Warning Num Samples Per Thread Reduced To 32768 Rendering Might Be Slower ((better)) Jun 2026

Now, with (e.g., on an 8‑core CPU or a GPU with 8 SMs), the naive samples per thread would be 524M / 8 = 65.5 million — far above 32768. But the renderer doesn’t process all samples for the whole image at once; it works in tiles or work chunks .

When you enable both CPU and GPU rendering (hybrid mode), the scheduler has to balance workloads. The per‑thread limit is applied to each device separately. A slower device (especially CPU threads) may cause the renderer to clamp the samples per thread across all devices to avoid synchronization issues. Now, with (e

By setting the samples per thread consciously (e.g., matching the warned number), you often achieve a stable, consistent render speed. Summary Table Why it Helps Decreases overall calculation complexity. Reduce Geometry/Textures Frees up VRAM and reduces computation. Optimize Lights Simplifies the ray-tracing workload. Set Samples per Thread Manually Stabilizes the rendering process. The per‑thread limit is applied to each device separately