More info about RSX from NVIDIA

mckmas8808 said:
A good example of TSAA. Look at the tree branches in the background too. Nice.

df1cb8d8-7f00-4c82-85e8-bb517f00c25a.jpg

Sorry for the ignorance, but how is that different from normal AA?
 
Sorry for the ignorance, but how is that different from normal AA?
multi sampling doesn't aa the alpha masks i believe. So any transparent things don't get aa'd . Its why things like grates or tree branches or windows don't have the aa in them


I could be wrong though
 
jvd said:
Sorry for the ignorance, but how is that different from normal AA?
multi sampling doesn't aa the alpha masks i believe. So any transparent things don't get aa'd . Its why things like grates or tree branches or windows don't have the aa in them


I could be wrong though

Oh so i don't suppose it's faster. If anything that sounds more complicated and possibly slower than the other bitchy AA methods.
 
Yeah, the first pic has "normal" AA. But it doesn't handle textures..that grass is a texture, not geometry.

There are better comparison pics, perhaps, in that article, with fencing from HL2.

Also, this is AA just for these textures, it's not AA applied generally to the scene. You'd still have "normal" AA running in conjunction with it. You can run the "normal" AA without it, if you wish (and vice versa, I suppose).
 
jvd said:
Sorry for the ignorance, but how is that different from normal AA?
multi sampling doesn't aa the alpha masks i believe. So any transparent things don't get aa'd . Its why things like grates or tree branches or windows don't have the aa in them


I could be wrong though
If you're right that will make a big difference in image quality in alot of games. I wonder what type of performance ie. bandwidth, clocks this takes.
 
london-boy said:
jvd said:
Sorry for the ignorance, but how is that different from normal AA?
multi sampling doesn't aa the alpha masks i believe. So any transparent things don't get aa'd . Its why things like grates or tree branches or windows don't have the aa in them


I could be wrong though

Oh so i don't suppose it's faster. If anything that sounds more complicated and possibly slower than the other bitchy AA methods.

I don't know it it would be faster . I belive its some where between multisampling and supersampling . Supersampling would fix up those tree branches also but its basicly rendering in 2x the res and scaling it down or 4x the res and scalling it down. Which is much more expensive than multi sampling .


As i've said i believe the r500 is capable of this too though i'm not a 100% positve
 
Do you know if speedtree uses alpha to render parts of their trees or is it all geometry?
Alpha. I've used an older version of it, and the branches are typically all geometry while the leaves are usually in patches of billboards. I'm sure the next-gen versions are a little less conservative with the geometry, but individual leaf geometry is beyond the scope of what PS7 and Xbox6 (I'm sorry, XboXtreme9000) will be able to handle.

I think you can, however, use all geometry for the SpeedTree MAX plugin and use full geometry for a pre-render.
 
ERP said:
ralexand said:
mckmas8808 said:
A good example of TSAA. Look at the tree branches in the background too. Nice.
Cool, does that new AA come for free?

My guess is that they are just enabling supersampling for transparent draw operations (usuall alpha kill stuff that MSAA doesn't work with). This was entirely possible even on NV2X, although not exposed anywhere in the API.

Well, TSAA I don't know about, but TMAA on the G70/RSX is just alpha-to-mask functionality. The R520 is supposed to have it to. It will work with multisampling and should essentially be free.
 
So I guess the next question is...

MS/ATI have stated that with Xenos 4xAA is basically free, and will be required (at 2x a minimum?) in every game. But are they talking TMAA or TSAA or the current form of AA which doesn't get this benefit.

There seems to be a noticable difference, and there's no point saying "Xenos can do it too" if it can but won't do it.

It looks good :)
 
PARANOiA said:
So I guess the next question is...

MS/ATI have stated that with Xenos 4xAA is basically free, and will be required (at 2x a minimum?) in every game. But are they talking TMAA or TSAA or the current form of AA which doesn't get this benefit.

There seems to be a noticable difference, and there's no point saying "Xenos can do it too" if it can but won't do it.

It looks good :)
Well you should be able to do both.

How many polys are we talking about?
Good question. Anyone know?
 
What i'm wondering is if it can do fsaa on hdr which is something the current nv40s can't and I don't believe the r520 can. Only the xenos can
 
thanks , btw where did u get the info that the r520 can do it ?




Anyway as nice as the aa on the alpha mask is , to me hdr fsaa is the most important , i really hope they have it in the rsx or we are going to have to trade one eye candy for another
 
G70/RSX supposedly supports sparse grid supersampling (according to Xmas), which is a very nice IQ boost to fix aliasing caused by pixel shaders which today is not helped by MSAA at all, and won't be fixed by alphatomask either.

If sparse grid supersampling is really supported, that you could render-to-texture a supersampled HDR buffer, and use that to get HDR FSAA but at superior quality (of course, it ain't free)
 
DemoCoder said:
G70/RSX supposedly supports sparse grid supersampling (according to Xmas), which is a very nice IQ boost to fix aliasing caused by pixel shaders which today is not helped by MSAA at all, and won't be fixed by alphatomask either.


Can it work on hdr though ?


Anyway i see your post in 3dtech , how about u link us to it
 
jvd said:
thanks , btw where did u get the info that the r520 can do it ?

This is not 100% certain information but a hint from wavey which i intepreted as r520 will be able to do fp16 with AA.
 
Back
Top