Downsampling: How to - Page 2

andressergio · 06-23-2011, 19:30 | posts: 773 | Location: Montevideo / Uruguay / SA

Quote:

Originally Posted by Noisiv

3200/1800 = 16/9

nice man will try that as the other is no go as you

cheers !
Sergio

eiren · 06-24-2011, 20:33 | posts: 72 | Location: UK

Quote:

Originally Posted by Noisiv

Why would you prefer this over large array of normal AA modes? When they are available that is.

As I said, it's supersampling without as much GPU overhead that works in all games. I would still apply a small amount of AA though in most cases, especially for transparency AA.

However, there's little point if it's all being supersampled by the downscaling anyhow?

SamW · 06-24-2011, 21:02 | posts: 528

Isn't this just SSAA or whatever? Actually it is a little worse since you would be using a regular grid of sampling points which makes it kind of worse.

I don't think it is anymore efficient than regular super sampling. I mean to get 2x anti aliasing on 1920x1080 you will render from 3840x2160 and down sample back to 1920x1080.

Well I guess it is faster because from your example you get the equivalent of 1.33x SSAA which is quite a bit fewer samples per pixel than 2x.

Actually why don't they graphics card drivers allow fractional sampling and blending with SSAA?

Psycho101 · 06-24-2011, 21:34 | posts: 216 | Location: England

It is indeed a brute force form of OGSSAA. SGSSAA would offer better image quality over all and RGSSAA (no recent card on the market currently exists that can do Rotated Grid) would be the ultimate form of AA in terms of IQ.

I can see it being slightly more efficient than nVidia's own OGSSAA modes and their hybrids (accessed via nVidia inspector) purely because no further driver intervention is required. However the performance advantage would vary by game and in most cases won't be very large.

The only tangibe advantage to it is in games that are impossible to AA any other way, or that have severe disadvantages from forcing AA. One could also argue that in games that do support MSAA which have noticable peoblems with shader/specular aliasing, downsampling would be usefull too.

eiren · 07-21-2011, 00:55 | posts: 72 | Location: UK

Found a real issue/bug with this, and have no explanation as to why this is happening.

I have a custom resolution of 2498x1404 setup and being scaled down to 1080p.

I have noticed recently that whenever I am playing a game or watching a film on my computer, that my internet speed slows down to almost nothing. This is including any other devices that are on my wireless network which just directly connect to my wireless router and don't touch the computer that runs the games and films.

I have a 50MB cable connection, so normally it's rocket fast.

When I was playing a game on my computer, I could run a speedtest.net test and I would suddenly go down from 45Mbps to 0.01Mbps.

I tried everything to find the issue, and in the end... it was simply changing the resolution would instantly give me back all that bandwidth. I could run the speedtest on my laptop, and as soon as the resolution changed on my PC... it either made the internet very fast, or it slowed it down to nothing.

Anyone else come across anything like this?

I honestly have no idea how a custom resolution could be causing issues with a connected wireless router?

Raiga · 07-21-2011, 06:23 | posts: 973

Removed.

Raiga · 07-21-2011, 06:28 | posts: 973

Removed.

Grahf · *Last edited by Grahf; 07-21-2011 at 14:10.*

Quote:

Originally Posted by SamW

Isn't this just SSAA or whatever? Actually it is a little worse since you would be using a regular grid of sampling points which makes it kind of worse.

I came to this conclusion after *months* of poking around the German forums, toying with downsampling and SGSAA, etc. Regular SSAA forced via Nvidia Inspector is simply better.

I've yet to find a game where downsampling looked superior to forced SSAA, even in games which didn't support it.

Quote:

Originally Posted by Psycho101

SGSSAA would offer better image quality over all.

The German fellows I spoke to swore up and down that this was the case, but it simply isn't so IMHO. I had a thread going in the PC Hardware (de) forums where I even posted side by side screenshot comparisons to prove it (Mass Effect 2)

In my experience, SSAA forced via Nvidia Inspector eliminates jaggies better than downsampling+SGSAA or SGSAA alone...

rewt · *Last edited by rewt; 07-22-2011 at 09:05.*

Removed

Raiga · 07-21-2011, 19:01 | posts: 973

Removed.

Noisiv · **Memory**: 4GB OCZ 5-4-4-4-12

ezy ... I use it in all Source games cept TF2

and a pinch of TrSS

rewt · *Last edited by rewt; 07-22-2011 at 09:06.*

Removed

Noisiv · **Memory**: 4GB OCZ 5-4-4-4-12

wrong answer

rewt · *Last edited by rewt; 07-22-2011 at 09:06.*

Removed

Psycho101 · 07-21-2011, 21:35 | posts: 216 | Location: England

It seems that you're confusing RGSSAA with SGSSAA.

They're two totally different things.

As for Down sampling or regular OGSSAA being superior to SGSSAA, I'm not going to even try to argue based on opinion and what "looks" better to someone else etc. If one looks better than the other, then use it by all means. However, looking at the sample patterns for both, downsampling/OGSSAA will not cover near horizontal and near vertical edges as well as SGSSAA and no where near as well as RGSSAA.

Apologies for the incorrect statement in #29. I was talking within the remit of nVidia manufactured cards, but got carried away a little and forgot to emphasize that.

rewt · *Last edited by rewt; 07-22-2011 at 09:06.*

Removed

Raiga · 07-22-2011, 06:27 | posts: 973

Removed.

rewt · *Last edited by rewt; 07-22-2011 at 09:07.*

Removed

Raiga · 07-22-2011, 08:53 | posts: 973

Removed.

rewt · *Last edited by rewt; 07-22-2011 at 09:07.*

Removed

Raiga · 07-22-2011, 09:01 | posts: 973

I agree its trolling, hence removed all my post.

rewt · 07-22-2011, 09:05 | posts: 1,217 | Location: Americas

Quote:

Originally Posted by Raiga

I agree its trolling, hence removed all my post.

Fine, I will do the same. I suppose I was never entirely on point to begin with.

Raiga · 07-22-2011, 09:07 | posts: 973

Quote:

Originally Posted by rewt

Fine, I will do the same. I suppose I was never entirely on point to begin with.

Trying to cover your tracks that you claim Nvidia supports full screen Sparse Rotated Grid Super Sampling Mode!.

rewt · 07-22-2011, 09:11 | posts: 1,217 | Location: Americas

Haha, give it up man.

The words sparse and rotated are just words to define the position of sub-samples. I was correct the entire time, but it doesn't have a whole lot of relevance to this thread (especially with you trolling about), so I removed it

Raiga · 07-22-2011, 09:38 | posts: 973

Quote:

Originally Posted by rewt

Haha, give it up man.

The words sparse and rotated are just words to define the position of sub-samples. I was correct the entire time, but it doesn't have a whole lot of relevance to this thread (especially with you trolling about), so I removed it

I know what sparse (scattered) and Rotated meant, also how they are used to calculate average color value.

One question.

How how can you manage to take both both sub-pixel values of both scattered and rotated location without irregularity of the pattern across all the pixels and without sub-pixel samples overriding each other on the screen...Obviously all that from a single upscaled image rendered in real time (also downscaled in real time into your screen), without the extra overhead [of the already upscaled image to be rendered with the "x" multiplier being used for the number of samples taken on the screen (like 2x, 4x, 8x).]