Warning Num Samples Per Thread Reduced To 32768 Rendering Might Be Slower

When you increase num_samples_per_thread , you are effectively asking the renderer to allocate a larger per thread. The total memory required is roughly:

If you have dozens of small, low-impact lights, reduce their number or combine them.

For those who want to understand the internal numbers, let’s do a quick calculation. This setting dictates how many samples a single

This setting dictates how many samples a single CUDA thread calculates before returning results to the main application.

[Log] Warning: Num samples per thread reduced to 113951, rendering might be slower. [Log] Warning: Num samples per thread reduced to 45210, rendering might be slower. [Log] Warning: Num samples per thread reduced to 32768, rendering might be slower. Use code with caution. [Log] Warning: Num samples per thread reduced to

Set a (0.01 is standard; 0.005 for ultra-clean prints).

If you are a 3D artist, animator, or visual effects professional working with GPU-accelerated path tracing renderers (such as Blender Cycles, OctaneRender, Redshift, or LuxCoreRender), you may have encountered a perplexing warning in your console or log output: If you are a 3D artist

In Blender Cycles, large tile sizes (e.g., 512×512) tend to increase per‑thread sample counts. Smaller tiles (e.g., 256×256) may stay under the limit. Experiment:

Which are you using? (Blender Cycles, Octane, V-Ray, etc.)