ATI Radeon HD 2000 Series Real-Time Demos

[Dave Baumann]

Yes, it was, I didn`t state anything to the contrary. What I meant is that Gabe will do anything to please you, and he has been doing that for a long time...

AFAIK they had the demo prior to us signing for the "Black Box" to be bundled with HD 2900.

 

[nyr]

Even if underlying hardware would count for nothing, if the API is making calls for the tesselator, not having one would still be no-go.

I'm not sure I understand what you're saying here. Are you suggesting that any game that uses the tesselator simply won't run on non-ATI hardware? In that situation any PC port of that game would appear to be commercial suicide.

 

[Jawed]

I'm not sure I understand what you're saying here. Are you suggesting that any game that uses the tesselator simply won't run on non-ATI hardware? In that situation any PC port of that game would appear to be commercial suicide.

No, it would just lose an IQ feature. Much like XBox 1 shadowing wouldn't work on ATI graphics cards when the game was ported to PC. Wasn't that what happened?

Jawed

 

[nyr]

No, it would just lose an IQ feature. Much like XBox 1 shadowing wouldn't work on ATI graphics cards when the game was ported to PC. Wasn't that what happened?

Jawed

Isn't it underplaying the potential of the tesselator to suggest that it's just an IQ feature? Doesn't that relegate it to being the equivalent of an Aegia Physix card, which also had potential but doesn't seem to have got further than adding a few extra particles to explosions?

Or am I overestimating the capabilities of the tesselator in terms of the generated geometry playing some part in gameplay?

 

[trinibwoy]

Or am I overestimating the capabilities of the tesselator in terms of the generated geometry playing some part in gameplay?

Based on the demos it looks like just another IQ feature like normal mapping or soft shadows. It's not going to affect how you interact with the environment. But if used properly it could probably have a dramatic impact on visuals.

 

[Razor1]

Based on the demos it looks like just another IQ feature like normal mapping or soft shadows. It's not going to affect how you interact with the environment. But if used properly it could probably have a dramatic impact on visuals.

yeah pretty much I know we have been asking for a tesselator for the past 4 years from both ISV's, the heightmaps used for parrellax bump maps can be used for tesselation and displacement, but, there is always a but lol, memory usage, and burden on the GPU, and bandwidth get to a pretty hefty for really getting all the fine bumpping detail into actual mesh detail, haven't tried anything on the r600, Dave gave a number of 2 billion triangles per frame?, but I guess that would be mean the tesselated objects can't be used for anything other then pure visuals I pressume.

 

[KimB]

Well, presumably you can make use of the geometry shader for tessellation as well. I am curious as to what the differences are here.

 

[Zvekan]

Well, presumably you can make use of the geometry shader for tessellation as well. I am curious as to what the differences are here.

If I remember correctly, AMD showed demos of R600 tesselating a model to approximately 15 mil. triangles with framerates exceeding 100-200 fps... Numbers could be higher, but I'm not certain anymore.

The good thing about tesselation is that models don't take additional space, everything is generated on the fly - while the model is passing through pipeline.

Using geometry shader, you could be forced to write results somewhere as it can output only 1024 vertices for one input, while tesselator (present in R600) can tesselate model up to 15 x.

 

[KimB]

Using geometry shader, you could be forced to write results somewhere as it can output only 1024 vertices for one input, while tesselator (present in R600) can tesselate model up to 15 x.

I'm failing to understand how only being able to write 1024 vertices for one input wouldn't allow one to tessellate a model by 15x or more.

 

[Zvekan]

I'm failing to understand how only being able to write 1024 vertices for one input wouldn't allow one to tessellate a model by 15x or more.

15x isn't multiplying starting vertices by 15 - it recursive tessellizationon done 15 times. Amplification is much greater that 15x in number of vertices as far as I understand.

Try tessellating in 3ds max or some similar programs - you'll see the explosion in number of vertices.

Zvekan

EDIT:

For a standard box (8 vertices) 8x tessellization (2 x 4 iterations) results in nearly 400k vertices in 3ds max

 

[KimB]

15x isn't multiplying starting vertices by 15 - it recursive tessellizationon done 15 times. Amplification is much greater that 15x in number of vertices as far as I understand.

Well, if it's recursive, then yes, it would be. But what would such massive recursive tessellation be useful for?

 

[Zvekan]

Well, if it's recursive, then yes, it would be. But what would such massive recursive tessellation be useful for?

That question is for developers to find, but I think even 4x is more than enough for games...

The advantage in having hardware tessellator is in storage requirements also - it works on fly without occupying additional space in memory.

Higher resolution models via tessellization + displacement maps can give much better visual results. I don't think that G80 GS is capable of giving same, or even similar performance as one demonstrated with R600 during launch.

Will developers use it as only one IHV supports it, remains to me seen. If history is indicative of something, I doubt that it will be used - unfortunate thing for AMD.

 

[AlexV]

"You"? I thought you were talking about Gabe's apparent obsession with ATI, not Dave.

Bah, you=ATi. My communication skills seem to be going down the drain due to the lack of sleep I`ve been experiencing these days. Goddamn