ATi Releases More Xenos Info On Website

[Hardknock]

http://www.ati.com/developer/eg05-xenos-doggett-final.pdf

Notes:

The Architect of the Xenos is ATI's Michael Dogett

- Features of System Architecture is New Rendering Performance and new GPU Architecture

- The Features are Unified Shader, Memory Export, Texture/Vertex Fetch, HDR Rendering, Displaced Subdivision Surfaces

- Features of the Daughter die (EDRAM) is Spix/Clix, 4xMSAA, Stencil and Z Test Alpha Blending which is done with the 10 MB EDRAM

- The Rendering Performances are 1) Daughter to GPU: 8 pixels/clk that means 32 BPP Color and 4 Samples Z lossless compression, 16 pixels/clk that means 4 Samples Z Lossless Compression and 2) Alpha and Z Logic to EDRAM interface: 256 GB/second, 32 Samples x 32 Bit color, 24 Bit Z and 8 bit stencil as well as Double Z which means 64 Samples x 24 bit z and 8 bit Stencil.

- Unified Shader means: A revolutionary Step in Graphics hardware, one Hardware design which performs both Vertex and Pixels and has Vertex Processing Power. Previous design was First Vertex was processed, then Pixels were Processed, Now both will be Processed at the same time thus increasing efficiency. It has GPU based vertex and pixel load balancing which means between vertex and pixel resource usage. It is Based on Directx 9, Shader Model 3.0+ .

- Memory Export means: Shader Output to computed address, Virtualise Shader Process by Multipass, Shader Debug and Scatter write. The Memory Export is from the GPU to the Unified 512 MB GDDR3 Ram. Randomly update data structures from Vertex or Pixel Shader which means Ray tracing acceleration structures, Physical simulation – GPGPU and Enabling exploration for the future

More in the slides

 

[expletive]

Aside from putting an end to the NEC bandwidth debate, any new info here?

J

 

[MonkeyLicker]

"Aside from putting an end to the NEC bandwidth debate, any new info here?"

So all is as it should be in 360 land?
I can't get the pdf to display correctly on this PC, so I just need the ultimate answer concerning the.....concern.

 

[Inane_Dork]

I'm certainly not well-versed in tesselation methods, but the one described in there sounds different than the one described in B3D's Xenos article. I found this link which has a lot of Stam's work. Maybe it'll help somebody.

 

[Acert93]

I'm certainly not well-versed in tesselation methods, but the one described in there sounds different than the one described in B3D's Xenos article. I found this link which has a lot of Stam's work. Maybe it'll help somebody.

It does seem there may be more than one way to skin the cat, errr Displacement Mapping, on Xenos it seems. Dave had mentioned the tesselation unit in his article (which is quite fast at 1 vertex every 2 clocks, i.e. 250M triangles/sec) but the Shader units also have texturing ability so all 48 ALUs could be dedicated to processing vertex texturing for such effects in a reasonable time.

The picture in the PDF, while simple, gives a good idea of how displacement maps could be used in games. I am looking forward to devs getting to play with this stuff

 

[Hardknock]

I'm curious, does anyone know what else Michael Doggett has designed?

 

[expletive]

I think we're just beginning to understand the flexibility of this architecture. Theres a slide about physical simulation/GPGPU as well.

J

 

[pakpassion]

I'm curious, does anyone know what else Michael Doggett has designed?

check his webpage: http://www.gris.uni-tuebingen.de/~miked/

Michael Doggett

ATI Research

62 Forest Street

Marlborough, MA 01752

(508) 303-3900 x3863 (o)

MDoggett@ati.com



Biography:Michael Doggett works as an architect on graphics hardware at ATI Research. He completed his B.S. degree in Computer Science in 1990, B.E. degree in Electrical Engineering in 1992, and Ph.D. in 1997 all at the School of Computer Science and Engineering at The University of New South Wales, Sydney, Australia. From 1996 to 1998 he worked as Chief Engineer at Conja Pty Ltd, a Special Effects, Animation and Design company. From 1998 to 2001 he was a member of the research staff of the Computer Graphics Laboratory (GRIS) at the Computer Science Department of the University of Tuebingen as a PostDoc where he worked on custom hardware for Volume Rendering and Displacement Mapping. He has been involved in teaching courses at the University of New South Wales and the University of Tuebingen. He is the paper co-chair for Graphics Hardware 2002 and has served on the program and review committee for serveral conferences. He has published numerous papers and is a member of IEEE Comuter Society, and ACM.

