Try older driver like 307 series or 341 series.

32xS is SuperSampling with MSAA 8x (it's AntiAliasing, not DSR to be strict).
Basicly for 1080p it means 4k with MSAA 8x on top.

agent_x007 wrote:

Try older driver like 307 series or 341 series.

Thanks for the tip but are you referring to something specific from above or is that just a general recommendation?

agent_x007 wrote:

32xS is SuperSampling with MSAA 8x (it's AntiAliasing, not DSR to be strict).
Basicly for 1080p it means 4k with MSAA 8x on top.

So SSAA is not exactly downsampling the same way as DSR would do it. I would imagine 4x DSR is simply averaging the dead center of 4 pixels into 1 pixel. But I think SSAA should provide similar benefits as DSR regarding the sharpness/detail of regular textures (something most AA methods just does not do). I might test it myself and post some screenies later.

This source: http://www.dahlsys.com/misc/antialias/ says:

When multiple samples are calculated per pixel, those samples are distributed across the pixel in patterns designed to maximize the AA effect. In particular, this means that the samples are not evenly distributed and do not line up horizontally or vertically. ..... For instance, with a resolution of 1920x1080 and 8xSSAA, the hardware must calculate the same amount of pixels as if the resolution was 7680x2160.

So in the case of 8xSSAA there are apparently 8 samples per pixel divided into 4 columns and 2 rows grid but thats not even true because they are not evenly distributed. Now I am curious where they got that 8xSSAA because my drivers allow only: 2x1, 1x2, 2x2, 3x3, and 4x4 SSAA. So there would need to be 4x2 option. Probably just an example though, could be an ATI/AMD setting too. So "2x2 SSAA" option is the one closest to "4x DSR" I think.

Technically the best option would be sparse-grid SSAA (SGSSAA) on NV. If you can get it to cooperate with your game.

AMD SSAA is similar to that.

DSR is certainly the least troublesome though.

swaaye wrote:

Technically the best option would be sparse-grid SSAA (SGSSAA) on NV. If you can get it to cooperate with your game.

AMD SSAA is similar to that.

DSR is certainly the least troublesome though.

You can actually do SSAA and transparency SGSSAA at the same time I think, for example I can select: "32xS (2x SS + 8x MS) + 8x SGSS". Not sure about the visual results (we are talking minuscule details here) but there was a noticeable dip in FPS compared to "only" "8x MS + 8x SGSS" so its definitely doing some extra work. If that is worth it, thats probably best decided over some screenies, which is ironic, but what the hell.

Regarding game compatibility, those AA options sometimes work without any compatibility flags, sometimes with 0x000000C1, 0x000010C1 or 0x000001C1, and other times with a flag listed here: https://docs.google.com/spreadsheets/d/1ekUZs … pub?output=html

Each game will be different, DSR on the other hand fails to scale the UI sometimes. My goal with this thread is to have more eye-candy options to play with on Windows XP. XP is usually considered as something where you cannot easily downsample but I think we can come up with some tricks to make it possible in some games.

I like downsampling because it improves texture details and I feel like lightning/post-process effects are more precise/nuanced, giving the final picture a more refined look. Let me show you an example (not from WinXP) with some of my old screenies, a game from 2009 (DX9) being rendered at 15360x7680 with tons of MSAA and then downsampled to 7680x2160. Thats over 50 FullHD monitors worth of pixels:

https://i.imgur.com/uJVkuDj.jpg
https://i.imgur.com/ZhrSoGK.jpg
https://i.imgur.com/v8Npnc1.jpg
https://i.imgur.com/zL7zRc5.jpg

Yeah that certainly looks excellent. I liked to use DSR when I had a 1360x768 home theater plasma TV. Rendering 2720x1536 and downsampling that naturally looked really good. The modern consoles even downsample to your native resolution in some cases.

I actually have not even considered looking at DSR for any of my XP gaming. Maybe because I rarely use such a powerful GPU in that setup and I'm also using a 1440p monitor so just running native resolution is a challenge.

Lets look at some seemingly identical screenshots 😀

NOTES:

- Click here for examples of notable areas >>>

- The difference in visual quality is much more apparent when you are moving around in-game compared to just looking at static screenshots, but it is easier to compare the various methods with screenshots because one can easily tab back and forth between them.

- HL2 main menu in all tests: Texture settings were set to "High", no AA

- Nvidia Profile Inspector settings = Default settings with only the following changes = AA Behavior Flags: None, AA Mode: Override, Texture filtering AF: 16x, Texture filtering: High Quality, and then the respective AA options from the screenshot labels

- 1080Ti tests use the latest steam build of HL2.

- 960 tests use an older nosteam build.

- Please ignore the gamma differences between various tests, they are related to my specific setup, which is not the same across all tests and is not the point of this test.

- DSR stands for 4x DSR factors at 0% smoothness.

