AMD Radeon HD 6970 already benchmarked? Enough to beat GTX480 in Tesselation?

[Scali]

Your probably both saying the same thing and speaking past one another.

I think what confuses him is that I'm not on nVidia's side. The GeForce FX was an abomination of a DX9 architecture, as I have always said.

 

[jones377]

In theory they could do that, but it would be cheating.
Namely, as AMD themselves promote tessellation: it should be adaptive.
That means that the level of tessellation is determined at runtime by parameters such as the distance or the projected size on screen.
This can (and generally will) change at every frame, so trying to buffer the geometry is pretty useless.... Unless ofcourse you're going to cheat and not actually going to do adaptive tessellation every frame, but just every X frames, re-using the buffered geometry for the remaining X-1 frames, rather than generating a proper adaptive set of geometry.

I don't mean they would break compatibility with DX11 tesselation code. But I'm curious, if you use adaptive tesselation, does the geometry grow linearly with the inverse distance or is it more like a staircase looking growth? If it's the latter, then buffering could still work since we'd be rendering at approx 60 frames per second. That's very little forward movement in a single frame. It could still wreck the minimum framerate however, anytime the camera makes too sudden movements or shifts to a new position.

 

[Scali]

I don't mean they would break compatibility with DX11 tesselation code. But I'm curious, if you use adaptive tesselation, does the geometry grow linearly with the inverse distance or is it more like a staircase looking growth? If it's the latter, then buffering could still work since we'd be rendering at approx 60 frames per second. That's very little forward movement in a single frame. It could still wreck the minimum framerate however, anytime the camera makes too sudden movements or shifts to a new position.

Tessellation is done with fractions, not with exact triangles (fractional-even/odd algorithms).
So every movement, ever so small, will result in the triangles at least reshaping, if not being added/removed.
So yes, any kind of buffering will break compatibility with DX11. Which is why I said it would be cheating.

 

[blastingcap]

Kitguru is predicting that Crysis 2 will feature "copious" amounts of tessellation courtesy of NV's $2MM investment in the game--enough to push even the GTX580 to the limit: http://www.kitguru.net/components/g...2-being-re-designed-for-gtx580-expect-delays/

 

[Genx87]

Kitguru is predicting that Crysis 2 will feature "copious" amounts of tessellation courtesy of NV's $2MM investment in the game--enough to push even the GTX580 to the limit: http://www.kitguru.net/components/g...2-being-re-designed-for-gtx580-expect-delays/

Sounds wonderful!

 

[Skurge]

Sounds wonderful!

If it actually makes the game look better and doesn't have diminishing returns, but I would expect that from you.

 

[Genx87]

If it actually makes the game look better and doesn't have diminishing returns, but I would expect that from you.

Expect what from me? Like a game making use of a major component of DX11? We want DX11 games right?

 

[Skurge]

Expect what from me? Like a game making use of a major component of DX11? We want DX11 games right?

Sorry. I was cross posting. That last part shouldn't have been there. I didn't meant anything by it.

 

[RussianSensation]

Kitguru is predicting that Crysis 2 will feature "copious" amounts of tessellation courtesy of NV's $2MM investment in the game--enough to push even the GTX580 to the limit: http://www.kitguru.net/components/g...2-being-re-designed-for-gtx580-expect-delays/

It may very well be that the biggest catalyst for tessellation will actually be consoles in 2012.

When NV and AMD both continue to increase tessellation in their GPUs, why didn't EA push tessellation in Crysis 2 in the first place since it's supposed to be "next generation" engine? If EA already did have "normal" tessellation and NV is paying to include some "extreme" variation of this, then it's going to require 3x GTX580s to run . Let's hope these tessellation differences actually improve the visuals rather than produce a massive performance hit as they usually do; and the game doesn't have more delays.

 

[OCGuy]

Kitguru is predicting that Crysis 2 will feature "copious" amounts of tessellation courtesy of NV's $2MM investment in the game--enough to push even the GTX580 to the limit: http://www.kitguru.net/components/g...2-being-re-designed-for-gtx580-expect-delays/

Sounds good to me!

I'm glad somebody is pushing developers to increase graphics standards.

 

[Skurge]

Sounds good to me!

I'm glad somebody is pushing developers to increase graphics standards.

No body needs to push CryTek to do anything, they do a pretty good job of pushing graphics standard to the limits of what current tech can do. They actually do a better job of going over them.

