Arauna

(L) [2006/08/25] [Phantom] [Arauna] Wayback!

Here's a preliminary demo of the Arauna ray tracer. Sorry for posting both here and in the visuals section.

[LINK http://www.bik5.com/demo_arauna.rar]

The demo does not support super sampling (that's too slow). Also, the secondary rays for reflections are not accounted for in the stats (kray/s). I would really like to hear how this performs on the Conroe though.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/08/29] [playmesumch00ns] [Arauna] Wayback!

Because it's running linux [SMILEY Smile] I downloaded the rar but could only see a .exe. Could you do a linux build?

We do have quad-core boxes, but they're on the render farm, so no display. [SMILEY Sad]

(L) [2006/08/29] [Ho Ho] [Arauna] Wayback!

That thing works rather nicely through Wine. Though I have no ideas how big is the performance loss. I'd like to know that myself so I'll just line up waiting for a source release [SMILEY Smile]
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/08/30] [playmesumch00ns] [Arauna] Wayback!

Can't do wine too well either. I'll give it a whirl and see. We're using a bastardised version of fedora1 here (we have a couple of hundred workstations so os updates tend to be rather slow) with a horrid mess of library configs. I'll give it a go and see what i get [SMILEY Smile]

(L) [2006/08/30] [Phantom] [Arauna] Wayback!

Good news: I finally found the last bug that caused some bad artifacts on intersecting reflecting surfaces. I'll remove my testing code (which is all over the place after three days of hardcore debugging) and then I'll upload another demo.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/09] [Phantom] [Arauna] Wayback!

Here's a fresh demo of my Arauna raytracer:

[LINK http://www.bik5.com/arauna_demo2.zip]

Instructions:

1. Unzip and run arauna_supersampling.exe to measure peak performance (~11M on my machine);

2. Open scene.txt and add '//' to the fifth line (speedtest);

3. Run arauna.exe to walk around the scene. Keys:

A, S, D, W to walk forward/backward and strafe;

cursor keys to rotate the camera;

R and F to move down and up (reversed, sorry, forgot to fix).

That's the basic stuff; there's more to try out using scene.txt, have a look at the file to see what's possible.

The executables are optimized for architectural walkthroughs and perform bad for solo objects (e.g., legocar).

Please post your timings!
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/09] [tbp] [Arauna] Wayback!

modulo defects in traversal, shadows etc...
_________________
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) [2006/09/09] [Ho Ho] [Arauna] Wayback!

Core2 @~3.1GHz through wine.

max 25.6Mray/s, mostly stays around 24M.

With the walkthrough I'm usually seeing ~25FPS. Staring at a lit wall really close I get ~60FPS, staring into the void I get 36FPS.

Is it normal that shadow intensity is dependant on view angle or is that those errors tbp is talking about?

I'm not sure if mu CPU is running at maximum speed. Since the motherboard is not that well supported it might be that it isn't. I've yet to find a tool that reports real speed of CPU and not the one it would have when FSB is at its default.

[edit]

Now someone should add a nice camera path to it to make a benchmarking tool out of it. I could probably get tens of test subjects in a matter of days [SMILEY Very Happy]

Also if it could be ran without X under Linux I could probably get some results from 16-core AMD box. Don't know if 32 or 64 bit though.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/09/09] [Phantom] [Arauna] Wayback!

The shading is odd because I multiply by 0.5 if in shadow, on top of Schlick, wich is evaluated regardless of shadows. It is thus not a result of being cheap or whatever, it's just a lousy way to do the ambient factor.

@tbp: What traversal problems did you encounter? Can you make screenshots of problematic cases? By the way, if you press 'P' while in freelook mode, the app generates a file 'camera.txt' that contains two lines that you can paste directly in scene.txt. If you then disable 'freelook' and set camradius to 0, you get a still rendering of the artifact, useful for debugging. You can also send these files to me, please specify resolution in that case.

@Hoho: 25.6M max is awesome, the best score I've seen so far. And: If you need source code to test it on those extreme boxes, I can provide you a private package. All code is aimed at 32bit, I don't use (or compile for) 64bit yet. Still not sure what to do with the sources. Seems a bit too valuable to just toss in the crowd. I've seen lesser projects of mine being abused for commercial purposes without even an e-mail to me.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/09] [tbp] [Arauna] Wayback!

My screenshots are too valuable to be tossed in the crowd [SMILEY Rolling Eyes]

Frankly i find a bit dishonest to advertise this as "optimized for architectural walkthrough" when you obviously don't try to address even the most common tricky cases, which aren't rare by any standard.
_________________
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) [2006/09/09] [Phantom] [Arauna] Wayback!

'Optimized for architectural walkthrough' hints at the poor performance of my MLRTA path for the legocar scene, that's all.

And what do you mean by the most common tricky cases?