- WBGDSR stands for tricking Window Borderless Gaming app to perform DSR-like downsampling. This method is finicky and has its own set of issues, depending on each game. For example in HL2, the main menu does not accept mouse for example (so you navigate the main menu with keyboard).

- All DSR and WBGDSR screenshots below are slightly "faked" because it is not (easily) possible to take 1:1 screenshots of the real result. Faking consisted of manually downsampling to the correct resolution via 3rd party software using "bicubic" (not "bicubic automatic") method. I paid close attention (visually, side-by-side) to this and the "bicubic" option seemed pretty much identical to the real result of both DSR and WBGDSR. "Bilinear" option produced more jaggies than the real result and "nearest neighbor" was even worse.

SOME OBSERVATIONS:

- During motion, anything that is less than perfectly aliased results in ants/shimmering near the edges of objects (less noticeable) or in transparent textures (more noticeable). Especially at longer distances (horizon) and/or against the skybox.

- The "2xSS+8xMS+8xSGSS+(WBG)DSR" option (subjectively) provided the best visual quality (during motion) but lowest FPS and highest input lag. The 8xMS is quite needed here to smooth everything out.

- Most of the "4x4SS" and "4x4SS+(WBG)DSR" tests have the same FPS for some reason, yet they look different (correctly I think).

- The rendering seems different between GTX960 and GTX1080Ti. Some spots have less or more shadows, on "Square" screenshots with 1080Ti the tree leaves are dark, etc. I am not sure how much is this the result of using 2 different builds of the same game, or if its the drivers/OS difference, or some combination. Not the point of this test though, just an observation.

- FPS is not everything, any form of AA/(WBG)DSR noticeably increased the (perceived) mouse input lag in all tests, the least laggy were probably the "(WBG)DSR" and "4x4SS" options. It was worse with 960 than with 1080ti because it is simply an older, less powerful card.

Half-Life 2 - Windows 8 - Geforce GTX 1080 Ti 1024x768

Station

1/1 - Win8 - 1080ti - x768 - NOAA
1/2 - Win8 - 1080ti - x768 - NOAA+DSR
1/3 - Win8 - 1080ti - x768 - 8xMS+8xSGSS
1/4 - Win8 - 1080ti - x768 - 8xMS+8xSGSS+DSR
1/5 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS
1/6 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS+DSR
1/7 - Win8 - 1080ti - x768 - 4x4SS
1/8 - Win8 - 1080ti - x768 - 4x4SS+DSR

Square

2/1 - Win8 - 1080ti - x768 - NOAA
2/2 - Win8 - 1080ti - x768 - NOAA+DSR
2/3 - Win8 - 1080ti - x768 - 8xMS+8xSGSS
2/4 - Win8 - 1080ti - x768 - 8xMS+8xSGSS+DSR
2/5 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS
2/6 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS+DSR
2/7 - Win8 - 1080ti - x768 - 4x4SS
2/8 - Win8 - 1080ti - x768 - 4x4SS+DSR

Corner

3/1 - Win8 - 1080ti - x768 - NOAA
3/2 - Win8 - 1080ti - x768 - NOAA+DSR
3/3 - Win8 - 1080ti - x768 - 8xMS+8xSGSS
3/4 - Win8 - 1080ti - x768 - 8xMS+8xSGSS+DSR
3/5 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS
3/6 - Win8 - 1080ti - x768 - 2xSS+8xMS+8xSGSS+DSR
3/7 - Win8 - 1080ti - x768 - 4x4SS
3/8 - Win8 - 1080ti - x768 - 4x4SS+DSR

Half-Life 2 - Windows 8 - Geforce GTX 1080 Ti 2496x1872

Station

4/1 - Win8 - 1080ti - x1872 - NOAA
4/2 - Win8 - 1080ti - x1872 - NOAA+DSR
4/3 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS
4/4 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS+DSR
4/5 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS
4/6 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS+DSR
4/7 - Win8 - 1080ti - x1872 - 4x4SS
4/8 - Win8 - 1080ti - x1872 - 4x4SS+DSR

Square

5/1 - Win8 - 1080ti - x1872 - NOAA
5/2 - Win8 - 1080ti - x1872 - NOAA+DSR
5/3 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS
5/4 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS+DSR
5/5 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS
5/6 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS+DSR
5/7 - Win8 - 1080ti - x1872 - 4x4SS
5/8 - Win8 - 1080ti - x1872 - 4x4SS+DSR

Corner

6/1 - Win8 - 1080ti - x1872 - NOAA
6/2 - Win8 - 1080ti - x1872 - NOAA+DSR
6/3 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS
6/4 - Win8 - 1080ti - x1872 - 8xMS+8xSGSS+DSR
6/5 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS
6/6 - Win8 - 1080ti - x1872 - 2xSS+8xMS+8xSGSS+DSR
6/7 - Win8 - 1080ti - x1872 - 4x4SS
6/8 - Win8 - 1080ti - x1872 - 4x4SS+DSR

