Neverending Upscaling/Resolutions/AA etc Thread #2 *Rules: post: #616 *

hi Quaz you mind checking Vegas 2, its UE3 also

http://static.videogamer.com/videogamer/images/pub/misc/vsmarch18rb6v2large03.jpg

http://static.videogamer.com/videogamer/images/pub/misc/vsmarch18rb6v2large01.jpg

I think I see more jaggies on the 360 side on the electric tower.

 

very excessive jpeg compression

 

really, it glue the ps3 and 360 version together, sorry bro.

 

I seriously doubt that. Downscaling with such small resolution difference will provide uneven "AA" (some parts of the edge will see no blur, some other will see a lot).
A proper 2-d LP filter should be able to do better with consistency, and certainly would not degrade textures more than downscaling.

Plus, they are probably loosing some memory due to downscaling.

Where are those shots from? Are they authentic enough?
That would be a first if they are.

 

they are from a comparison on this site

http://www.videogamer.com/features/article/18-03-2008-342.html

taken with HDMI + RGB full range on PS3 and component with refenrence level set to extended on 360. By the way the color and contrast look, seems very real because that usually the way it is when you compare a PS3 and 360. Im more surprise that the PS3 version have less aliasing than the 360 in comparison without much inspection on these screen shots. Dont think there are any UE3 games on PS3 have AA at all. Could it be another blur filter like GRAW2? Or the first to have some port of AA.

 

not me

 

i don't really interested to analyse very bad screenshot like this. MSAA2x and Quincunx have really similar edge aspect, impossible to differentiate with lot of compression like that. you can just use "feeling" but feeling don't interest me (blur can have lot of different origin other than quincunx)
and when you can make Quincunx, you can make MSAA2x too (it's even little more easy) then ...

 

685/(720 ~= 19/20

That is every 19 pixels on the downscaled version correspond to 20 pixels on the original.
So around every 20th pixel you should see worst kind of blur ("avarage" for bilinear interpolation, nothing better for bicubic), and in the middle of those areas almost no correlation between neighbors (no "AA"), and something in between for the rest.

Again, this cannot possibly be better than a simple 2D LP filter.

 

yes i know that of course (and i inevitably see that with my eyes when i counted )

it's better because preserve more details of the original image (760p downscaled in 720 can even add detail in relation to native 720p), i see lot of game that use filter for hide AA and it's disaster every time... i hate that

 

How about Virtua Fighter 5 on PS3. The game has no AA, 1024x024 rendering resolution, however in the options there is a 'soft mode' which seems to give the impression of slightly less jaggies in some areas than proper 2X AA.

Is that similar to the soft mode found in Tekken 4 and 5 on PS2 when run in 480i 6hz mode, a cheap way to do AA?

In VF this soft mode doesn't seem to blur or ruin the image much at all, so would it not be possible for developers to use the technique in games which are too memory intensive to use true 2X AA?

 

Overall 1.056xSSAA is better than 1xAA in terms of accurate edge information but that's not we are discussing, since they are not comparable in terms of resource usage (10 percent more pixel shaders, BW etc).

The question is whether 685p 1.05xSSAA better than 720p 1xAA with LP in terms of edge and texture quality, and I claim there is a simple LP convolution matrix that makes the answer "no" for both edges and textures, with the bonus of consistency.

I'd think bad examples you observed, are simply results of poor blur filters (such as averaging) which may be very cheap on RSX or Xenos (I wouldn't know).

 

Speaking of VF5, that reminds me of Yakuza 3, Quaz you mind checking this

http://image.com.com/gamespot/images..._screen002.jpg

demo also on JP PSN

Yakuza 3, seems to be running on the VF5 engine or at least they both look very similar, wonder if its 1024X1024 also.

and is this possible to check at all LOL?

http://www.watch.impress.co.jp/game/...0224/val11.htm

Battle Field Valkyrie

 

Okay, this most likely have been posted before and discussed, but my search was a bit inconclusive so it does not really matter if I do it again, I hope

Was looking over the Insomniac web site and the R&D page had a Resistance Fall of Man debrief

http://www.insomniacgames.com/tech/articles/0108/files/RFOM_Debriefing_public.pdf

Code:

             Frame Buffer Setup
  
 * Display Buffers

   *  2 x 1280x720 32-bit RGBA
   *  Both display buffers share a single tile to reduce wasted memory
   *  We did not support 1080, couldn’t afford the memory
   
* Render Buffer

    * 1280x720, 32bit, 2xMSAA
    * 32-bit depth buffer
    * Color and depth in their own tiles with compression enabled
    * Always render 1280x720 down sample for NTSC and PAL
   
 

And

Code:

* Alternate Render Buffer

   * 1280x704, 32bit, 2xMSAA
   * 704 is the magic height as it obeys all the restrictions of depth and color tiles
   * Center on the 720 front buffer, giving 8 pixels of black top and bottom
   * No wasted memory due to alignment. Over 1Mb saved from using 1280x720 and 2% faster

* Idea was too late to be useful on Resistance

I remember people talking about why doing 704 instead of 720 (Ratchet?), well here we have the reason, atleast what their reason was.

 

2% faster..so whats the point

 

And 1MB+ extra, everything counts on consoles (on PC to but not to the same degree).

 

Also it makes better sense from the theoretical design POV. 1280x704 better fits digital data and processes as it's an array of 64x64 quads - it's all base-two friendly. The chosen resolution 1280x720 is based on an arbitrary 16:9 aspect without regard for real-time processing or image creation. If computer engineers were designing a display for games and image processing, 1280x704, a 20:11 aspect ratio, would be a better choice. Same with 1080p. 1920x1088 would be a better choice than 1920x1080.

It's only because the TV companies chose certain aspects and resolutions that everyone is now used to that makes a more sensible base-two design seem weird!