Rendering Performance and Quality

 

A render is not finished when the “kernel”->”maxsamples” are reached. It is finished when the Viewport image quality is free of noise and all details have been rendered to a fine level. For some simple scenes, this will be after 500 samples. For some complex scenes it may take 10,000 samples.

 

Tips to increase render speed (by increasing sample/sec or reducing the samples required to finish the render):

  1. Use the “directlighting” kernel. This may provides a lower quality output but is much faster than “pathtracing” and “pmc”. Only use this option if the reduction in render quality is acceptable.
  2. Activate the Spectral AI Denoiser from the Camera Imager settings.
  3. Minimise the use of SSS (“diffuse”->”scatter”)
  4. Set “imager”->”hotpixel_removal” to 0. This eliminates most ‘fireflys’
  5. Increase “kernel”->”caustic_blur” (to whatever level needed – even ‘1’). This smooths out the noise in the render
  6. Use a higher-powered graphics card, or add additional graphics cards to your system
  7. If the light illuminating the scene has come through a specular (glass) material, using “pmc” rather than “pathtracing”. In this situation you can also reduce “kernel”->”direct_light_importance” to give the light that has passed through the glass more priority. Also, you can let more direct light through the glass by reducing the glass material’s opacity (from 1 to say 0.2).
  8. If illuminating the scene by emitters, make sure those emitters have as few polygons as possible, and are as large as possible. If using emitters, it is highly recommends that you an IESAn IES light is the lighting information representing the real-world lighting values for specific light fixtures. For more information, visit http://www.ies.org/lighting/. file in the “distribution” pin to direct the light as required.
  9. If you are getting hotspots or noise that is not quickly resolving, reduce the GI Clamp setting in the Kernel (down to as low as 1 if needed).

 

IMPORTANT: Remember to check “kernel” -> “rayepsilon” for each scene. The “rayepsilon” will be smallest for close-up renders, and increase as the camera moves further away from the subject.