nvidia rtx perf.

Practical and theoretical implementation discussion.
Post Reply
mpeterson
Posts: 58
Joined: Fri Jan 06, 2012 3:09 pm

nvidia rtx perf.

Post by mpeterson » Thu Sep 05, 2019 3:29 pm

as the discussion is going on, day by day, here is a statement: 10grays is a tech. possible limit for non-realistic scenes: that is: a single triangle
without shading. in practice you get something around 1-3 grays and cannot do something useful with it without any filtering as post-proc.
these numbers were already achieved years ago (7 to 8 years) with our highly optimized ray-tracers => so there is really nothing new to the problem today.
the quality everyone expects needs much more compute-power ! end of statement.

jbikker
Posts: 220
Joined: Mon Nov 28, 2011 8:18 am
Contact:

Re: nvidia rtx perf.

Post by jbikker » Thu Sep 05, 2019 5:31 pm

:)

Well the RTX hardware does help, but the 10G figure is not realistic, unless you store only intersection distances for the rays. I currently store barycentrics, t, prim index and instance index, which peaks at 4G on 2080, probably 5G on 2080ti. That is without any shading, like you said, but it still allows me to render roughly 4x faster than before overall. That is mostly because my shading code is built for realtime, i.e. one ubershader with limited functionality (Disney). For a production renderer I imagine the relative gains are smaller, except if the pre-RTX ray tracing was bad.

4G is using the latest Lighthouse 2, https://github.com/jbikker/lighthouse2 .

mpeterson
Posts: 58
Joined: Fri Jan 06, 2012 3:09 pm

Re: nvidia rtx perf.

Post by mpeterson » Sat Sep 07, 2019 2:16 pm

jbikker wrote:
Thu Sep 05, 2019 5:31 pm

4G is using the latest Lighthouse 2, https://github.com/jbikker/lighthouse2 .
yes, i had a look and browsed thru the code. good work as always. mp

Post Reply