FSAA to take a fillrate hit and not a bandwidth hit

K.I.L.E.R

Retarded moron
Veteran
Is it possible in the future? Assuming we use holograms that are stored in atoms and those atoms use a very large space? The atoms can transfer as much data as atomically possible and travel at the speed of light or close to.

Bandwidth made irrelevant, so does that mean fillrate will be impeded by FSAA? (assuming fillrate with triangles and such are used to fill in the hologram)

Is there any current method to allow a fillrate ONLY hit? Not SS as that takes bandwidth as well as a fillrate hit.
 
Possibly the only way to avoid a bandwidth hit is to combine the computed sub-samples before writing them out to the main memory.

But this means you have to process every triangle for a given pixel, before writing it out.
So you eighter need sufficient on-chip memory to store the entire framebuffer, or you have to divide up the framebuffer to areas (eg. tiles) which can fit on-chip and process them one-by-one.

Of course it takes internal bandwidth, but it's much cheaper to improve than the external bandwidth.

Until then will see MSAA with compression (FAA can be treated as a compression scheme as well), since MSAA produces highly compressible output. (While SSAA does not!)
 
Or you could do something like change your rasterization method to output edge equations, and then analytically integrate them on scan-out.

Mathematically intensive (so it would incur a huge fillrate hit, relative to sample-based AA), but it wouldn't have a measurable bandwidth hit.
 
But with such a thing, gking, you always have the problem of variable data sets per frame. This automatically eliminates many efficiencies that today's 3D hardware uses.

That doesn't make it impossible, but I just don't see it happening, most particularly as triangle counts increase.
 
The question presupposes a fillrate hit vs no bandwidth hit, otherwise I could think of a fillrate/bandwidth free sollution *ducks*.
 
I never said the system was practical -- in fact, pretty much any system that involves analytical evaluation of equations is guaranteed to be impractical. It just happens to cost lots of fillrate (lots of integrands per pixel), while not having a bandwidth hit (primarily due to the fact that it's impossible to measure how much bandwidth an aliased frame would consume because there is a variable amount of data per frame).
 
Ailuros said:
The question presupposes a fillrate hit vs no bandwidth hit, otherwise I could think of a fillrate/bandwidth free sollution *ducks*.

What? Not rendering? *ducks*
 
Chalnoth said:
Ailuros said:
The question presupposes a fillrate hit vs no bandwidth hit, otherwise I could think of a fillrate/bandwidth free sollution *ducks*.

What? Not rendering? *ducks*

Or just a simple gaussian type blur filter heh. I had more an exotic algorithm on a Tiler in mind.
 
Heh, yeah, I know. Just couldn't resist...though a tiler method would have a very slight fillrate hit, just as MSAA does...

I'd be very interested to know what these companies are going to do when polygons start approaching sub-pixel size.
 
Back
Top