Fillrate limited

This topic contains 1 reply, has 2 voices, and was last updated by  Joe Davis 6 years, 11 months ago.

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
    Posts
  • #30375

    GeraldJB
    Member

    Hello,

    Just a little question, what is the best method to make a rain specialFX on powerVR architecture?

    Actually I blit many bigs quads on the screen (the rain drops are in these quads), it’s working well on devices with a 480*320 resolution.

    The problem is when I’d try this method on a device with a bigger resolution (1024*768), the rendering is a lot slower.

    I know powerVR doesn’t like much transparency because the tile based deferred rendering, but the problem is I don’t see how to make a rainFX without transparency.

    So I had think to this forum, because I’m sure there’s somebody of your staff which know the best way to do it.

    On a paper I had read, I saw a tip which suggest to send more geometry to the gpu, instead of blitting big quads with empty space in it.

    The problem is it’s a lot of drops to draw to fill the screen !

    What is the best way to do it ?

    Many thanks,

    Gerald

    #34563

    Joe Davis
    Member

    Hi Gerald,

    It’s not so much that our hardware takes a bigger hit with translucency than other architectures, it just that there isn’t the big gain our architecture offers when rendering opaque objects 😉

    The paper you’ve read sounds similar to the approach I would suggest. By using a particle system to draw your rain, you can significantly reduce the amount of your screen that requires blending. That being said, this will increase your CPU load, and you may lose some of the advantage this approach offers if you are drawing a huge amount of rain in your scene (i.e. as you approach the screen being almost completely covered in rain, you will be taking the same blending hit as your full screen would but will also have the cost of submitting the geometry).

    The efficiency of one technique over another is completely dependent on the end result you are trying to achieve. Can you provide us with screen shots or video of your current approach so we can suggest changes? Either post this as a link in this forum, or send them directly to devtech@imgtec.com if you do not want this to be publicly visible.

    One question I do have on your current technique – how many full screen transparent quads are you currently drawing? You can quickly improve your current implementation by reducing the number of translucent layers you are drawing. I still think the particle system is likely to offer a better result though. Joe2010-11-25 15:55:02

Viewing 2 posts - 1 through 2 (of 2 total)
You must be logged in to reply to this topic.