Angle-independent AF on Xenos?

[kyleb]

I am extatic about the new high quality anisotropic flitering on the r5xx chips, so I can't help but wonder if the r500 is designed to do the same. Anyone know?

 

[Acert93]

I am extatic about the new high quality anisotropic flitering on the r5xx chips, so I can't help but wonder if the r500 is designed to do the same. Anyone know?

I asked around yesterday and no one seems to know, at least those I chatted with. Dave may know... but as it stands there are a LOT of little details on Xenos we do not know yet.

Maybe ERP, Dave, DeanoC, or someone can come to their thread and shake their head in disgust... but let us know if their head is shaking up and down or left and right

 

[Guden Oden]

Didn't dave's xenos article state it's a fixed pattern? Considering the relatively light transistor count for such a high-class piece of hardware, I'd be surprised if it had esoteric features such as this. Particulary in a product aimed at a market that is unlikely to appreciate the effects of it...

 

[Dave Baumann]

I've been trying to get some clarification on this myself. they descibed Xenos has not having the issues the R300-R420 series as having, but that was all. I've since mailed asking about it but those on the R520 side don't know and those on the Xenos side haven't replied.

I would guess that it probaby does use the new filter as I doubt they would want be be using too many separate alrgorithms on concurrent parts, but at this point this is nothing more than a guess. If I get further clarification, you'll know.

 

[Acert93]

I've been trying to get some clarification on this myself. they descibed Xenos has not having the issues the R300-R420 series as having, but that was all. I've since mailed asking about it but those on the R520 side don't know and those on the Xenos side haven't replied.

I would guess that it probaby does use the new filter as I doubt they would want be be using too many separate alrgorithms on concurrent parts, but at this point this is nothing more than a guess. If I get further clarification, you'll know.

Thanks Dave!

Sounding promising. High quality AF... sweet! If it does have such, hopefully it has a very low performance hit. AA and AF can make a plain game have that really smooth "clean" look that is very enjoyable IMO.

 

[Inane_Dork]

On a related note, I'd like to know if the transparency AA stuff is possible on Xenos. Did I miss any discussion or article on this?

 

[Acert93]

On a related note, I'd like to know if the transparency AA stuff is possible on Xenos. Did I miss any discussion or article on this?

I believe the answer is yes, yes it can. I am 95% sure of this. So unless I am randomly remembering something wrong I am pretty sure that is correct.

My question: Can Xeno do vertex texturing? I know in the Xenos article it talks about the tesselation unit and the unified shaders being a launching point to finally do displacement mapping in games. Due to the R520 drama, I did some reading arround it seems vertex texturing tends to be importain in many displacement mapping models. Is this an issue for Xenos? If I had to make a guess, as it is unified shaders, I am guessing it has the silicon in the shader units to do this... but that is only a guess.

Edit: An Xenos AF

Each of the filtered texture units have Bilinear sampling capabilities per clock and for Trilinear and other higher order (Anisotropic) filtering techniques each individual unit will loop through multiple cycles of sampling until the requested sampling and filtering level is complete. The texture address processor has some general purpose shader ability and is able to apply offsets from the input texture co-ordinates which can be used with higher order filtering techniques. The Anisotropic filtering capabilities adapts the number of samples taken dependant on the gradient of the surface that it is sampling, which is fairly normal for Anisotropic filtering mechanisms, ATI says that the anisotropic filtering quality is improved from previous generations of hardware. As Xenos is the controller of a UMA, the entirety of system RAM is available to the texture samplers, although they will not perform any operations on the eDRAM memory.

Whatever it is, it is better than what is out on older hardwre.

 

[3dcgi]

Xenos can do vertex texturing. All unified shader architectures should support vertex texturing very well.