Apparently I have offended you somehow; if it's the closed source thing, I would like to clarify that I'll gladly pass anyone the source code, I just don't want to see the next version of some random modeller suddenly fitted with an 'awesome realtime raytracer' without any credit. As soon as I find a way to prevent that, the code goes to the crowd. 'Crowd' does not include you or other attendees of this forum, they are welcome to request a private package right now.

I would like to add that my code is built on the knowledge of others: Tbp, Toxie, Wald, Benthin, Reshetov and countless others, so I feel an obligation to share (as I always have done, just check Google).
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/09] [Phantom] [Arauna] Wayback!

Quick initial source release:

http:/www.bik5.com/arauna_src.zip

Proper package will follow. For now, assume GPL.

Enjoy. And be aware, this is a work in progress. As tbp pointed out, it has flaws.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/10] [Ho Ho] [Arauna] Wayback!

I've gotten two other people with C2D to run it under Windows.

E6300 7x400=2.8GHz 2M L2 got maximum of 24.5M

E6600 9x344=3.1GHz 4M L2 got maximum of 28.2M

Seems like double the L2 gets a bit less than 4% speedup assuming linear scaling and wine doesn't cause too big speed drop. Of cource massively different FSB and memory settings can alter the results quite a bit. I'll get some better numbers once I get the code to work under Linux.

So far there are no big problems, mostly just restrict instead of __restrict__ and shadowing variables. Of course I haven't got to winapi and threading yet. I hope tbp won't mind if I borrow his code for those things to make my life easier [SMILEY Smile]

[edit]

Oh, yeah. One x2 [LINK mailto:3800@2.6GHz 3800@2.6GHz] got 10M too. I wonder why its so slow considering fpsunflower got 10M with only 2x2GHz.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/09/10] [Shadow007] [Arauna] Wayback!

I get 2730 K peak perfs on my "poor old" Athlon 64 3000+ @reference clock speed ...

Besides, I compiled the project with VS2003 (7.1) without problems except fo a missing intrinsics (_mm_cvtss_f32) which seems to be provided by the Intel Compiler.

After having googled for it, I found the following "replacement" :

(L) [2006/09/10] [Ho Ho] [Arauna] Wayback!

No, it's 2x2GHz with 512k L2 per core at default. That one was OC'd to 2.6GHz.

Now I got another report, 2GHz X2 got ~15M rays/s. That sounds a bit too high.
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/09/11] [Ho Ho] [Arauna] Wayback!

Phantom, add support for 8 cores and let [LINK http://www.xtremesystems.org/forums/showthread.php?t=115384 this] guy to run your proggie [SMILEY Very Happy]

"Lemme know waht benchies u wanna see on this Octal Core Monster"
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/09/12] [Phantom] [Arauna] Wayback!

It already allocates 8 threads; this can be increased in common.h. So any number of cores is supported, basically. I would love to see those figures. [SMILEY Smile]
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/12] [Ho Ho] [Arauna] Wayback!

I was just asked if this program supports 16 threads. One guy has access to some nice Opteron boxes and he was willing to run it there. He also has some 8-core ones. Is it correct that your program will use all those cores without modification? Only problem is I'm not sure if they are running Windows or Linux. I know he once offered me to run it on a similar Linux machine.

He also said he has some 32-core Sparc he could test. Anyone willing to replace all that SSE2 stuff with whatever the Sparc has ? [SMILEY Smile]
_________________
In theory, there is no difference between theory and practice. But, in practice, there is.

Jan L.A. van de Snepscheut

(L) [2006/09/12] [Phantom] [Arauna] Wayback!

Ho ho: If you can compile the code, you can simply increase the 'THREADS' count in common.h which is set to 8 by default. This is the number of worker threads; I found that 8 doesn't decrease performance (compared to two), so I set it to 8 to be prepared for faster machines. I didn't take 16 cores into account, so it needs to be increased. If you can't compile, let me know, I'll gladly provide a new win32 executable.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/12] [Phantom] [Arauna] Wayback!

Hm, that might be an option, I'll check it out. I'm rather busy right now, so that's not a solution for this week.
_________________
--------------------------------------------------------------

Whatever

(L) [2006/09/13] [andy] [Arauna] Wayback!

i've compiled the src ok with vs8, fixing up an error about zero length arrays which it didn't like, but i'm getting nowhere near the performance of the binary you posted. (20fps with your binary, ~5fps with my binary and ~2fps with bih).

does the src release match the binary release in terms of kdtree build parameters, max depth etc? do you have any crazy compile flags?

any other ideas? it can't just be the different compilers, ms isn't *that* bad.

(L) [2006/09/13] [Phantom] [Arauna] Wayback!

Yes that's typical. Note that the code is optimized for icc; I never looked for improvements using VC8.

By the way: I can't get 'profile guided compilation' to work under icc, and I have no idea why it doesn't work. It keeps reporting that there is no instrumentation file, despite the fact that I followed all the steps.
_________________
--------------------------------------------------------------

Whatever

Arauna back