BIH demo

(L) [2006/07/10] [Phantom] [BIH demo] Wayback!

I have uploaded a small demo of my 'Arauna' ray tracer:

[LINK ftp://mamoth.ompf.org/arauna.zip]

The demo shows four balls bouncing on an imaginary floor. Total tri count is 8k. Hierarchy is updated in realtime, of course. Runs at 11-13fps on my 1.4Ghz monocore, should do a lot better on decent machines.

I think this at least clearly proves how well BIH handles dynamic scenes. Enjoy. [SMILEY Smile]
_________________
--------------------------------------------------------------

Whatever

(L) [2006/07/11] [greenhybrid] [BIH demo] Wayback!

I wanna see, but I need username+password for the download [SMILEY Sad]
_________________
[LINK http://greenhybrid.net/] | [LINK mailto:root@greenhybrid.net root@greenhybrid.net]

(L) [2006/07/11] [tbp] [BIH demo] Wayback!

You hurried too much to snatch the pebble, grasshopper.

[LINK http://ompf.org/alpha/bikker/arauna.zip]
_________________
May you live in interesting times.

[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]

(L) [2006/07/11] [tbp] [BIH demo] Wayback!

fps: 33-42, peak 12M ray/s.

But you should really use a better scene, it's not proving much; 4 disjointed spheres is very easy a scene for any BVH-like hierarchy, there's not much coverage and not many triangles to begin with.

Heck, a kd-tree compiler not tuned for that kind of light load, on a slightly trickier scene (legocar) takes ~80ms - unthreaded - and peaks at 22M ray/s. [LINK http://ompf.org/ray/wip/pix/20060708-00-radius-memcpy.png]

Imagine it threaded, that's 40ms to compile + 12ms to render or 19 fps all of that with 36% triangles, more coverage and a trickier topology.

And we already know toxie had a bining compiler that could handle 20k tri q2 scenes in real time.

So i think you ought to bump the complexity quite a bit.

BTW there's a bug at the start of each line, with a couple of garbled pixels.
_________________
May you live in interesting times.

[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]

(L) [2006/07/11] [Guest] [BIH demo] Wayback!

perhaps those scanline beginnings are something to do with threading?

anyway, i second the need for a more dynamic scene [SMILEY Smile] where are those "completely random triangle soup"s? and where's the coherent reflection and refraction? [SMILEY Wink]

PS. the logging in gremlins are back with a vengeance. i can't even stay logged in for the 3-click-trip from front page to this thread :/ for the record i've been logged in at other phpbb forums for months now...

(L) [2006/07/11] [Phantom] [BIH demo] Wayback!

I was too anxious to get something out. Still, to my defense:

40ms on a 2.4Ghz AMD != 5ms on a 1.4Ghz P-M. Plus, your 'shading' is hardly the 'simple shading' Toxie suggested. I need to access much more data, and I need to process intersection points individually to calculate shading.

But indeed, spheres are too easy for BVH, so I need to come up with a better scene for demonstration purposes. And obviously, I need to show off rt-specific features; at least some shadows from multiple light sources and at least one reflecting ball, preferably one refracting (as that's quite hard to do with rasterizing hardware). I'll strive for a 50k scene or so; the final demo needs to be capable of doing 15fps at least on a medium dual core machine @ 512x384 to be convincible, imho.

The 'random pixels' are indeed the theading indicators.

Can people post some performance figures along with configuration details?
_________________
--------------------------------------------------------------

Whatever

(L) [2006/07/11] [Phantom] [BIH demo] Wayback!

Should scale linearly with nr of cores, I'm really looking forward to those figures. [SMILEY Smile] It's clear it needs to be more complex though... Anyone got a nice game idea that doesn't require balls? (erm, that's not what I mean)
_________________
--------------------------------------------------------------

Whatever

(L) [2006/07/11] [Phantom] [BIH demo] Wayback!

Toxie: No I didn't, but that 8k tri ball scene takes no time to build anyway, so it's just the rendering. I don't have a dc right now, hopefully tomorrow. Threaded building should not be hard btw now I got the worker thing done for rendering.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/07/11] [Ho Ho] [BIH demo] Wayback!

