V-Ray GPU with high-end Ampere, Ada and Blackwell GPUs

Introduction

The NVidia Ada Lovelace architecture, found in high-end GPUs like the RTX 4090, delivers a significant leap in rendering performance. This article provides helpful tips for configuring your system to get the best performance from such graphic cards.

For reliable monitoring of GPU performance, we recommend using a dedicated hardware utility such as GPU-Z or MSI Afterburner.

Interactive rendering

Since the default settings are optimized for best interactivity on mid-range GPUs, for optimal performance on high-end GPUs, it is necessary to adjust the interactive rendering settings. 

Read more about the GPU IPR Settings:

• 3ds Max
• Maya
• Houdini

Recommendations

• IPR Rays per pixel: 2 - 8
  Increasing this value increases the utilization of higher-end cards.
• IPR ray bundle size: 128 - 512
• Undersampling: personal preference

Sample presets

• Preset for high-end GPUs (RTX 3090 and RTX 4090):
  IPR Rays per pixel: 4
  IPR ray bundle size: 256
  Undersampling 2
   
• Preset for multiple high-end GPUs:
  IPR Rays per pixel: 8
  IPR ray bundle size: 512

Video demonstration of the IPR performance in Maya

Production rendering

To maximize the performance of high-end Ampere, Ada and Blackwell GPUs, follow these recommendations:

• Set Post Effects Rate to 0

  This will ensure Denoising and Lens effects are calculated only on render end.

  

  Where to find the Post effects rate setting:

  • 3ds Max
  • Maya
  • Houdini

   

• Use the less Render Elements

  Render elements have a big impact on GPU performance as seen in the graph below, which compares a scene with and without 25 render elements. Notice the larger fluctioations in GPU utilization. 

  

   

• Mind the VFB log for warning messages

  If the following message appears in the VFB log, this means that V-Ray GPU doesn't have quite enough VRAM for rendering the scene and must compromise performance to work with it. Optimizing GPU memory usage is recommended.

  Num samples per thread reduced to xxxxx, rendering might be slower

  Learn more about Optimizing memory (VRAM) usage for GPU rendering.
   

• Use Progressive Image Sampler for production rendering.

  Generally, it is faster and more memory efficient than Bucket mode.

  

   

• Light Cache is now caculated on the GPU.
  Note that only a single GPU is used for calculating Light Cache on GPU. Because of that, GPU usage will drop below 100% for a few seconds at render start, which is expected.
   
• Avoid having empty space in your frame whenever possible, since it affects GPU performance and usage.
   
• Avoid mixing 2 different architectures of GPUs.
  This is especially important if one of the GPUs is significantly faster than the other.
   
• Use the recommended driver for V-Ray GPU.
  Recommended driver are regularly updated in the minimal and recommended system requirements page.
   
• Use a CPU with at least 6 physical cores per high-end GPU.
   
• Thermal throttling for GPU and/or CPU will affect GPU performance and usage. Ampere and Ada GPUs will throttle around 81°C. Good airflow and undervolting are recommended.

If you still experience low GPU usage in your project, do not hesitate to contact support.