Kit 109.0 Release Highlights#

Release Date: November 2025

This targeted Feature Branch release of Omniverse Kit 109.0 focuses on critical platform enhancements for enterprise customers, specifically benefiting robotics, manufacturing, and simulation workflows. Key upgrades deliver improved performance and features essential for digital twins and industrial automation.

Major Advancements Include:

  • Platform Reliability and Stability: Foundational string safety upgrades and critical fixes enhance stability, including for use cases on DGXC.

  • Enterprise Scalability: Improvements in observability and logging capabilities are introduced to support large-scale cloud and enterprise deployments.

  • Simulation and Rendering: Significant feature and performance upgrades are delivered for sensor simulation, vital rendering workflows, and point cloud streaming scalability.

Note: Several platform-level changes in this release may necessitate updates to custom extensions and workflows.

What’s New in Kit 109.0#

Major Platform Updates#

NumPy 2.3.1 (NumPy 2.x)#

Kit now uses NumPy 2.3.1 by default (upgraded from NumPy 1.x), enabling seamless integration with Isaac-Newton and modern robotics frameworks. This update ensures compatibility across the expanding Omniverse robotics ecosystem and brings improved performance and new capabilities from the NumPy 2.x series.

CUDA 12.4.1#

Updated CUDA SDK to version 12.4.1, providing the latest GPU compute capabilities and optimizations.

Note: This update requires driver version >=550.54.15 (Linux) or >=551.78 (Windows). Please verify your driver versions before upgrading.

Memory Management Improvements#

mimalloc as Default Allocator (Windows): Kit on Windows now uses mimalloc as the default memory allocator, delivering improved memory performance, better fragmentation handling, and enhanced application stability. For compatibility scenarios, the system allocator remains available as a fallback via kit-sysalloc.exe.

Performance & Infrastructure#

Fabric Scene Delegate Now Default#

Fabric Scene Delegate (FSD) is now enabled by default in Kit 109.0. Applications no longer need to explicitly enable FSD in .kit configuration files.

Fabric Performance Breakthroughs#

Multiple significant optimizations to the Fabric scene representation layer deliver measurable performance improvements:

  • Zero-Copy Memory Sharing: Sharing Fabric memory with FSD and RTX through VtArray’s Foreign Source Data enables up to 50% memory savings on stages where mesh or point data is dominant.

  • Fine-Grained Copy-on-Write: Shallow copy array pointers instead of deep copying every array, dramatically improving array attribute operations

  • Trivially Copyable Handles: Implemented trivially copyable fabric::Token and fabric::Path handles for correct handle lifetimes when storing in attribute memory

Breaking Change: String Token Safety in Fabric introduces API/ABI breaking changes to fabric::Token and fabric::Path handles. Review the Migration Guide for details on updating your extensions.

String Safety & Security#

Comprehensive string safety upgrades throughout Kit and Carbonite replace unsafe string functions with secure alternatives. New safe-only interfaces for ISettings, IError, and IDictionary prevent security vulnerabilities from unsafe string operations.

User Experience Improvements#

Customizable Camera Manipulator#

New flexible camera controller system enables developers to implement application-specific camera navigation and manipulation behaviors. Key bindings for camera gestures—panning, tumbling, looking, flight mode, and zooming—can be remapped to match user preferences or legacy workflows through Python extensions.

This architectural improvement enables custom camera control tailored to different industries and projects, eliminating the need for users to relearn navigation methods when switching between applications. For implementation details, see the omni.kit.viewport.docs documentation.

Observability & Telemetry#

OpenTelemetry Integration#

Full integration of OmniMetrics into Carbonite SDK with enhanced metrics functionality, new helpers and tools, and support for OTEL_RESOURCE_ATTRIBUTES environment variable. Enterprise deployments can now export metrics to OTLP endpoints for comprehensive monitoring, with improved logging guidelines and documentation.

For configuration details and implementation guidance, see the Carbonite SDK Observability documentation.

Platform & Extension Updates#

Events 2.0 Transition: All Kit App Template and sample extensions have migrated to the new Events 2.0 APIs for consistency and improved event handling.

Rendering & Graphics#

Kit 109.0 brings significant rendering and sensor simulation improvements that enhance visual fidelity, performance, and workflow flexibility for robotics and digital twin applications. These updates enable physically accurate lighting simulations, customizable sensor behavior, and optimized memory usage for large-scale scenes.

Physical Lighting: RTX now fully supports the UsdLux 25.05 specification with new photometric APIs for real-world lighting units, enabling physically accurate lighting simulations and seamless transfer between compliant OpenUSD applications.

Sensor Simulation: Customizable camera ISP pipeline with runtime parameter updates for camera, lidar, and radar sensors. New rolling shutter support and runtime motion raytracing toggle that automatically activates only when needed, avoiding the ~40% performance overhead.

Advanced Features: Reworked mesh lights sampling reduces memory usage and improves visual fidelity; new custom projections for projector lights allow almost any projection to be encoded.

Note: For comprehensive details on all rendering features, improvements, and fixes, see the RTX Renderer 109.0 Release Notes.

Breaking Changes and Compatibility#

This release contains changes that may require extension and workflow updates:

  • String Safety APIs: Introduction of string_view-based interfaces may require code updates

  • Extension Dependencies: Updated dependency requirements for several core extensions

  • CUDA Requirement: Minimum driver version updates required

Important: Thoroughly test custom extensions in a staging environment before deploying to production.

Additional Resources#