Forums » Work in progress
Pages: 1
The progress of working in CentiLeo ^_^
Testing. Sharing. Feedbacking.
 
User  Posts: 7
Aug 15, 2016 07:21
Hi there! Here’re some interior shots rendered in 0.40! Scene has apprx 7 million polygons and 100+ lights. While interactive works really fast, scene converges fast as well, there are too many bugs so far and too many things should be implemented for comfortable workr. As for me, first of all you should make UVW coordinates to work correctly, for habitual faster texturing, make it visible when the lights are turned-off and make resource collector alive))) Exposure correction, gamma and other colour tuning seem pretty nice!

Anyway, the idea with a good prospect, looking forward for fixes for dat big amount of bugs)

I begin testing how 0.41 works and and try this scene with slightly more geometry and features. will keep you in touch guys.

And the pix! 8)
2222.jpg (1.63 Mb)
333.jpg (1.59 Mb)
1111-3.jpg (1.63 Mb)
clean.jpg (1.12 Mb)
Edited: Bakha - Aug 16, 2016 11:13
 
Administrator  Posts: 933
Aug 15, 2016 08:01
Hi Bakha, thanks for testing! Nice WIP :)

Still noisy, how long is was it rendered and on which GPU?
What is on the street? Geometry or envmap?

I think 100 lights may converge not very fast but we will work in this area.

How is it easy to use our material cntlStdMat? Have you understood it by yourself or have you read the documentation which we provided with 0.41?

Noticed that you spots look black for direct visibility. It's our bug. Already fixed, will upload it soon.

Can you also post a topic on the Wishlist forum with all the things we need.
CentiLeo Chat: https://t.me/centileochat
 
Sami Timkala
Guest  Posts:
Aug 15, 2016 09:17
Hi Bakha, progressing well! Wait your new works.

On the 3rd image there is broken highlight. Is it our forum bad conversion, jpeg too high gamma usage?
 
User  Posts: 7
Aug 15, 2016 11:07
Hi, guys!
Quote
Still noisy, how long is was it rendered and on which GPU?
What is on the street? Geometry or envmap?
Yeah, its a bit noisy, but i didn't aim to make perfect result, was just curious how the final shots could look like. It took ~40-45 mins on GTX 970, 1080p res, per shot.
Nearest pavement and street lamps are geometry and the distant urban view is envmap (not in a so high res btw ^_^).

Quote
I think 100 lights may converge not very fast but we will work in this area.
To be honest, the scene converged noticeably faster with 20-30 lights and it's good direction to optimise! Anyway, 40 min at 1080p resolution for this scene is a decent result i, think)
Quote
How is it easy to use our material cntlStdMat? Have you understood it by yourself or have you read the documentation which we provided with 0.41?
i have some expirience in several renderers, so it was quite easy. Started testing before the documentation was uploaded, but, anyway, thank U for uploading it. I'll check some moments)
Quote
Noticed that you spots look black for direct visibility. It's our bug. Already fixed, will upload it soon.
Good! Keep on, guys!
Quote
Can you also post a topic on the Wishlist forum with all the things we need.
Yeah. Need to remeber and list the wishes. Try to post in a few hours.

Quote
On the 3rd image there is broken highlight. Is it our forum bad conversion, jpeg too high gamma usage?
It's my fault i guess. Experimented with some post-production stuff and uploaded not the best one :oops:



Working with new 0.41 vesrion now. Some bugs appear to be fixed! Good! Upload some pix later, then.
Edited: Bakha - Aug 15, 2016 11:08
 
User  Posts: 7
Aug 16, 2016 11:16
One more clean example. Need fixing materials
clean.jpg (1.12 Mb)
 
User  Posts: 7
Aug 16, 2016 11:26
There is no problem loading and applying large textures. 4K, 9K - it all ok! But need to clean disk from caches.

Noticed some additional blur visible for lowres texture seen through perfect window - must fix it.
 
User  Posts: 7
Aug 16, 2016 14:35
Seems that with firefly_killer = 0.5 or 1.0 the noise goes away faster in this scene and Ms/sec is higher. It seems to be coool!

Why different render_iter_power changes Ms/sec too much?
 
Sami Timkala
Guest  Posts:
Aug 16, 2016 19:40
Quote
Bakha wrote:
One more clean example. Need fixing materials
Intersting work! Thank you

Quote
Bakha wrote:
Seems that with firefly_killer = 0.5 or 1.0 the noise goes away faster in this scene and Ms/sec is higher. It seems to be coool!
With firefly_killer there are different internal algorithms involved and so they have different performance characteristics resulting in better performance for some heavy lightin scenes and no more than 2x worse performance for relatively simple lighting. These algorithms (enabled by positive firefly_killer) are experimental and not switched on by default. They are a little bit clever for some situations but have some memory overhead.
Quote
Bakha wrote:
Why different render_iter_power changes Ms/sec too much?
Smaller render_iter_power simply allocates less number of samples to compute one iteration. So the the image iteration computes faster with render_iter_power = 0 but it brings less valuable information per iteration and saturates the GPU less. However, it can be usefull for small GPUs and interactive previews while modeling. For performant GPUs like 980s it is recommended to enable render_iter_power = 1
 
User  Posts: 7
Aug 24, 2016 20:29
Hello!!! Played with 0.42 is a lot more stable, never crashed excpet when I changed the GPU to another one while IPR was running.

Couple of new renders. Here a lot more centileo shaders utilized in the scene, they are very useful, triplanar projection, two reflection layers. And I really like surface mask shader on multi material. But I need dirt When will you do it?
Next time will dig deeper into SSS and also would like to apply more displacement.

This image is 2 hours render in 1920x1080 on single 970. There are 100+ light sources here and white balance 10000 and I think it's very fast for unbiased engine with this amount of lights.

UPD: Uploaded some new shots!


3.jpg (1.51 Mb)
2.jpg (1.46 Mb)
1.jpg (1.46 Mb)
0.jpg (1.69 Mb)
Edited: Bakha - Aug 25, 2016 23:22
Pages: 1
Users browsing this topic