AMD: R8xx Speculation

my friend in Poland is developing game around unigine engine for over a year now. Apparently unigine blokes are pleasure to deal with...
I will ask him for an update

 

Did you see the performance hit with tessellation enabled? It is huge, I went from 55fps next to the dragon to around 15-18fps enabled. Of course the demo looks ugly without it so I guess really this is a tessellation demo. very impressive though when enabled, it looks a ton better than any futuremark demo, so I guess I have to give them credit for making something that is cool looking even if it is a bit slow.

 

Someone with 5800 card to cap FPS numbers with FRAPS during the timedemo (tessellation on/off), and plot them as nice graph in Excel, please!
Or just gimme the FRAPS output dump and I'll do the Excel thing.

 

Yes, it's a pretty massive hit to performance, in addtion to Tessellation when you enable it, it's also doing Advanced SSAO on everything that's tessellated. As well as realtime lighting and shadows on tessellated surfaces.

Also, that dragon statue just has an absolutely insane number of triangles, way more than it needs by a few orders of magnitude. I think it's that way to try to stress the game engine/video cards.

Just ran it again and I must say the roof shingles at midday are particularly impressive with the lighting and shadowing.

It's also interesting to put it in wireframe and watch the dynamic tessellated LOD as you move around.

In a real game, I'd imagine they'd have to find a balance with tessellation and post processing on tessellated surfaces. But it's first and foremost a tech demo/benchmark of their Game Engine.

I wish they also showed a count of how many triangles were on screen at any particular time, as there's a LOT of triangles depending on where you are in a scene.

And it's pretty much entirely GPU bound. I don't think it used more than perhaps 50-60% cpu time on one core while I was moving about.

I wish you could create a light source in the engine demo/benchmark and move it around. As is, I'm constantly amazed watching the movement of shadows over the exaggerated cobblestone roads.

Regards,
SB

 

Directx 11 is shy. He only comes out when I take a peak, once fraps tries to look he hides.

I would have uploaded the data you wanted, but FRAPS doesnt seem to work with directx 11. Unless you know of another fraps like benchmark utility.

 

FRAPS v2.9.7 works for me under DX11 mode -- at least the FPS stats recording, the overlay FPS counter is not displayed, though.

 

What about RivaTuner or ATT?

 

Well, now at lest we can use AF with the nice bumpy surfaces and ditch that POM mapping.

 

Any reason why these shouldn't run on the 5700 series? ATI seem to only focus on 5800 series everywhere, drivers, demos etc. No mention of the apparently ignored step-child 5700 series?

 

Yeah after seeing those tech demos ATI presented (using that engine) a few weeks back, and playing Risen lately, I see all these opportunities for tessellation.

 

It seems it is a pretty heavy benchmark!

A youtube friend of mine did some tests.

Q9550 stock, ATI 5850 stock

1920X1080 2XAA DX10 no Tesselation 41 fps
1920X1080 2XAA DX11 no Tesselation 42 fps
1920X1080 2XAA DX11 + Tesselation 26 fps <-edit

It seems there's quite a big performance impact from the tesselation after all!

I thought it was like free or something! :S

 

As above in this thread, you're forgetting the SSAO!

 

Maybe the framedrop isn't caused by tesselation itself, but by SSAO, which is more complex on the tesselated surfaces.

//ed.: neliz was faster...

 

Most likely!

If I am not mistaken there's an abmient occlusion option before starting the benchmark, that can be turned off. If this would be turned off, would we have a more clear picture of the impact of tessetalion itself?

 

TANSTAAFL. Also, without knowing what they're doing, it's pointless to do too much extrapolation. Do you want to see the cost of tessellation? Use one of the samples in the DX11 SDK, or one of the samples ATI provided in their Tessellation SDK, and play with varying amplification factors.

 

Isn't SSAO based on the Z-buffer? It should not be slowed by more geometry...

Consider instead that vertex count gets an order of mangitude higher, effecting the shadow buffer renders on top of the primary framebuffer. LOTS of triangles.

 

triangle: pixel ratio might be affecting things too ?

 

Probably, although the meshes don't seem to be that dense in the wireframe parts on the youtube video. As long as a triangle is larger then 4-6 pixels, it should not cause that much trouble.

Some thorough testing could answer the question, run the demo with tesselation, but in a relatively low resolution, run it with and without SSAO, and so on... I bet there'll be a tech website with an article soon.

 

Couple benchmarks:
Q6600 @ 3GHz, ATI 5850 stock

1600x1200 0xAA 4xAF DX11 Tesselation 33.5 fps
1600x1200 8xAA 4xAF DX11 Tesselation 22 fps

 

On HD3870
Effectswise DX11 = DX10 > DX9 (naturally, since this isn't DX11 hardware)
Sppedwise DX9 > DX11 > DX10

DX9:


DX10:


DX11: