Re: Only 4 multisamples on IR?

New Message Reply Date view Thread view Subject view Author view

Orad Hi-Tec Systems (orad_u++at++netvision.net.il)
Thu, 10 Oct 1996 09:07:46 +0300

Rob Jenkins wrote:
>
> On Oct 9, 6:54pm, Orad Hi-Tec Systems wrote:
> ..snip..
Rob Jenkins wrote:
  
> The people who really want alpha are typically those wanting to blend gfx over
> video, where the gfx has transparency in ( eg virtual sets ). They want high
> quality multisampling too - Mmmmm. You can only get alpha with 12 bit visuals
> on iR, so if you can jiggle your vof such that you have 12 bits/comp RGBA with
> 4 multisamples ( 1280x1024 with 1RM I think ) there's an OpenGL extension to
> support accumulating two multisampled images hence increasing image quality. On
> iR you should be able to 2 pass render your scene in real-time. Especially if
> you hava a small ( video sized ) scene or more than one RM - Note 2RM boards
> you can acheive RGBA12, 4 multisamples up to 1920x1080. Playing with
> ircomine/glxinfo and all that is the clearest way to see this.

Even more, in a video application scenario, you can set tup a cmb that
manages less than 1280x1024 pixels, only large enough to contain a
video-sized window (720x576, 720x486), plus maybe some small GUI area.
The monitor channel(s) can still get 1280x1024 format (below which most will
blank) by using the DG4 bi-linear interpolator.

The suggestion about 2-pass rendering is usually NOT applicable for a strict
real-time application like virtual studios. It is true that you can pile RM6's,
but pretty quickly you'll get bounded by the geometry. You have to go through
all the geometry transformations, lighting, etc. again for the second pass.

You can't add more GE's,
and also the CPU->IO4->GE path will bottleneck first. Maybe not on the Onyx2...
Speaking about the Onyx2, is the iR pipeline there free of the 8-samples + alpha
problem?

Just a thought, if the GE kept the data it sends to the RM's (2D-triangles
with Z and RGBA per vertex) and re-send it for the second pass, the geometry
calculations will not be repeated. This will also
work for accumulation-buffer, multi-pass anti-aliasing since the jitter is in
screen-space.

Another thing. The two-pass rendering will get you better anti-aliasing results.
This is what it was designed for. Will it also help the multisampled transparency
(double the number of transparency levels)? I am not sure.

Moshe Nissim
Orad Hi-Tec Systems
=======================================================================
List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
            Submissions: info-performer++at++sgi.com
        Admin. requests: info-performer-request++at++sgi.com

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:53:45 PDT

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.