Is 360 cursed with sub 720p resolution in games using deferred renderer engines?

[Ruskie]

I have been looking at Frostbite 2 thread and there was a talk about creating a G-buffer on 360 so it would fit(20 megs?) in eDRAM with two passes,it was never finished so was I wondering...Seeing alot of games using deferred rendering engines on 360,some(GTA IV,Blur.NFS HP) are native 720p,even with 2x-4xMSAA and some(Crysis 2,Split Second,Halo Reach?) are all cut to 1152x720 resolution(presumably because they also have to fit HDR), can you set up G Buffer that way that your game(using deferred renderer) is native 720p with HDR on 360

Edit.

Oh and I should have pointed out in the tittle also while using HDR.

 

[AlNom]

To avoid tiling, yes.

 

[Kameradschaft]

To avoid tiling, yes.

Sorry for the question guys but the two passes (20mb) trick in BF3 actually means that the game will run at 1280*720p? If yes then it will be interested to see if they will be able to implement any AA solution...if it's possible considering the limitations of edram.

 

[Ruskie]

Sorry for the question guys but the two passes (20mb) trick in BF3 actually means that the game will run at 1280*720p? If yes then it will be interested to see if they will be able to implement any AA solution...if it's possible considering the limitations of edram.

MSAA is no no,but I'm wondering if MLAA or other kind of post process AA(DLAA,FXAA...) would affect buffer size?

 

[Boxercide]

*mod snip of reply to OT post*

Back to the topic: Has there been any discussion from DICE on what the AA solution for 360 may be? Or how they are going to set up the PC options? I remember seeing a comparison of different techniques, but i was wondering if they'll let users set them to their own preference.

ie: Some may like MLAA + 4xMSAA while somebody else may prefer nVidia's FXAA, etc.

 

[AlNom]

MSAA is no no,but I'm wondering if MLAA or other kind of post process AA(DLAA,FXAA...) would affect buffer size?

There's a memory cost incurred in storing the frame so that the shaders (SPUs on PS3) can operate on the frame in a later pass.

 

[Ruskie]

There's a memory cost incurred in storing the frame so that the shaders (SPUs on PS3) can operate on the frame in a later pass.

So,MLAA is kind of useless on 360 in deferred renderer game?Since its memory footprint is 2x the size of the backbuffer.That would make frame buffer twice as big,and you would have to tile even more,right?

 

[AlNom]

I'm not sure I understand what you mean by more tiles. Say you try to render 4 MRTs that's 1280x720. Then the four targets are written to in edram until it gets filled. I'm a little unclear here, but I think at this point you can do post-processing on what's rendered so far (at least, the object outlines for Crackdown are done at this stage). That's all resolved to main mem. The rest of the frame is rendered in eDRAM and so on. Of course, there's going to be an issue at the frame boundaries where the edram had to split up i.e. tile the frame. The object outline in crackdown saw this issue, and I imagine it's conceptually not that different from any other post-edge rendering. They had to re-render a few extra lines from the previous tile so as to get consistency between the tiles. Of course, not having to tile means not having to worry about post-process image discontinuities too. MSAA is more problematic from a shading power point of view (shading at sample frequency for correctness).

hm... so at least with the edram there, there shouldn't be extra storage required since it's all done in the temporary edram memory space.

 

[Ruskie]

Oh my bad,I amm...confused g buffer and frame buffer :smile:

Still,in description it says that MLAA has twice a footprint of back buffer(like 2xMSAA) and that from my understanding requires additional tilling,no?What I mean now,720p alone is ~7 megs but with MLAA it will be 14 megs meaning using it you will have to tile,without it you wouldn't have to.

Sorry for confusing you

 

[AlNom]

Are you referring to that 360 MLAA paper with the Lionhead guy? IIRC, the double footprint is from having to use a particular format, which is wasteful. I can't remember if it was due to XNA... It's been awhile.

 

[Ruskie]

Are you referring to that 360 MLAA paper with the Lionhead guy? IIRC, the double footprint is from having to use a particular format, which is wasteful. I can't remember if it was due to XNA... It's been awhile.

Yea that one.They have port using XDK,but it wasn't Jimenez or Lionhead,some other developers did it and timings are 1.3ms. So with XDK it doesn't require double memory footprint?

 

[scently]

Yea that one.They have port using XDK,but it wasn't Jimenez or Lionhead,some other developers did it and timings are 1.3ms. So with XDK it doesn't require double memory footprint?

No, it was posted on their official twitter page by Jimenez himself: http://twitter.com/#!/iryoku1/status/62502539333087232, cutting the render time from 3.7 ms in XNA to 1.3 ms in an XDK build.

As for BF3, on of their devs(?) managed to get FXAA rendering on the 360 at 1.3 ms too: http://timothylottes.blogspot.com/2011/04/nvidia-fxaa-ii-for-console.html

 

[Gitaroo]

seems like FXAA also blur the IQ judging from the screen.

 

[Rangers]

Would Brink (1120X720 on 360) be a new example of the topic title?

 

[AlNom]

I believe so.

 

[sebbbi]

some(Crysis 2,Split Second,Halo Reach?) are all cut to 1152x720 resolution

Split Second's and Trials HD's deferred renderers use 1280x672 (860160 pixels, 93.3% of 720p) resolution. That provides slightly more detail than 1152x720 (829440 pixels, 90% of 720p). Split Second scales the 1280x672 to 1280x720. Trials HD resolves it to the center of 1280x720 final buffer (and renders UI on top) to keep everything pixel precise (less blurring), but this results in more wide view (24 pixel black up/bottom borders - that are fortunately often masked out by TV overscan).

Yea that one.They have port using XDK,but it wasn't Jimenez or Lionhead,some other developers did it and timings are 1.3ms. So with XDK it doesn't require double memory footprint?

Usually you only have the active render target in the EDRAM (and you copy it to the main memory when you finish rendering to it). So when you are rendering the post processing effects, your g-buffer layout doesn't really matter anymore. If you want to save main memory, you can use the same memory areas as you are using for the g-buffers to do all your post processing (since you do not need the g-buffer contents after the lighting pass).

This basically means that MLAA doesn't use any extra memory on Xbox 360.