Benchmarking Canvas render, does it reduce memory consumption?

Ruris

I was doing some benchmark using a one female model + classroom scenes.

 

Run 1: 2min 6 second for 103 iteration for a full scenes (Benchmark, no canvas render)

Run 2: 2min 6 second for 140 iteration for model only (Canvas beauty, node select model only)

Run 3 : 2min 6 second for 74  iteration for classroom only (Canvas beauty, node select everything except model)

 

Analysis:

Run 2 make sense, since the texture for the classroom was removed, so you should get higher speed. 

Run 3:Seems a bit odd, maybe the initialization stage take longer due to purging memory after multi run? Render window are not closed.

 

Checking the Log File:

Texture memory consumption: 1.75127 GiB (for both card)

Same for all 3 render. 

Analysis:

All the visible scene objects are loaded into GPU, it will not go through a second filter via canvas 'Node'. Hiding object in canvas node does not save memory nor boost performance. 

 

My observation ,canvas is good for postwork preparation, but not for meddling around for performance boost. let me know if this is true or false.

JonnyRay

That actually makes sense to me based on my understanding of Iray canvases. When Studio is preparing the scene, it still has to send everything to Iray because the division and node selection is being done by Iray, not by Studio. While for simple cases it might be logical if Studio would pre-filter, once you get to more complex ones like canvases that use Light Path Expressions (LPEs) it may not be technically feasible for Studio to be able to determine everything that would or would not be included in those calculations without replicating everything Iray is doing to generate the canvases.

Ruris

Hi Ray,

Thanks for sharing your experience. Now that I give it more thought, 'node hiding' and 'scene hiding' has completely different effect, in the sense that, scene hiding will totally ignore the reflected light (blue walls, etc) while 'node hiding' knows that the wall are blue therefore giving everything a blue tint, so using this theory, the texture memory are 100% loaded.  Haven't got to testing LPE layer yet, the canvas process flow is mind boggling enough, heheh.