 

[Inane_Dork]

My question: Can Xeno do vertex texturing? I know in the Xenos article it talks about the tesselation unit and the unified shaders being a launching point to finally do displacement mapping in games. Due to the R520 drama, I did some reading arround it seems vertex texturing tends to be importain in many displacement mapping models. Is this an issue for Xenos? If I had to make a guess, as it is unified shaders, I am guessing it has the silicon in the shader units to do this... but that is only a guess.

It has to. It has SM3.0+, and VS3.0 requires vertex texturing capabilities.

And yes, to do displacement mapping on the GPU, the vertex shader must be able to read from texture.

 

[akira888]

It has to. It has SM3.0+, and VS3.0 requires vertex texturing capabilities.

Not necessarily, it seems. ATI got around this on the R5XX ASICs by enabling the "Vertex Texturing" DX cap bit and not enabling any surface formats for vertex texture maps.

But given that vertices and fragments will run on the same execution units on Xenos/C1 I see no reason not to have fully featured (trilinear, AF, all surface formats) VT.

 

[DeanoC]

R520 doesn't do vertex texturing which is a real shame. R2VB isn't a replacement, no matter what the PR claims. I'm really disappointed that R520 actually has a worst vertex processor (IMO) than the NV4x However R500 (XeGPU) has a fantastic vertex processor, in a different league compared to either of the PC chips. It has identical texturing capability to the pixel engine, so can even use DXT compression etc.

Of course given that both vendors PC chips have fairly abysmal vertex texturing caps, its just means the vertex texturing will be something left to the Dx10 generation in the PC space.

 

[Shifty Geezer]

I too would like to know for sure about transparent texture filtering. My first use of AA on a PC was a bitter disappointment because the worst jaggies weren't the least bit effected as they textures. Only I didn't understand that back then and thought all AA was supersampling. The nVidia examples of the tranparent texture filtering were a marked improvement. I don't remember any explicit mention of this feature on Xenos. I do remember some PGR grabs through a wire fence that were antialiased but don't know how that was managed.

 

[Inane_Dork]

Not necessarily, it seems. ATI got around this on the R5XX ASICs by enabling the "Vertex Texturing" DX cap bit and not enabling any surface formats for vertex texture maps.

Ack.
I wasn't aware of such hideousness.

Thanks for telling me.

 

[Acert93]

I too would like to know for sure about transparent texture filtering. My first use of AA on a PC was a bitter disappointment because the worst jaggies weren't the least bit effected as they textures. Only I didn't understand that back then and thought all AA was supersampling. The nVidia examples of the tranparent texture filtering were a marked improvement. I don't remember any explicit mention of this feature on Xenos. I do remember some PGR grabs through a wire fence that were antialiased but don't know how that was managed.

I believe this feature is the same (at least based on the discussions a few months ago in the GPU forum):

http://www.beyond3d.com/articles/xenos/index.php?p=04

So, as far as the operation is concerned, once pixel data has come through the shader array and is ready to be processed into colour values in memory the Z data of the pixel is matched with the correct colour data coming out of the shaders. Xenos supports an "Alpha to Mask" feature, which allows for the use of Multi-Sampling for sort-independent translucency

Of course I could be wrong.

 

[Shifty Geezer]

Unclear to me too. Sounds maybe right, assuming sort-independant translucency is an alpha'd texture.

 

[Titanio]

Anyone know the cost of filtering beyond Bilinear on Xenos? Some games have me wondering about the kind of filtering they're using.

 

[3dcgi]

R520 doesn't do vertex texturing which is a real shame. R2VB isn't a replacement, no matter what the PR claims. I'm really disappointed that R520 actually has a worst vertex processor (IMO) than the NV4x However R500 (XeGPU) has a fantastic vertex processor, in a different league compared to either of the PC chips. It has identical texturing capability to the pixel engine, so can even use DXT compression etc.

Of course given that both vendors PC chips have fairly abysmal vertex texturing caps, its just means the vertex texturing will be something left to the Dx10 generation in the PC space.

What specifically is bad about R520's vertex processor besides the obvious lack of vertex texturing?