~33FPS/11m ray/s max on 3.85GHz P4 dualcore through wine. BIH switches between 1 and 2 ms. Speed would probably be faster if wine would handle threads better. Currenyl CPU's are running under 150% load instead of 200%.

I would think that FPS gets a serious drop once every ray hits something. But still, couple of years ago not many would have dreamed about such speed.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/07/12] [River] [BIH demo] Wayback!

I'm currious about one thing, how do you proceed with annimation? Generate all the transformed triangles right at the beggining or you keep matrices alive that you multiply when you need (when building the tree and testing intersections)?

(L) [2006/07/12] [toxie] [BIH demo] Wayback!

Quake 3 has a ton more polygons and also "ugly" features like multitextures, blending fx, fake fog, etc.

So the shading code alone should be a pain in the ass to do on the CPU.

But maybe a simple shade (f.e. depth buffer rendering only) tech-demo could be done if one also uses

the precalc BSP/PVS data built into the Q3 levels to clip most of the level??!
_________________
what do you expect to do if you don't know what to do when you've got nothing to do?

(L) [2006/07/12] [Phantom] [BIH demo] Wayback!

River: I transform all vertices at the beginning of a frame (just translation, so I don't even have to recalculate or transform normals), then I build the tree and finally I render it. Only rendering is multi-threaded at the moment.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/08/15] [monkeyman] [BIH demo] Wayback!

btw is the source for this also coming out? Id like to see the BIH construction and traversal and whats different about it...
_________________
This is the funniest video I have ever seen!:

[LINK http://video.google.com/videoplay?docid=5430343841227974645&q=internet+is+for+pron]

(L) [2006/08/15] [Phantom] [BIH demo] Wayback!

Source code for kd/sah is available. BIH is very simple to implement (see Toxie's paper), and traversal is almost identical to kd and described in detail on this forum.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/08/23] [Ho Ho] [BIH demo] Wayback!

Quite old topic but I wanted to brag [SMILEY Very Happy]

65-75FPS, max 22.5m ray/s. Bih stays at 1ms.

System is Core2 e6300 [LINK mailto:1.86@3.08GHz 1.86@3.08GHz] with some really crappy memory settings. Again this runs through Wine and both cores are not under 100% load.

About 2x faster than my old dualcore [LINK mailto:P4@3.85GHz P4@3.85GHz] [SMILEY Smile]

This CPU is awesome, it doesn't need extra voltage to reach that kind of speed. Only problem is that it needs some expensive motherboard to reach >1.75GHz FSB. It should go up to 3.4GHz but it will take a while until I get my system up an running.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/08/24] [toxie] [BIH demo] Wayback!

I'd say it's 2x faster because of the -real- 4x SSE 1 cycle execution (old P4's/PIII's could only do "2x SSE", so >= 2 cycles per SSE opcode).

And with the highly optimized SSE code we all have built into our rt-cores, this could really result in this measured 2x speedup.
_________________
what do you expect to do if you don't know what to do when you've got nothing to do?

(L) [2006/08/24] [Ho Ho] [BIH demo] Wayback!

Actually per-clock its more than 2x spedup. P4 was running at 3.85GHz, Conroe 3.08. Linearly scaling to same speed it would mean conroe was ~2.5x faster than P4 running same speed.

I'll try to get 64bit Gentoo working also and get some native tracers working to see how well Core2 fares then. I have quite high hopes for it. only problem is that currently I have no 64bit tracers and the mortherboard is giving me a lot of pain because I can't get some of my drives working. I'll keep you posted if I have something interesting.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/08/24] [tbp] [BIH demo] Wayback!

Heh. Granted my business plan didn't include options for a) taking over the world b) profit.
_________________
May you live in interesting times.

[LINK https://gna.org/projects/radius/ radius] | [LINK http://ompf.org/ ompf] | [LINK http://ompf.org/wiki/ WompfKi]

(L) [2006/08/24] [Phantom] [BIH demo] Wayback!

No but seriously, what does it take to get help from Intel for research? I know they are looking for benchmarks for future machines, I know they are interested in ray tracing, heck I even did an article on it for them. But the last Intel employee / researcher that I spoke to (the guy that was supposed to team up with me for that article) didn't seem very interested. Sigh. O well I'll probably buy a Conroe myself. [SMILEY Smile]
_________________
--------------------------------------------------------------

Whatever

BIH demo back