UE3 Engine Favor Dyanamic Branching or Vertex Fetch

[memberSince97]

Which way is this engine going or is it flexible enough to make everyone happy?


Looking at SCCT UE2.5+ ATI's do well but previous implemtations are a mixed bag.

Does anyone have a opinion or will this be a flexible engine or CPU vertex ruling monster.

pardons if this is the wrong forum.

 

[Reverend]

Would it matter either way?

 

[memberSince97]

It matters to big business IHV's , benchmark whores, enthusiasts aka fanbboys ,but not general consumers. Except the sticker on the box they read.

 

[Rys]

Which way is this engine going or is it flexible enough to make everyone happy?


Looking at SCCT UE2.5+ ATI's do well but previous implemtations are a mixed bag.

Does anyone have a opinion or will this be a flexible engine or CPU vertex ruling monster.

pardons if this is the wrong forum.

It took me about 10 reads to understand that you want to know if UE3 will run faster on ATI or NVIDIA hardware. Should have just said so.

And Rev is right, it doesn't really matter just now.

 

[Bouncing Zabaglione Bros.]

Which way is this engine going or is it flexible enough to make everyone happy?


Looking at SCCT UE2.5+ ATI's do well but previous implemtations are a mixed bag.

Does anyone have a opinion or will this be a flexible engine or CPU vertex ruling monster.

pardons if this is the wrong forum.

Looking at what Epic are trying to achieve, you should expect everything and the kitchen sink. Whether the game developers will use those features, or whether hardware will be fast enough to make them usable in a game is another question altogether.

 

[Mintmaster]

memberSince97, those are two very different, independent features, so it makes no sense to say they'll go one way or the other.

PS branching is something they've already demonstrated on NV40 with soft shadows. I think that if you want high quality shadows, ATI will be the way to go, oddly enough. Remember, dynamic branching is in essence a speed optimization (sometimes a dramatic one), and they can still run the same pixel shader code on PS2.0 hardware. I would be very surprised if they don't use this quite heavily in UE3.

VTF, on the other hand, is a special feature handy for displacement mapping and various eye-candy simulations of cloth, water, and particles, which I absolutely love. Right now, it really hammers NVidia's vertex throughput (please, someone give me numbers! :smile, so it's likely that Epic will do such things on the CPU, IMO, and if they want to offload to the GPU, then even on NVidia hardware they'll get better performance using render to vertex buffer, assuming they support it.

 

[Demirug]

Mintmaster, nVidia claims 33M fetchs/s with one fetch per vertex for a 6800 Ultra.

 

[Mintmaster]

Thanks, Demirug. That's basically GF2 transform rates.

I guess that means since the 6800 has 6 vertex shaders at 400 MHz, then a texture access costs 70 (!) instructions or so. Weren't they saying 20-30 cycles latency in a pdf?

I suppose that's still not that bad. If your game runs at 100 fps, then adding a 200,000 vertex mesh with one VTF would reduce framerate to 60fps. I think a simple physics simulation on a 3GHz CPU could easily be faster, though, especially if it's sitting idle as it often does at high resolutions.