 

[AlphaWolf]

I'm curious, does anyone know what else Michael Doggett has designed?

The main focus of my research at WSI/GRIS was Computer Graphics Hardware. I worked on the VIZARDII project for hardware accelerated Volume Rendering, and displacement map rendering using traditional polygonal rasterization hardware techniques.

From here

 

[mozmo]

Does anyone know if the improvements in branching that are seen with R520 are also present in the xenos. ie does xenos have fast branching? I believe going forward for complex shaders fast branching will be a big performance.

The rest of the achitecture looks amazing, very impressed and definately way ahead of what we have on the desktop now in terms of shading power and flexibility.

 

[Inane_Dork]

I don't think Xenos' ALUs are as efficient as the R520's at branching. The batch sizes are much larger on Xenos which increases wasted cycles for at least some tasks.

 

[nAo]

Xenos works on 64 pixels wide batches AFAIK

 

[Shifty Geezer]

Aside from putting an end to the NEC bandwidth debate

This .pdf was published before the NEC article. It doesn't put an end to it as we still don't know who's figure is right, NEC's or ATi's. Consensus is that the NEC figure was wrong, but nothing's confirmed.

Regards article, I did like the Displaced SDS! Not sure how well it'll work in motion using a texture to specify geometry as the texture will warp with UV's, causing bending structures that should remain inflexible and the like. It'd be nice to hear Laa-Yosh's opinion on how useful this'll be for things like characters or inanimate tyres and scenery and bits. Maybe KZ's tyres aren't in the realm of impossibility?

 

[therealskywolf]

This .pdf was published before the NEC article. It doesn't put an end to it as we still don't know who's figure is right, NEC's or ATi's. Consensus is that the NEC figure was wrong, but nothing's confirmed.

Regards article, I did like the Displaced SDS! Not sure how well it'll work in motion using a texture to specify geometry as the texture will warp with UV's, causing bending structures that should remain inflexible and the like. It'd be nice to hear Laa-Yosh's opinion on how useful this'll be for things like characters or inanimate tyres and scenery and bits. Maybe KZ's tyres aren't in the realm of impossibility?

Lol at people wishing it is true. Anyway cool find.

 

[expletive]

This .pdf was published before the NEC article. It doesn't put an end to it as we still don't know who's figure is right, NEC's or ATi's. Consensus is that the NEC figure was wrong, but nothing's confirmed.

Yes but i believe the pdf is dated AFTER the 360 was finalized and went into mass production. I'd like ot believe if there had to be a change in spec, ATI would have corrected this information before they put it up on their website. I agree nothing is officially confirmed but imo theres enough evidence that it was someone misinterpreting BW diagrams at NEC or there was a typo. (Honestly i just need to try and put an end to the rumor becuase its killing my buzz )


J

 

[Shifty Geezer]

theres enough evidence that it was someone misinterpreting BW diagrams at NEC or there was a typo.

I and most other people agree it seems. As for your buzz there's no reason to be disappointed, as even if true it wouldn't make any difference to your experience. Without any idea what the numbers for components in my PS2 might be or may have changed from original intentions, I was still able to enjoy the games on it.

Remember the old mantra, - It's the games that matter.

 

[expletive]

I and most other people agree it seems. As for your buzz there's no reason to be disappointed, as even if true it wouldn't make any difference to your experience. Without any idea what the numbers for components in my PS2 might be or may have changed from original intentions, I was still able to enjoy the games on it.

Remember the old mantra, - It's the games that matter.

Agreed. I just want all these new consoles 'to be all they can be'. I dont like any of them being specced down becuase at some point i cant help but feel that it lowers the ceiling on potential.

J

 

[dukmahsik]

anyone see ray tracing on one of the slides?

 

[Carl B]

ATI's R520 Toy Store slides indicated a sort of 'per pixel' ray-tracing function being run on certain elements in the scenes, could be something similar they are talking about. It caught my attention though in both those slides and in this more recent R500-related info.

 

[Shifty Geezer]

The raytracing comment seemed to be raytracing assist.

MEMEXPORT
Randomly update data structures from
Vertex or Pixel Shader
• Ray tracing acceleration structures
• Physical simulation – GPGPU

These are the areas it's been mentioned before where MEMEXPORT can be used to have the GPU perform non rasterizing tasks.​

I think it's primarily an optional capability for things like reflections. It's certainly not talking about hardware implementation of one of our beloved realtime raytracing engine pipedreams!​