[ChrisRay]

Alright, These are just some basic questions. I have heard several different optimisation possibilities for both the NV4x/r420 derivitive. Obviously treating the Nv4x as an r300 doesnt yield the best possible results and vice versa.

Some things I have noticed about NV4x

1) Still needs proper compiling for optimal performance. However not nearly as sensitive to register space as before.

2) FP16 still benefits the NV4x slightly. However my own testing among a few different ((Current)) games suggests it doesnt benefit it as much as I originally believed. ((we're talking roughly 5-10%))

3)NV4x seems to do be able to do FP16 "Normalise" instructions for free. ((Thanks Dave))

4) Shaders using .nrm. Perhaps could also further improve FP16 performance.

My Information on the r300/r420 is somewhat limited so please forgive me.

1) Obviously seems to prefer floats to texture lookups and normalised cubemaps.

2) Seems extremely sensitive to CG compiled runtime ((This could very just be an compiler optimisation issue))

Now. I'm mostly interested in the kinds of tweaks that are going to best going to optimise for the Nv4x verses r300 derivitive chips. If anyone has anything they'd like to add. Or Fill in the blanks please do

[PeterAce]

Originally Posted by ChrisRay

Alright, These are just some basic questions. I have heard several different optimisation possibilities for both the NV4x/r420 derivitive. Obviously treating the Nv4x as an r300 doesnt yield the best possible results and vice versa.

Some things I have noticed about NV4x

<snip>

4) Shaders using .nrm. Perhaps could also further improve FP16 performance.

You might want to take a look at this thread :

http://www.beyond3d.com/forum/viewto...amp;highlight=

My Information on the r300/r420 is somewhat limited so please forgive me.

1) Obviously seems to prefer floats to texture lookups and normalised cubemaps.

2) Seems extremely sensitive to CG compiled runtime ((This could very just be an compiler optimisation issue))

Yes, it seems to be a compiler issue I think the CG compiler output lots of unnessisary swizzles, look here :

http://www.beyond3d.com/forum/viewto...=190182#190182

and

http://www.beyond3d.com/forum/viewto...ght=cg+swizzle

Now. I'm mostly interested in the kinds of tweaks that are going to best going to optimise for the Nv4x verses r300 derivitive chips. If anyone has anything they'd like to add. Or Fill in the blanks please do

You have to also realise that a shader written for R3X0 isn't always exactly the best otimized shader for R4X0 - just take a look at the performace increases (for R4X0) in Farcry using PS 2.0b longer shaders.

Another example of this is also how on ATI's ASHLI viewer you can see performance differences (on sum on the provided shaders) when compiling for different targets (R300 and R420).