X3 Reunion graphics again. This time raytraced.

Kaeru Gaman · Post by **Kaeru Gaman** » Wed Apr 18, 2007 10:09 am

what smoothing do you use?

the first example's triagle resolution is completely sufficent for my taste.
I mean, triangles per object or triangles per detail.
with that "resolution" one could create really complex scenes with under 1M triagles...

Hades · Post by **Hades** » Wed Apr 18, 2007 3:12 pm

Thanks!

With those two scenes, I just wanted to show that primitive count doesn't matter that much anymore, if you have a working acceleration structure. So even scenes with several million primitives shouldn't slow down the raytracer too much, unless memory access gets to random.

The smoothing is very simple. In most cases, the normal at the vertices of each triangle is computed by averaging the normals of all adjacent triangles. Those normals are just interpolated over the triangles. If your looking close at picture 1, especially the torus, you see that the normal should really be normalized. I think the lighting it's quite ugly, but that's an easy fix.
It just wasn't what I wanted to show.

Hades · Post by **Hades** » Sun May 20, 2007 6:28 pm

Hi, me again!

I need your help!

Before coding any further, I'd like to get some feedback from you, so I've uploaded a small Demo (1.2MB, Windows).

It's nothing special, just that X3 Reunion ship again. But without shadows, and no bilinear filtering.

Does it run on your Rig? How fast (from start position)?

WSAD - Movement
Cursor keys - Rotation
1-7 - Resolution
Escape - End

Thanks!

Hades

THCM · Post by **THCM** » Mon May 21, 2007 8:07 am

Nice work! Resolution 7 takes 218ms to render one frame. Resolution 1 to 3 takes 16ms because of vsync. Nice to see that your code uses all 4 cores!

Derek · Post by **Derek** » Mon May 21, 2007 11:10 am

File seems to be missing from the servers, just tried to get it from 3 places and it's not there.

Hades · Post by **Hades** » Mon May 21, 2007 3:15 pm

@Derek

I've just downloaded it, and everything was fine. Strange...

@THCM

Great to hear! You've got a Kentsfield? How fast?

In theory the program should use any number of cores you throw at it. How high is your CPU utilization?

By the way, thanks for testing and reply!

PS: Resolution 7 is 2560x2048 rays, rendered to your desktop resolution. So if you've got 1280x1024 it's 4 rays per pixel. Doing this in 218 ms means 24 million rays per second. Damn, you've got a nice mashine!

Derek · Post by **Derek** » Mon May 21, 2007 3:53 pm

My fault, I had getright turned on.

I get:
1 16ms
2 31
3 47
4 62
5 94
6 156
7 594

Details in my sig.

citystate · Post by **citystate** » Mon May 21, 2007 4:02 pm

MAN you've coded a nice demo there - well done

my speeds are much more modest than THCM's (but then again we can't ALL have supercomputers

)

res 1: 30ms
res 2: 53ms
res 3: 77ms
res 4: 114ms
res 5: 171ms
res 6: 277ms
res 7: 1016ms

on a Celeron 3.33GHz with 2GB ram 1280x960 screen res

so when are we going to see X4 (powered by Hades StreamRay) - or failing that, a DLL for us mere-mortals

Hades · Post by **Hades** » Mon May 21, 2007 4:29 pm

@citystate

Thanks!

so when are we going to see X4 (powered by Hades StreamRay)

lol - I'd really LOVE to do something like that! It just may take some time...

or failing that, a DLL for us mere-mortals

That' a little more probable.

It's just... there is so many dependency everywhere. You can't just call one function and get a nice picture. I don't know how to do it to be usable for other people, and I'd hate to write a doc for it.
But I would like to share it, if you have ideas how!

@Derek

I suspect your C2D is at stock speed!?

You've got the same speed as my Athlon X2 @2,6 GHz.
Your C2D should be faster. But then... the code is hand optimized for AMD. Maybe I'm just missing something. I'll have to read that Intel optimization guide again...

THCM · Post by **THCM** » Mon May 21, 2007 5:09 pm

@Hades I have a Q6600 running at stock 2,4 Ghz. Here are my scores:

ms / %
16/15
16/18
16/25
31/57
47/72
62/80
218/95

My desktop resolution is 1920x1200x32. I'll do some tests later using my overclocking setup (3,4 Ghz).

Hades · Post by **Hades** » Mon May 21, 2007 5:42 pm

@THCM

Damn, I really don't like those numbers.

My rig runs at 88% for res 1, and 97-100% for all other resolutions. Without vsync it looks very ugly at low fps, but I should have removed it for this test. But there has to be another problem with my code on your machine... Ok, still a lot of work ahead. :roll:

Derek · Post by **Derek** » Mon May 21, 2007 6:06 pm

Yeah, stock speed so that's only 1.86 and my graphics card is VERY, VERY slow.

My %'s are 54,75,83,88,92,96 and finally 99 at resolution 7.

Hades · Post by **Hades** » Mon May 21, 2007 6:24 pm

@Derek

Your graphics card shouldn't matter, because all the hard work is done by the CPU. The graphics card is only used to render a single quad with the raytraced picture as a texture. An AGP bus could be a bottleneck for high resolutions though. For resolution 7 that texture has more than 20 MB.

What really bugs me... your CPU utilization is to low, too. It would only make sense, if vsync works different on your rigs, than on mine. Very strange.

Edit: I've removed that vsync and another possible performance killer, and uploaded a new exe (removed, better one below) . Could you please replace the old one, and report the new performance numbers (especially THCM)?

Many thanks!

Hades

Edit2: I've just replaced the above exe for a more stable one (*hopefully* no more crashes when switching resolutions).

THCM · Post by **THCM** » Tue May 22, 2007 8:08 am

Here the scores for your latest exe (always 100% usage):

0-16ms
0-16ms
0-16ms
16-32ms
31-42ms
47-62ms
203-218ms

Numbers are hard to read, they are changing too often.

Derek · Post by **Derek** » Tue May 22, 2007 9:11 am

Here's mine, 100% cpu usage as well.

1 0-16
2 16-31
3 31-47
4 47-63
5 74-96
6 140-170
7 578-594 and a massive jump in the size of the pagefile usage!