AMD's Radeon HD 6870 benchmarked? (updated more screens)

[RPGWiZaRD]

AMD needs to start getting their logos in a few games now.

My thoughts exactly. Now would be a good time to do it, something corresponding to NVIDIA's "TWIMTBP". They've already had quite an upswing with HD 5xxx series when it comes to cooperating with game devs (well they were way ahead with DX11 support so no suprise), with HD 6xxx they could gain trust even easier.

[Florinmocanu]

Well, one thing saaya.

tesselation performance is not directly correlated with resolution when doing a unigine test, but.

Rendering is done per pixel. More pixels = more area to render, which means more shading/displacement/geometry etc....

I can tell you about normal CGI rendering, which shares a lot of similarities.

Displacement takes quite a lot of time to render and when the resolution is smaller, rendering goes faster. If you leave the same displacement and up the res, you increase linearly the rendering time, because the objects being displayed/rendered share a bigger pixel space. Displacement is done on screen, not for the whole scene, it would be crazy to do it like that.


So, resolution increase means also lower performance, even in Unigine.

[deeperblue]

I can tell you about normal CGI rendering, which shares a lot of similarities.

Displacement takes quite a lot of time to render and when the resolution is smaller, rendering goes faster. If you leave the same displacement and up the res, you increase linearly the rendering time, because the objects being displayed/rendered share a bigger pixel space. Displacement is done on screen, not for the whole scene, it would be crazy to do it like that.

Well... CGI rendering is actually quite different from the graphics cards rendering pipeline.
Take a look at this:
http://i.msdn.microsoft.com/Ff569022...s,VS.85%29.png
From Input Assembler until after the Geometry Shader everything is done resolution independent! (as somebody mentioned assuming same aspect ratio) The Rasterizer and Pixel Shader are then operating on the actual pixel level of the current rendering target.

[Florinmocanu]

Well... CGI rendering is actually quite different from the graphics cards rendering pipeline.
Take a look at this:
http://i.msdn.microsoft.com/Ff569022...s,VS.85%29.png
From Input Assembler until after the Geometry Shader everything is done resolution independent! (as somebody mentioned assuming same aspect ratio) The Rasterizer and Pixel Shader are then operating on the actual pixel level of the current rendering target.

Nope. In CGI effects are calculated resolution independent as well, but then to display them, it takes more time at higher resolution than lower one. Displacement is the best example for this, it is done usually with physical scale in mind (the amount the model is tessellated (1 polygon every 1 mm or 3 mm or 5etc..) determines how detailed the model will be after the displacement is done). So, even if i render at 640x480 or 1024x768, the displacement is done independently on the model.

But, doing displacement at 1024x768 usually gives linear increases in the time of rendering vs 640x480. That's why a lot of times i avoid using displacement, because it's a resource hog.

[deeperblue]

Nope. In CGI effects are calculated resolution independent as well, but then to display them, it takes more time at higher resolution than lower one. Displacement is the best example for this, it is done usually with physical scale in mind (the amount the model is tessellated (1 polygon every 1 mm or 3 mm or 5etc..) determines how detailed the model will be after the displacement is done). So, even if i render at 640x480 or 1024x768, the displacement is done independently on the model.

But, doing displacement at 1024x768 usually gives linear increases in the time of rendering vs 640x480. That's why a lot of times i avoid using displacement, because it's a resource hog.

And that's exactly the point.
The DX11/OpenGL tesselation we are talking about are spawning millions of new polygons in the middle of the pipeline - independent of the rendering resolution.

Your CGI renderer most likely is implementing displacement mapping in a sub-pixel displacement way without generating tons of micro-polygons like we have in the GPU pipeline - so the renderer is resolution dependent.

[Dami3n]

I don´t know if all the leaked numbers are fake, but real performance must be very good. Otherwise i don´t think that AMD would kill the Ati brand.

[Iconyu]

I don´t know if all the leaked numbers are fake, but real performance must be very good. Otherwise i don´t think that AMD would kill the Ati brand.

Interesting theory, AMD won't have a flop for their first AMD GPU...

[Florinmocanu]

And that's exactly the point.
The DX11/OpenGL tesselation we are talking about are spawning millions of new polygons in the middle of the pipeline - independent of the rendering resolution.

Your CGI renderer most likely is implementing displacement mapping in a sub-pixel displacement way without generating tons of micro-polygons like we have in the GPU pipeline - so the renderer is resolution dependent.

I said physical scale. So, i can say to the model, i want a polygon every mm of the model (that implies working the model up to scale). So... if the model will be distant then i use some low tessellation on it, if the model is close to the camera than i will input a higher value to get more detail.

Still, even then resolution plays a huge role into the render time since the actual displacement is done through the shader, through a black and white map (exactly how you do it in DX11 as well). So, more pixels = more space to shade/displace.

Remember, tessellation means you divide the polygons of the model into more polygons.
Displacement means you then use a black and white map on the tessellated model and you displace extra geometry via shader = which means more render time when you increase the resolution. especially when you use higher AA (you have that in CGI as well, you can use low AA or high AA, depending on the scene or desired render time/quality).

Also, shading is done usually at a sub-pixel level, because you use an AA filter for subpixel transitions (better, crisper renders), like Catmull-Rom for example, so as you up the resolution and AA levels, the displacement takes longer to do.


I can go on about this forever, but i think you get the picture. There is a clear distinction between tessellation (done via geometry shader in DX11) and displacement, done after that.