[PixResearch]

Imagination Technologies purchased Caustic.

[rpg.314]

Originally Posted by hoho

We hope someone figures out some new algorithms by then. It's kind of similar to how people talk about simulating brain. "In N years we build a computer having same processing power of brain! (but we have no clue how to simulate it anyway)".

We can hope. But hope is not a policy.

[rpg.314]

Originally Posted by Rodéric

Caustic sold hardware accelerating Ray Tracing already, and Imagination Technologies is licencing the IP...

To whom, for which market, when are the products going to be out?

[Mintmaster]

Originally Posted by rpg.314

In 6 years, 100W GPUs of today will shrink to 10W GPUs then. If we can't do real time ray tracing today, why will we be able to do it then?

For one, we're going to have better software. straaljager pointed out HLBVH2 as building a raytracing structure for a few million polys in realtime today. We'll also see more compute oriented architectures in one form or another.

Maybe I was too optimistic, but I think we'll at least see hybrid rendering by then. You'll have a separate structure for raytracing to do secondary lighting and composite it with higher resolution rasterized details.

If moore's law helps ray tracing so much, then why would it not help rasterization too?

The problem is that 1000fps isn't much better than 100fps for rasterization. The only thing left for photorealistic realtime rendering, IMO, is secondary lighting, and you need hacks like AO to do it with rasterization. If artists/programmers can make rasterization good enough, then transition will be pushed out further, but eventually it'll just be easier to switch to raytracing.

[rpg.314]

Originally Posted by Mintmaster

For one, we're going to have better software. straaljager pointed out HLBVH2 as building a raytracing structure for a few million polys in realtime today. We'll also see more compute oriented architectures in one form or another.

Maybe I was too optimistic, but I think we'll at least see hybrid rendering by then. You'll have a separate structure for raytracing to do secondary lighting and composite it with higher resolution rasterized details.

The problem is that 1000fps isn't much better than 100fps for rasterization. The only thing left for photorealistic realtime rendering, IMO, is secondary lighting, and you need hacks like AO to do it with rasterization. If artists/programmers can make rasterization good enough, then transition will be pushed out further, but eventually it'll just be easier to switch to raytracing.

HLBVH2 is nice. But without dedicated hw, it would be tough to replace rasterization. If it was a competition between sw rasteriztion and sw ray tracing, the difference would be much less. Besides, there remains the problem that the ff hw for rasteriation (like for AA) keeps growing, and the effects we want to add keep growing.

For secondary rays, we are looking at a problem that is needs a big break through, the kind we haven't had yet in several decades. I am not sure that throwing more flops, more bw and more caches is the solution to it when these things benefit rasterization (and rasterization based effects) just as much as ray tracing.

As for AO, there exist rasterization based methods (ambient occlusion volumes) that match ray traced AO at a fraction of the cost. I think there is a DICE presentation showing that they are considering it (have they implemented it?).

It's like chasing a moving goal post. The laggard is not gaining on the leader and the goal post is not moving towards the laggard either. How is catching up even possible?