Half-Life 2 - Windows XP 32bit - Geforce GTX 960 1024x768

Station

7/1 - WinXP - 960 - x768 - NOAA
7/2 - WinXP - 960 - x768 - NOAA+WBGDSR
7/3 - WinXP - 960 - x768 - 8xMS+8xSGSS
7/4 - WinXP - 960 - x768 - 8xMS+8xSGSS+WBGDSR
7/5 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS
7/6 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS+WBGDSR
7/7 - WinXP - 960 - x768 - 4x4SS
7/8 - WinXP - 960 - x768 - 4x4SS+WBGDSR

Square

8/1 - WinXP - 960 - x768 - NOAA
8/2 - WinXP - 960 - x768 - NOAA+WGBDSR
8/3 - WinXP - 960 - x768 - 8xMS+8xSGSS
8/4 - WinXP - 960 - x768 - 8xMS+8xSGSS+WBGDSR
8/5 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS
8/6 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS+WBGDSR
8/7 - WinXP - 960 - x768 - 4x4SS
8/8 - WinXP - 960 - x768 - 4x4SS+WBGDSR

Corner

9/1 - WinXP - 960 - x768 - NOAA
9/2 - WinXP - 960 - x768 - NOAA+WGBDSR
9/3 - WinXP - 960 - x768 - 8xMS+8xSGSS
9/4 - WinXP - 960 - x768 - 8xMS+8xSGSS+WBGDSR
9/5 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS
9/6 - WinXP - 960 - x768 - 2xSS+8xMS+8xSGSS+WBGDSR
9/7 - WinXP - 960 - x768 - 4x4SS
9/8 - WinXP - 960 - x768 - 4x4SS+WBGDSR

Half-Life 2 - Windows XP 32bit - Geforce GTX 960 2496x1872

Station

10/1 - WinXP - 960 - x1872 - NOAA
10/2 - WinXP - 960 - x1872 - NOAA+WBGDSR
10/3 - WinXP - 960 - x1872 - 8xMS+8xSGSS
10/4 - WinXP - 960 - x1872 - 8xMS+8xSGSS+WBGDSR
10/5 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS
10/6 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS+WBGDSR
10/7 - WinXP - 960 - x1872 - 4x4SS
10/8 - WinXP - 960 - x1872 - 4x4SS+WBGDSR

Square

11/1 - WinXP - 960 - x1872 - NOAA
11/2 - WinXP - 960 - x1872 - NOAA+WBGDSR
11/3 - WinXP - 960 - x1872 - 8xMS+8xSGSS
11/4 - WinXP - 960 - x1872 - 8xMS+8xSGSS+WBGDSR
11/5 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS
11/6 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS+WBGDSR
11/7 - WinXP - 960 - x1872 - 4x4SS
11/8 - WinXP - 960 - x1872 - 4x4SS+WBGDSR

Corner

12/1 - WinXP - 960 - x1872 - NOAA
12/2 - WinXP - 960 - x1872 - NOAA+WBGDSR
12/3 - WinXP - 960 - x1872 - 8xMS+8xSGSS
12/4 - WinXP - 960 - x1872 - 8xMS+8xSGSS+WBGDSR
12/5 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS
12/6 - WinXP - 960 - x1872 - 2xSS+8xMS+8xSGSS+WBGDSR
12/7 - WinXP - 960 - x1872 - 4x4SS
12/8 - WinXP - 960 - x1872 - 4x4SS+WBGDSR

I think GTX 960 simply isn't good enough for all those AA options to be enabled (will work fine for quality comparisons). Since you say best option had highest input lag, I guess it had low 1% or .1% numbers ?

Main issue : GTX 960 is 1/2 of GTX 980.
32 ROPs on 128-bit bus and 7GHz memory with 2GB (or 4GB) VRAM.
I think you might simply overtax this GPU when you use so many different AA options on top of each other. Getting GTX 780 Ti or GTX 980 (maybe Ti ?), might be a better option for XP.

I have not measured the input lag (dont have the necessary equipment) but the lag was definitely noticeable which is I think important to note. I think this is normal to happen even when you adjust in-game post-process or AA effects. When I play Battlefield(BC2) online for example I only use 4xDSR and select SSAO instead of HBAO because I felt like this setup is the best looking without compromising the input lag too much (which is crucial if you want to stay "competitive"). One would really need an extremely overpowered card to not feel any extra input lag from the extra eye-candy (Q3A on 1080ti with AA+DSR is fine). I think GPU usage from Afterburner could be a good rough indicator if you are pushing it too much (and probably increasing input lag).

I agree that GTX960 is a bit slow for things like this. I plan on getting 980ti eventually (if I decide to actually use a setup like this). I already had GTX960(2GB) lying around so I thought I would put it through its paces 😈