Slow data generation rate for offline synthetic data

I have a decent workstation with multiple A series GPU but when going through this topic, the data generation is pretty much slow. However, when monitoring the CPU and GPU, there is almost no load. Two cores for the CPU and 3% of each GPU. Is there any way to accelerate the process?

Currently loading and re-loading materials can take a while loading scenes/meshes/materials from disk to memory the GPU/CPU can idle.

You can try re-loading the dome materials every X frame using the interval parameter:

with rep.trigger.on_frame(interval=self.dome_interval):
    rep.randomizer.randomize_domelight(dome_texture_paths)

You can also try optimizing your scene by setting up multiple cameras (render_products) in the same scenario, this would allow for more parallel work.

1 Like

Thanks for the tips. Are there any future updates for enabling multicore processing on the horizon for the replicator?

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.