Multi-Core/Thread Use in 3d Coat Render Room

[Rkhane]

Hey guys, recently upgraded from a 2.66 core 2 duo 8gigs ram to a i7-4790 4GHz 16gigs ram. Expected some performance gains, ran some test renders against my old machine for rendering out a massive image in the render room.
 

Both machines used my Nvidia 750ti, 2 gigs ram

 

Test: 28million poly render to PNG at 15360x8224 pixels. (For printing art for a 16x20in poster at 300dpi...)

 

All tests were done in 3dCoat 4.1.17

2.66 core 2 duo - Rendered the PNG in 18:07min - 3dCoat64BitCuda - OpenGL

i7-4790 4GHz - Rendered the PNG in 19:37min - 3dCoat64BitCuda - OpenGL

i7-4790 4GHz - Rendered the PNG in 31:19min - 3dCoat64BitCuda - DirectX - What, LOL!

i7-4790 4GHz - Rendered the PNG in 19:32min - 3dCoat64BitOpenGL - No Cuda

 

Also noticed some interesting results.

 - The OpenGL version taxed 1 thread at 100% and used another thread here and there, 25% at the most. 3dCoat64BitCudaOpenGL

 - The DirectX version taxed two threads evenly at about 10% total load as opposed to one thread heavily.

3dCoat64BitCudaDirectX

 - Neither render used more than 2 threads out of 8 available.

 

So things I learned possibly:

1. 3d Coat Render Room is heavily GPU reliant and not CPU reliant as my new CPU had no positive effect on the render time? After doing a lot of research into 3d rendering, I thought this was mostly CPU based?

2. Open GL Rendering seems to run DRASTICALLY faster. That was a shock...

3. The Render Room seems to only take advantage of 1-2 cores at best.

4. CUDA does not seem to affect render times as the Cuda and non-Cuda renders were very close.

 

I do a LOT of rendering in the Render Room for my work and would like to figure out how to speed up my render times drastically in 3d Coat. Hoping for some help and understanding in this, thank you!

Edited April 6, 2015 by Rkhane

[Rkhane]

Installed GPU Z today, GPU stayed at 99% load during the entire render for 28mil poly file. Before the render GPU was much lower, 3-4% load.

I then loaded a 4mil poly model that the graphics card handles with ease. When rendering GPU load stayed at 99% again, sometimes dropping to 98%.

Both renders were at 15360x8224 pixels. During both renders, 1 CPU core and its second thread stayed at 100% the entire time.

Memory controller stayed at 44% while moving, manipulating and later rendering the 28mil poly model. (128bit bus, GDDR5). The 28mil poly file used 1400 megs of ram out of 2000.
The 28mil poly model rendered in 19+ minutes.

The 4mil poly model rendered in 4+minutes.

So from this I would possibly conclude:
3d Coat Render Room relies mostly on GPU as opposed to CPU.

Does not use Cuda Cores to render faster.

 

Well thanks for listening to me work through this

Just shocked me that 3d coat renders with GPU as opposed to CPU. When doing research, you always here about rendering with CPU. I guess 3d Coat renders out like a video game engine would, pretty cool, good to know, lol