Show-off, reference material & tools.
-
mpeterson
- Posts: 59
- Joined: Fri Jan 06, 2012 3:09 pm
Post
by mpeterson » Thu Sep 05, 2013 9:18 am
hallo, we just released our tracing kernels for intel phi. enclosed a simple test scene. it runs at
1.2 grays/s on latest 7xxx series ! btw: it is also around 2x faster than intels embree 2.0.
maik
-
voxelium
- Posts: 15
- Joined: Fri Apr 27, 2012 11:10 am
-
Contact:
Post
by voxelium » Thu Sep 05, 2013 10:53 am
Impressive performance! Are your kernels publicly available somewhere?
-
graphicsMan
- Posts: 167
- Joined: Mon Nov 28, 2011 7:28 pm
Post
by graphicsMan » Thu Sep 05, 2013 1:20 pm
Interesting. Do you have numbers for diffuse interreflections?
-
straaljager
- Posts: 24
- Joined: Thu Dec 01, 2011 9:45 pm
- Location: Switzerland
-
Contact:
Post
by straaljager » Fri Sep 06, 2013 1:44 am
Looks like Whitted ray tracing, not even ao. Are you sure this quality is still relevant today?
-
jbikker
- Posts: 225
- Joined: Mon Nov 28, 2011 8:18 am
-
Contact:
Post
by jbikker » Fri Sep 06, 2013 8:17 am
Regardless of ray distribution, I think 1.2G is amazing, and it's very cool that you managed to break the 1G barrier! Wald wrote in his slides for a SIGGRAPH course a few years ago 'ray tracing performance is expressed in *millions of rays*, not thousands'. Looks like this has to be adjusted for today.

Also, for that kind of scenery, a Titan can perhaps reach 500M, so the phi appears to be vastly better for ray tracing. That's a good thing if you want to sell a new architecture.
-
mpeterson
- Posts: 59
- Joined: Fri Jan 06, 2012 3:09 pm
Post
by mpeterson » Mon Sep 09, 2013 9:11 am
@graphicsman: not yet. more "in-deep" test for any kind if distribution will come in the next weeks.
just ordered a titan to see and compare different scenarios. pt results as soon as it arrives.
this is a nice research project. we have to build a system that can trace 256 rays/pixel@1080p > 10fps
for a given set of shaders. now we have to select hw that looks most promising when running our
kernels on top.
mp
-
jbikker
- Posts: 225
- Joined: Mon Nov 28, 2011 8:18 am
-
Contact:
Post
by jbikker » Mon Sep 09, 2013 7:17 pm
Even with total rays, that's like 5G, minimum. You intend to use 4 phi's?
By the way, I'm not sure how the phi will behave with divergent rays; CPUs tend to suffer badly if you can't use large packets anymore (as in: 10x slower). GPUs do much better in those situations. If you expect to use a random ray distribution, you may want to look into that.
-
jbikker
- Posts: 225
- Joined: Mon Nov 28, 2011 8:18 am
-
Contact:
Post
by jbikker » Tue Sep 10, 2013 7:39 am
Do you happen to have that car scene available? I managed to find a similar model of the car online, but I would love to test the exact scene on a Titan.