NVLink FAQ

This article covers information on setting up OptiX and NVLink for usage with V-Ray. It examines their purpose and answers questions such as: 

• What is NVLink and what is it used for?
• How to set NVLink?
• How does V-Ray benefit from NVLink?
• Which NVidia driver is best for using V-Ray GPU?
• How to determine if a GPU device is compatible with V-Ray GPU?

Understanding NVLink

NVLink is NVIDIA’s high‑bandwidth, low‑latency interconnect that lets supported GPUs exchange data directly, bypassing the PCIe bottleneck.

• It provides fast peer‑to‑peer (P2P) communication between matching GPUs connected with a compatible NVLink bridge.
• In multi‑GPU rendering, this can reduce inter‑GPU data transfers through the CPU and can improve scaling on large scenes.
• Allows multiple graphics cards to share their available memory. This is beneficial for GPU rendering, since it allows more memory to be used for large scenes that otherwise would not fit on a single device. 
• Support depends on the exact GPU model and generation (for example, many RTX professional cards support NVLink; newer GeForce RTX 40‑series cards like the 4090 do not).

Compatibility and connector types

• NVLink can only bridge identical graphics cards.
• Connector type depends on the physical spacing between GPUs and the GPU family (GeForce vs. Quadro/NVIDIA RTX). Use the appropriate NVLink bridge for your setup.

 
Performance considerations

• Expect a slight decrease in render speed with NVLink enabled due to the overhead of accessing memory across GPUs.

• To reduce performance impact, frequently used data is duplicated on all GPUs. As a result, memory is not fully pooled; some data resides on each device.

   

Setting up NVLink for GeForce Devices

• Requirements:

  1. Identical GeForce devices with NVLink support
  2. Enabled SLI-mode from the NVIDIA Control Panel

  3. SLI Ready Motherboard

     Note: GeForce devices support only a single NVLink connection per workstation.

• Enabling NVLink on GeForce

  1. Open the NVIDIA Control Panel.
  2. Go to "Configure SLI, Surround, and PhysX".
  3. Under "SLI configuration", select "Maximize 3D performance" and click Apply. 

      

     

      

  4. To verify that the setup is complete, run a test render and check the V-Ray log (located in the V-Ray Frame Buffer) for the following message:

     NVLink Detected!
     Device[0] (<GPU model>) is using NVLink with ID: 0
     Device[1] (<GPU model>) is using NVLink with ID: 0
     Number of CUDA devices: 2

Setting up NVLink for Quadro Devices

• Requirements

  1. Identical Quadro devices with NVLink support
  2. All Quadro GPUs should run in a special TCC driver mode
  3. Monitors should not be attached to a device that is running in TCC mode
  4. A dedicated graphics card should be installed for a monitor instead

  5. SLI Ready Motherboard.

      

• Enabling NVLink on Quadro Devices by switching the driver mode to TCC

  Switching the driver mode to TCC is done through the nvidia-smi command-line utility. The utility is installed along with the NVidia drivers (default location: C:\Program Files\NVIDIA Corporation\NVSMI\). 

  Step by step:

  1. Open the Windows command prompt by pressing Start and typing cmd.

  2. Enter the following command to switch the default directory to the directory of the nvidia-smi tool:

     cd C:\Program Files\NVIDIA Corporation\NVSMI
      

  3. Use this command to find your GPU IDs:

     nvidia-smi -L

     The result is printed in the command prompt:
     
     GPU 1: GeForce RTX 2080 (UUID: GPU-21600cd2-5f9f-fb16-e1dd-e45e99e85a01)I
     
     In this example, the {GPU_ID} is 1.

      

  4. The command to change the driver to TCC mode is: 
     
     nvidia-smi -g {GPU_ID} -dm {0|1}
     
     The value for -dm is either:
      0 = WDDM or 1 = TCC.
      

     Example command to switch the driver mode to TCC on GPU ID 1:

     nvidia -smi -g 1 -dm 1

     

      

  5. To verify that NVLink is enabled, run a test render and check the V-Ray log (located in the V-Ray Frame Buffer) for the following message:

     NVLink Detected!
     Device[0] (<GPU model>) is using NVLink with ID: 0
     Device[1] (<GPU model>) is using NVLink with ID: 0
     Number of CUDA devices: 2

Additional considerations

• NVidia GPU architecture requirements
• Compute capability requirements

• Recommended GPU drivers

  Learn more in V-Ray GPU minimal and recommended system requirements.