Vertex texture look-ups on non-shader model 3.0 hw

[K.I.L.E.R]

http://hem.bredband.net/grahei/web_res/progs.html

This demo was created to demonstrate that we can effectively move the vertex texture look up from the GPU to the CPU by simply making a function that does the look up and sends it as a variable to the GPU. It is very useful when you need to perform vertex texture look up but do not have the hardware that supports shader model 3.0.

This is neat, I wonder if developers are going to use this?
I hope this isn't old news to you guys, I've just found this out at Gamedev.

 

[991060]

how does he do that actrually? there's no source code.

 

[Hyp-X]

Yeah, try to use the CPU for a lookup in a render target texture...

 

[K.I.L.E.R]

Yeah, try to use the CPU for a lookup in a render target texture...

What's wrong with that?

 

[nAo]

What's wrong with that?

it's slooooooow

 

[RejZoR]

No,you're wrong. It's sloooooooooooooooooooooooooooooooooooooooow

 

[K.I.L.E.R]

Yeh, but if your CPU isn't doing much then why not use it?

 

[Ladis]

Yeh, but if your CPU isn't doing much then why not use it?

It's appliable for technological demos only. You use CPU power for other things in real games (collisions, physics, pathfinding, AI, ...).

 

[K.I.L.E.R]

Yeh, but if your CPU isn't doing much then why not use it?

It's appliable for technological demos only. You use CPU power for other things in real games (collisions, physics, pathfinding, AI, ...).

Most real games don't come close to using 100% CPU time, so why not use some extra time for vertex texture lookups if it would increase overall framerate?

 

[trinibwoy]

Yeh, but if your CPU isn't doing much then why not use it?

It's appliable for technological demos only. You use CPU power for other things in real games (collisions, physics, pathfinding, AI, ...).

Most real games don't come close to using 100% CPU time, so why not use some extra time for vertex texture lookups if it would increase overall framerate?

I thought this was a way to implement vertex texturing using the CPU. How does that bring a performance increase compared to no vertex texturing at all? I can only see performance going down even if there are a few cpu cycles to spare.

 

[Frank]

Yeh, but if your CPU isn't doing much then why not use it?

It's appliable for technological demos only. You use CPU power for other things in real games (collisions, physics, pathfinding, AI, ...).

Most real games don't come close to using 100% CPU time, so why not use some extra time for vertex texture lookups if it would increase overall framerate?

How many do you want to look up? What do you want to do with it? And what processor do you need to do that?

 

[K.I.L.E.R]

Okay, I get the point now. The "how many?" question just got me.

Thanks.

 

[KimB]

Most real games don't come close to using 100% CPU time, so why not use some extra time for vertex texture lookups if it would increase overall framerate?

Well, for one, if you're talking about the results of a render-to-texture operation, transferring that back to system memory is going to cause a massive performance hit.

 

[Mintmaster]

Does anyone know of a good tutorial about superbuffer/uberbuffer extensions?

The only one I know of is the particle system presentation on ATI's web site, and I'm looking for something that explains the fundamentals of these extensions a bit better.

This is probably even better than vertex texturing for performance, provided you don't have to change render targets too often.