle.

Shows again what I said about benchmarking: you have to know what you're measuring, and how to interpret the results.
The GM965 may be more 'feature-rich', but it isn't actually faster than the Q35. Even with swvp, the Q35 system had an edge.
Looks like the GM965's X3100 is little more than a 'DX10-ified' version of the GMA3000: they didn't add any extra pipelines, they just upgraded the pixel pipelines to SM4.0 unified pipelines, which made them capable of vertex processing.

Falcosoft wrote:

So it can be concluded that on the notebook front many popular chipsets at that time were not quite Aero ready.

It can? I never once used a system where I felt that the GUI was somehow slowing me down. And I always ran full Aero Glass.
So I would like to see that for myself first.
I don't know how much better the GMA3000 would be than a GMA950, if at all, but I wonder if it makes a difference.

Actually it was one of the faster Intel chips with Aero support 😀 . At the release time of Vista more Intel notebooks used the 940/945M chipset with GMA 950. GMA 950 was SM2 capable but lacked HW TnL and vertex shaders
...
That conclusion is flawed 😀

Where is a conclusion at all in that sentence (that can be flawed) ? 😀 I have not written it was slower because of the lack of TnL and vertex shaders. It was simply a part of the GMA 950's description.
I concluded that your 965M chip was the faster one, because you had not experienced problems with it using Aero.

It can? I never once used a system where I felt that the GUI was somehow slowing me down. And I always ran full Aero Glass.
So I would like to see that for myself first.

I hope I can make a demonstration video about it for you tomorrow.

Edit:
Ok I have created a video about an Nx6325 with a resolution of 1024x768. The problem is the screen capture itself is quite taxing for this PC and I had to force it to a lower performance state
to demonstrate the issue (as I have described above). So the outcome on the video is much worse than in real life. But I had switched between Aero enabled/disabled states so the difference can be noticed. The point is at 1400x1050 the situation is worse and problem free Aero usage is only guaranteed if no p-state changes are allowed. Unfortunately I cannot get an NC6320 with Intel core duo and GMA 950 anymore. But at that time Aero was not fluent on them either.
https://youtu.be/hn54lj15Qew

Falcosoft wrote:

Where is a conclusion at all in that sentence (that can be flawed) ? 😀 I have not written it was slower because of the lack of TnL and vertex shaders.

Well, you say it was one of the faster chips with Aero support. Which is a conclusion. Perhaps not drawn from TnL and vertex shaders, but a conclusion nonetheless. And a wrong one at that. It's a horribly slow chip, slower even than some of its less capable brethren.
Perhaps not the slowest, but certainly among the slower chips.

Falcosoft wrote:

I concluded that your 965M chip was the faster one, because you had not experienced problems with it using Aero.

Again, I don't think overall speed and problems with using Aero are necessarily related.

Hi ,
I have just edited my answer above since when I started editing it still was last one.

95DosBox wrote:

I think the memory usage for the game and the video memory will play a significant part. […]
Show full quote

gdjacobs wrote:

Hmm... then it would be interesting to know where the bulk of inefficiency comes from with Vista and later.

I think the memory usage for the game and the video memory will play a significant part.

The laptop had only 512MB video memory and a dual core with 4GB.

I think around 8GB you will see a definite change as it helps Vista and W7.
16GB will probably notice all three evening out pretty much or the other two catching up to XP or possibly edging XP out.

But a lot of hardcore overclockers purposely install XP to max out benchmark performance. If XP really was worse I don't see why they go through all the trouble to get it to install and run it.

Memory subsystems have not traditionally ramped as quickly as processor throughput, so this bothers me somewhat. What is driving memory consumption? Are developers precomputing significantly as an optimization?

To the average Joe, I suppose it's alright for memory requirements to just grow as they're often getting their new software with new hardware. However, as I've stated before, if an OS update requires more resources to run, I want to know it's worth it. MS Word, for instance, long ago hit a feature plateau, but system requirements still seem to grow with each new release.

I'm also interested in how the renderer and libraries stacked up for BeOS, as I remember it being extremely responsive and very clean as far as presentation.