Crashes every time

Started by Mr_Lamppost, October 21, 2007, 01:12:53 PM

Previous topic - Next topic

Mr_Lamppost

I was trying to re-create the image mentioned in this thread:

http://forums.planetside.co.uk/index.php?topic=2536.0

More or less using the methods suggested by Oshyan.  There is nothing special, no wild settings or fancy functions. The moon is not even image mapped yet.  If I leave it for any length of time to get a detailed preview or do an actual render TG just bombs out with one of those "tgd.exe has encountered a problem would you like to tell Microsoft", messages.  The crash comes at different points,

I am running XP Pro SP2. 
Pentium D 2.8 Ghz
1 GB Ram
(Yes I know it's cheap just now and I should get some more).

tgd attached.
Smoke me a kipper I'll be back for breakfast.

Oshyan

Interesting. It does crash near the last 1/4 of the image, but I'm not sure why. I'll have development look at it and see if it helps suss out some of our remaining instability issues.

- Oshyan

Mr_Lamppost

I have had it crash after completing only a few percent of the render, fortunately the problem occurs with small images, no need to wait two days. 

If it helps I have just made tests with GI, Ray traced shadows and finally both turned off.  The crash came in a similar place each time,  the screen grab if from the final test with both GI and Ray traced shadows turned off.  No image editor running in the background either, I started that after the crash.

Smoke me a kipper I'll be back for breakfast.

cyphyr

I had it crash about two thirds of the way through the render. Odd, you don't seem to have any unusual settings, your cloud samples are not ott (over the top). It will be a beautiful image when finished. I can only guess however that it has something to do with the cloud settings but what I cant say. have you tried rebuilding the scene from scratch or alternatively using a different seed number for the cloud fractal.
Richard
www.richardfraservfx.com
https://www.facebook.com/RichardFraserVFX/
/|\

Ryzen 9 5950X OC@4Ghz, 64Gb (TG4 benchmark 4:13)

Mr_Lamppost

If you get any distance through the render before it crashes or let the preview run for a bit you will notice that the old sorting order problem starts to show up, brown line near the top edge of the clouds.  :(    I had another look at the original scene and decided that I was taking the wrong approach; trying to do it properly is not going to work while the sorting order problem persists.  This is a shame because making the moon by placing a dim "Sun" behind a non shadow casting sphere, which can be correctly lit by an out of scene sun worked well here:

http://forums.planetside.co.uk/index.php?topic=2259.0

Looking again at the scene I am attempting to duplicate I decided that the "Moonlight" is actually coming from a point light somewhere in the middle distance with either a textured sphere between it and the camera or more likely a bit of post work.  :P

I have started again with this cheat method and am just now making the final tweaks to the cloud layer prior to doing a high quality version.  I am still using a power fractal to add features to the moon but I can always patch in a crop once I dig out the image map. 

I have been at this for a couple of hours and not a single crash.   ;D  ;) ;D

The aim was only to do the scene as pure TG with no post work and the cheat version will do for that, I will come back and see what I can do with the proper one another time.  I have another project that is just needing a couple of custom objects to finish it off and I wanted to sketch out a new idea.  That's not counting all the normal things I should be doing. :-[
Smoke me a kipper I'll be back for breakfast.

Mr_Lamppost

When the alpha with the horizon sorting problem came our I tried this scene again but it still crashed.  Memory leak most likely, oh well wait for that to be fixed.  Along comes build 1.9.98.1, without the memory leak.  Still crashes. 

I looked at the scene again and noticed that the cloud height is exactly half the thickness. This makes the lower extent of the cloud layer coincident with the planet surface.  I raised the cloud layer 1m and it renders fine. Cross checked in build 1.9.04.1 and the modified scene renders there as well albeit with the horizon sorting problem. 

I doubt this will be a common problem but something to be aware of.

I have posted the final image on Ashundar:
http://www.ashundar.com/gallery/displayimage.php?pos=-5418
Smoke me a kipper I'll be back for breakfast.

Matt

#6
I was able to catch one crash and fix the problem. I don't know for sure if it is exactly the same crash that you encountered, but it is highly likely. It was a bug in the acceleration cache (it would not occur if you switched the acceleration cache to "None").

Matt
Just because milk is white doesn't mean that clouds are made of milk.

Mr_Lamppost

Sorry I have had house guests and then have been away so haven't being paying as much attention to these forums as I should or else I would have seen this sooner. 

To be honest I had not been able to reproduce the original bug with similar scenes involving the lower limit of a cloud layer intersecting an un-displaced terrain.  If an otherwise undiscovered problem was found as a result them I do feel justified in causing this topic have become active again by publishing a solution to the original problem.

I like to think that I am not someone who screams BUG at the first sign of a problem; if I do say bug then I try to make tests to find out exactly what causes it.  This one had me stumped, I would have never thought of the Acceleration Cache.  ;D ::)
Smoke me a kipper I'll be back for breakfast.