is TG2 subject to tiredness ???

Started by chaps, October 03, 2008, 09:41:14 AM

Previous topic - Next topic

chaps

Hello,

I started a picture using Bigben files for river and mountains. I changed a little the riverbed to get it smoother and deeper. I tried to use Lightning grass clump, but revert to TG2 embedded ones, I was thinking about a relatively fast test. :-\

The progress of rendering was OK for the first 8 parts, and I noticed that TG2 was using pretty well the 2 cores of my laptop, with an average of more than 90% of CPU used, even when the memory usage was 810 000 Ko. ::)
after 8 hours, less than 1/9th of the image was remaining, only one core used, with 50% of total CPU. This look fine, but in fact the image progress is really slow now. After 26 additional hours it still not finished. ???
I have made a difference after 5 hours of rendering, it is really few (see the picture).
Does some one have an Idea of what I should modify in the scene and or setting to accelerate rendering?

Bye

rcallicotte

That's sort of a good question.  I've seen the foreground stuff slow down quite a bit and actually have centered my renders around this fact.  In actuality, if there are objects closer to the foreground (closer to the camera), the memory issues go through the roof.
So this is Disney World.  Can we live here?

chaps

Finally, it reachs the end.
;D

In fact my son stopped my PC after more than 50 hours  >:(.  So I had to relaunch a render limited to a cropped region. The render restarted, dividing the region into 8 smaller parts. The render is very slow when there is grass clumps under and above water (+ fake stone?); otherwise it is quite fast. I will try to verify this with a modification of the blending shader for grass.

I noticed also one thing: in my first rendered image, the shadow of the rock in the foreground looks quite weird. In the partial render, it is still weird, but with a different shape. Is there some option to have really calculated shadows? I join the 2 results for better understanding.

Bye.

rcallicotte

Another thing that might help us is to turn off double-sided on our objects.  This will eliminate some of the render time.  It would be nice to have this for the entire image.
So this is Disney World.  Can we live here?

MacGyver

@calico: is there something like "Back-face Culling" from Terragen 1 in TG2 yet? I often see that water is rendered beneath the actual landscape using unnessecary rendertime. :-\
What you wish to kindle in others must burn within yourself. - Augustine

rcallicotte

@MacGyver - No.  That's what I was asking about.  It's available in a sense for objects by disabling the double-sided checkbox on an object.  But, nothing for the entire scene.
So this is Disney World.  Can we live here?

PorcupineFloyd

Does it work only per object basis or also for whole populations?

rcallicotte

Since the populations are based on the one object, I'd wager it's the entire population.  But, you would need Planetside to confirm this.
So this is Disney World.  Can we live here?

PorcupineFloyd

Sounds logical but as long as we don't know specifically how do populations work we can only assume that it applies to whole population. I'll have to try disabling double sided checkbox and see how big will the improvement be.

Oshyan

Backface Culling is already implemented, it's just not as easy as you might think to cull things before they're rendered. In a procedural environment it's much more difficult to tell ahead of time whether something will cover something else up.

http://forums.planetside.co.uk/index.php?topic=3703.msg38295#msg38295

As to the problems in this scene, I suspect it is the setup of the scene that is causing the extreme slowdown. If you have odd surface/displacement intersections, or are using a merge shader to blend displacements, or in some other situations, it can create very confusing situations for the renderer. I somehow doubt it is just the basic scene that you see in the finished render. There is likely some unnecessary configuration that is causing this.

- Oshyan

rcallicotte

Thanks Oshyan.  I forgot Matt said he would work on the pre-subdivision function.  Good reminder of what's happening in Matt's Dark Basement of fun and games.
So this is Disney World.  Can we live here?

bigben

I've seen foreground water slow down, but probably not to this extent.  From m latest test I'd probably point my finger at the reflective shader you have for the "wet" rocks.  On my riverbed test adding a reflective shader really slowed down the rendering of the bank i the foreground.

chaps

#12
Hello,

If it can help this interresting (for me  :D) discussion, i join a zip of the file I use for this image.

I Have a reflective shader for the wet rocks, but this par of the render was fast. The part which was really slow was the place were it exists a terrain, a grass population, a water shader  and fake stones.

In this part of the picture, there was roughly 4 passes for rendering. The first one with small dots (I gess it is a kind of z buffer preparation?). Then the terrain or something which looks like terrain. Then comes the very slow part: water with reflexion. and last someyhing quite fast: grass cover, if any.

To Oshyan, it is hightly probable that there are useless nodes in the scene. I don't fully understand the role of some of them, and the type of input/output.  :-\.

I still have to make a modification to avoid to have grass in the water, but I have had no time for that yet.

Bye,

Pascal.

rcallicotte

The reflective shader is fine, unless you use Ray Traced Reflections.  While Ray Traced looks cool, it costs a lot of time and energy.



Quote from: bigben on October 07, 2008, 09:08:00 AM
I've seen foreground water slow down, but probably not to this extent.  From m latest test I'd probably point my finger at the reflective shader you have for the "wet" rocks.  On my riverbed test adding a reflective shader really slowed down the rendering of the bank i the foreground.
So this is Disney World.  Can we live here?

chaps

Hello,

I finally found some time to make a test. I removed the grass, and move down the water surface. My guess was that if the water surface goes under terrain (as it actually does), an if the 2 surfaces are close enough, TG could have difficulty to decide which surface is above the other.

The time for render was better (roughly 1/2). That sounds normal because the scene uses much less objects, as the memory consumption confirmed it. But I got the same dramatic slow down, in the same part of the picture. What should be avoid to get faster render:

water a little above terrain
water a little under terrain
mix of water, terrain, fake stones
??? All this conditions exist in other part of the scene, where the render time is OK.

any idea is welcome.

Bye,

Pascal.