NVIDIA IndeX Remote Client#

This extensions can be used to run a remote client for distributed cluster/network rendering with NVIDIA IndeX.

An IndeX rendering cluster consists of one “head node” and multiple “remote” worker nodes:

  • The head node will run a normal Kit session with RTX / NVIDIA IndeX Compositing (omni.rtx.index_composite) or RTX Scientific (omni.index.kit.rtx_scientific). Clustering can be enabled and configured in “Render Settings / Scientific (IndeX) / IndeX Cluster / Network Settings”.

  • On the remote nodes, Kit will typically run headless and with only this extension loaded. See omni.app.nvindex-remote for an example. Network configuration can be applied via command line arguments.

Cluster Configuration#

  • Network Mode (/nvindex/networkMode): tcp, udp, off (default)

  • Discovery Address (/nvindex/discoveryAddress): <address>[:port] (selects discovery server; only for TCP mode)

  • Multicast Address (/nvindex/multicastAddress): <address>[:port] (only for UDP mode)

  • Cluster Interface Address (/nvindex/clusterInterfaceAddress): <address>[:port] or if://<interface-name> (selects network interface for outgoing packets)

  • Minimum Cluster Size (/nvindex/minimumClusterSize): Rendering will pause until this many nodes have joined the cluster.

  • Service Mode (/nvindex/clusterServiceMode): rendering_and_compositing, rendering, compositing, none (service mode of the node)

  • Cluster Identifier (/nvindex/clusterIdentifier): Optional identifier, must be the same on all hosts.

For more information about the configuration options, see the documentation of INetwork_configuration in the NVIDIA IndeX API reference guide: https://raytracing-docs.nvidia.com/nvindex/reference/dice/classmi_1_1neuraylib_1_1INetwork__configuration.html

On this page