A Compact Mesh Representation for Interactive Ray Tracing

Oiram · Post by **Oiram** » Sun Sep 16, 2007 1:42 am

Ray-Strips: A Compact Mesh Representation for Interactive Ray Tracing

1) It is able to reduce up to 80% of the memory footprint over prior approaches

2) It is able to obtain performance improvements on ray tracing of large datasets due to increased SIMD utilization and more coherent memory access compared to a standard acceleration structure

http://gamma.cs.unc.edu/~lauterb/rs_techreport.pdf

Zom-B · Post by **Zom-B** » Sun Sep 16, 2007 2:25 am

This one seems to be really interesting!

The main benefit would be a higher maximal render resolution because of the reduced memory footprint...
The Speed up could be worth the pain too

Would be cool if you Ono could drop an eye on this and tell us your thoughts about this technique!

Thanks for the paper Oiram!!

Post by **OnoSendai** » Sun Sep 16, 2007 3:24 am

Looks interesting.
Their data structure looks like it takes a while to build, tho, and the results they give actually make it look slower in terms of rendering peformance in many cases.

Kram1032 · Post by **Kram1032** » Sun Sep 16, 2007 3:36 am

so, for now, the simple KD-tree is better?

And what's about that one as a BIH-replacement? would that be better?

Post by **OnoSendai** » Sun Sep 16, 2007 3:24 pm

I wouldn't call the kd-tree simple

I think kd-tree will be faster than their technique.

Kram1032 · Post by **Kram1032** » Sun Sep 16, 2007 11:51 pm

simple is the wrong word, yeah...

but you didn't answer my question^^

I asked, whether it's slower than BIH, or not...
that's not the Kd-tree, is it?

Post by **OnoSendai** » Sun Sep 16, 2007 11:52 pm

Well, it might trace faster than the BIH.
But one of the nice things about the BIH is the super-fast build time, that this new method doesn't have.

Kram1032 · Post by **Kram1032** » Sun Sep 16, 2007 11:54 pm

yeah, they mentioned that.

good point.