Read Ahead CPU preparation Tweet

[Janman]

In animation it would be good to limit the time GPU's are waiting on the CPU to render.

Before rendering on the GPU there's always some preparation time on the CPU. (depending on how much animated objects there are)
If the CPU could read ahead and cache X number of frames to feed the GPU's,
rendertimes could increase significantly on long animations.

I've tried to make it clear in this Exaggerated graphic;


I don't even know if this could be possible ...
thanks,
Jan
C4D R17 - OCTANE 3.04.5 - OCTANE4C4D 3.04.5

[bartolos]

You mean that while the GPUs are rendering, say, frame 1, the CPU could start preparing for frame 2, so the GPUs can start rendering it as soon as the preprocessing for frame 1 is finished...
While the idea seems quite good, the benefits would probably affect only simple scenes, where GPU part is comparable time-wise to preprocessing time by the CPU. In that case you can leverage free GPU cycles by running 2 rendering processes simultaneously, so the GPU calculations should "fill in" the idle time.

B.

[Janman]

You're right about the "simple scenes", but there are more conditions where this would apply.
When rendering scenes with a lot, or heavy mesh-updates,
when cloning (mograph) a lot of animated meshes
and last but not least, when previewing your animations (+- 150 samples)

In my case,
a lot of my 3D projects are around 5000 frames,
if I even would gain 1 second every frame when previewing and rendering,
that would be 01h20 saved every run.

And that's huge !

[Janman]

In that case you can leverage free GPU cycles by running 2 rendering processes simultaneously, so the GPU calculations should "fill in" the idle time.
B.

Can you explain how you would do that using C4D, C4D-plugin and Octane ?
My experiments halted when running out of memory on my GPU's.

thanks,
Jan

[aoktar]

In animation it would be good to limit the time GPU's are waiting on the CPU to render.

Before rendering on the GPU there's always some preparation time on the CPU. (depending on how much animated objects there are)
If the CPU could read ahead and cache X number of frames to feed the GPU's,
rendertimes could increase significantly on long animations.

Sorry it's not possible with current design of plugin+renderer. You can try to run/render multi C4D copies and divide your frames.

[bartolos]

Can you explain how you would do that using C4D, C4D-plugin and Octane ?
My experiments halted when running out of memory on my GPU's.

As I wrote, that works for simple cases only. Scenes that render faster than they get prepared for rendering.
Maybe you could try exporting to standalone but with huge scenes it could be more trouble than benefit.

Cheers,
B

[Janman]

Sorry it's not possible with current design of plugin+renderer. You can try to run/render multi C4D copies and divide your frames.

OK,
thanks aoktar,
maybe one day ...