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

[Muhammed_Hamed]

Introduction,

The new Ada GPUs bring a big leap in performance powered by the Ada Lovelace architecture. In this article I discuss GPU usage with high-end GPUs like the RTX 4090, plus helpful tips on getting the best performance out of your GPU(s)

Note, Windows Task Manager is not helpful on reporting GPU performance, we recommend using an external tool like GPU-Z or MSI AfterBurner for monitoring GPU resources.

Interactive Rendering,

It is necessary to adjust interactive rendering settings if you have one or more high-end GPUs to get optimal performance and GPU usage. The default settings are optimized for best interactivity on mid-range GPUs. Below are ready-to-use presets for high-end GPUs



Preset for high-end GPUs(RTX 3090 and RTX 4090)
Rays Per Pixel 4
Ray Bundle Size 256
Undersampling 2

Preset for multiple high-end GPUs
Rays Per Pixel to 8
Ray Bundle Size to 512

IPR Rays Per Pixel(Value range = 1 to 8 ) is the number of rays that are traced per pixel for one image pass. The greater the value the smoother the image, plus it increases GPU usage on high-end cards like the RTX 3090
IPR Ray bundle size(value range = 128 to 512)
Undersampling value is personal preference, feel free to experiment with. More on using interactive rendering in V-Ray GPU here

Production Rendering,

• Use the least amount possible of Render Elements, they have a big impact on performance. For IPR you can disable them until final frame rendering.
  This graph shows the impact of Render Elements on GPU usage, in this example there are 25 Render Elements like Cryptomatte, Denoiser, Lights Selects and so on.
  
• Check the log for this warning, this happens when V-Ray GPU doesn't have enough GPU memory for rendering the scene. This affects performance considerably. You need to avoid this warning by optimizing GPU memory usage for your scene using my guide here
  

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

  
• If you just changed your GPU drivers or V-Ray build or your hardware, V-Ray GPU will compile rendering kernels for the first time it runs. This takes around 20 to 30 seconds during which the GPU usage will be low. This happens only one time after changing hardware or drivers or V-Ray versions
• Use Progressive Image Sampler for rendering, it is usually faster and more memory efficient than Bucket mode.
  
• Don't use Low Priority option in the device selector if you have one or more high-end GPUs, this option is only meant to be used for mid and low-end GPUs that struggle with drawing the UI during rendering.
  
• GPU Light Cache now renders on GPU, and it is usually very fast taking a handful of seconds at the start of the frame. only one single GPU is used for calculating GPU LC, and GPU usage drops below 100% for a few seconds at the start of the render, this is normal.
• Avoid having empty space in your frame whenever possible, this affects performance and GPU usage
• Avoid mixing 2 different architectures of GPUs, specially if one of the GPUs is significantly faster than the other devices.
• Use the recommended driver for V-Ray GPU.
• You need to have a CPU with 6 physical cores at the very least, for high-end GPUs.
• Thermal throttling for GPU and/or CPU will affect performance and GPU usage. Ampere and Ada GPUs will throttle around 81 degrees C, good airflow and undervolting could help with that.
• If you still see low GPU usage with your scene, reach out to us by submitting a ticket.

Let me know if you have questions,

Best,
Muhammed