 

[Lonbjerg]

Expect what from me? Like a game making use of a major component of DX11? We want DX11 games right?

Didn't you get the memo?

"We" only want AMD DX11 level games :whiste:

 

[PingviN]

Didn't you get the memo?

"We" only want AMD DX11 level games :whiste:

If we get Metro-style tesselation, I'd rather be without. Either implement tesselation properly, or don't bother at all.

But then, I seriously doubt you care whether it looks better or not, as long as the game runs better on Nvidia GPUs.

*facepalm*

 

[Lonyo]

Didn't you get the memo?

"We" only want AMD DX11 level games :whiste:

You mean we only want games that no card can play with all the options on?
The fastest single GPU at the moment can only manage 50fps at 1920x1200 with 4xAA and most settings on level 3 of 4 (Gamer) in terms of quality.

You think adding lots of tessellation is going to enable ~any~ single GPU to give playable framerates?
If they do add an 'extreme' level, it would probably be dual GTX580s only that would be able to run it without chugging horribly.

 

[Skurge]

You mean we only want games that no card can play with all the options on?
The fastest single GPU at the moment can only manage 50fps at 1920x1200 with 4xAA and most settings on level 3 of 4 (Gamer) in terms of quality.

You think adding lots of tessellation is going to enable ~any~ single GPU to give playable framerates?
If they do add an 'extreme' level, it would probably be dual GTX580s only that would be able to run it without chugging horribly.

Which is exactly what he wants.

 

[Skurge]

AMD DX11 level games

What the hell does this even mean?

 

[Lonyo]

What the hell does this even mean?

I think it means playable level.
Also apparently tessellation is the only DX11 feature in existence.
Not like Depth of Field in Metro 2033 kills framerates or anything (reducing by 1/3rd is nothing).


Which is actually another thing. It will be interesting to see if the new architectures (going back to the original topic) bring anything to other DX11 features, other than just tessellation.
When something like DOF can kill FPS by a lot, will they make improvements in other areas, or will that just come with a 'better'/bigger architecture/chip?

 

[Skurge]

I think it means playable level.
Also apparently tessellation is the only DX11 feature in existence.
Not like Depth of Field in Metro 2033 kills framerates or anything (reducing by 1/3rd is nothing).


Which is actually another thing. It will be interesting to see if the new architectures (going back to the original topic) bring anything to other DX11 features, other than just tessellation.
When something like DOF can kill FPS by a lot, will they make improvements in other areas, or will that just come with a 'better'/bigger architecture/chip?

Im more interested in DirectCompute. AMD have already used it for MLAA. With sometimes works great, looks like AA with no performance hit. CIV5 also uses it, but the only thing people see is tessellation.

 

[AtenRa]

I believe that we don’t need any form of Anti-Aliasing when we have Tessellation.

 

[Keysplayr]

I believe that we don’t need any form of Anti-Aliasing when we have Tessellation.

And why is that?

 

[AtenRa]

Well, I believe that if the entire screen is being tessellated in High Resolutions at 1920x1080 and up, then the need for Anti-Aliasing is very small. The polygon count is getting so much denser and graphics look smoother, so AA do not contribute allot more.

If you run Alien vs Predator benchmark at 1920x1080, the Alien models don’t get a lot from AA because they already being Tessellated. Only the environment gets better with AA because its not Tessellated.

Have a look at those pics bellow, both at 1920x1080 same settings. First pic no AA, the second pic is with 4x AA. Can you see a deference in the Alien model ?? the difference is very small.
We can see smoother surfaces in the environment, in the pipes, the window etc etc.

No AA 16xAF

4x AA 16x AF

 

[Keysplayr]

So, you "believe" that one of Tesselation's purposes is to eliminate Jaggies? You don't see any difference in those two shots you just provided with and without AA? I'm starting to "believe" that this is just a "theory" of yours instead of an astute observation.

 

[AtenRa]

There is a very small difference with 4x AA in the Alien model but not that much to justify the performance penalty and its only visible in still pictures.

I believe in future games with better Tessellation the need for AA will be even smaller.

 

[vshin]

I'm not totally sold on tesselation replacing AA. Not sure why they can't coexist.

I will agree that AA has a relatively low performance/benefit ratio though so it's the first thing I turn off to keep my minimum fps as close to 60 as possible.

 

[BathroomFeeling]

Tessellation needs to be in the subpixel range throughout the entire screen before it can successfully and completely replace AA.