Thea Render Revision 568

[giannis]

I don't understand the percentages above...

We have made extensive testing with various scenes. The speed up depends on the scene and it is better seen on TR1/TR2/IR. It is more evident on heavy scenes and where instancing is present.

Note also, there is a difference between time and speed. A 20% reduction in time is 25% speed increase, a 30% reduction in time is ~43% speed increase, and so on. We have seen 50% reduction in time on some scenes with the exception of heavy instancing scenes where reduction was much much more.

best wishes

[gargon]

Thank you very much Thea team for the work.

[gelbuilding]

Shear brilliance.
Thanks to all the Thea Team,
This new revision is awesome.

Working like a charm. Speed and user interface is incrediblely smooth.

Just love it so much.

Thanksyou. thankyou, thankyou.

[HuMa892]

Could you please inform us, when Colimo 1.2 is coming out? I need to get a Prkoject done till Saturday in the Morning and want to present it in Colimo with different textures.
Do You see a chance for a release today?

[seanser]

Thanks for all your hard work, it is much appreciated.

I do think I may have found a little bug, it seems I am no longer able to adjust hdr intensity during rendering.

[giannis]

Do You see a chance for a release today?

Yes, I believe that the final version will be out in any moment.

[giannis]

I do think I may have found a little bug, it seems I am no longer able to adjust hdr intensity during rendering.

Good!

[notareal]

Thank you very much for the new version of thea! Many new things to learn!!

Yesterday I made some "speed tests":
v1.0.10.505 TR1 100%
v1.1.01.568 TR1 80%
v1.0.10.505 BSD 100%
v1.1.01.568 BSD 80%

That means v1.1.01.568 is 20% faster then the older version. About this I am a little disappoint because it is not close to 100% like told... I do not know anything about this and I do not want to seem arogant it is just that for me the speed is still a very important factor. I can imagine that maybe other scenes work faster. But in the scene I used are most basic materials. It is an interieur scene.

Perhaps you can post your test scene? For unbiased TR1/TR2 you always need to compare the noise level, not the time used for a render pass. Easiest way is to compare the file size. Lets say, if you do a 5 min render with RV505 and then do few test renders with RV568 (maybe 5 min, 3min and so) and then compare what test render file size is close to the RV505. Now you can see the actual speed difference vs noise.

[giannis]

To be honest, the speed up is directly related this time to the render times. In many other cases, the speed is related to the noise level, but this time it is raw acceleration power.

best wishes

[ThomasHelzle]

giannis: I tested some more and it seems both the suboptimal CPU usage as well as the long stopping time are related to me rendering the Normal, Depth, Alpha and Material ID channels at the same time as the main image.
I'm rendering at 2560x1600, but I have 14 Gig of RAM, so I'm not sure what the problem may be.
Also, I measured how long it takes to stop the rendering and it is over 2.5 minutes on average. In this time, CPU usage is around 2-3 % - not sure what's going on there...
Copying stuff around in memory?

The scene is one single object with ~450.000 polys on 5 surfaces.
If needed I can provide the scene, but I tried some more and see the same behaviour on other scenes as well.

Just speculating here, but with FPrime for Lightwave (which works similar to unbiased in some ways) at one point there was a speed decrease with many cores and fast CPUs because FPrime synchronized too often to the file on disk, since it was optimized for slower machines.
Now I wonder if rendering several Channels is slow because there is too much switching between the Buffers or something like that?
Now that building the Environment is multithreaded and superfast, would that be possible for the stopping phase as well somehow?

And a last question: I have 8 cores and the Thea Threads setting has a max of 8. I saw a speed increase with some other renderers when setting Threads to 16 (although I have no Hyperthreading). Is the 8 Threads limit hardcoded in Thea or does it set this limit based on physical cores available?
Now that there are 16 and more core machines available, I wonder how Thea scales on those...
I personally would like to see a higher Thread limit, even if it is just for testing purposes

Cheers and thanks!

Thomas