26.10.0#
Release Date: October 2026
Omniverse Streaming on NVCF 26.10.0 extends the Reference Architecture to include customer-managed on-premises Kubernetes clusters, updates the platform to NVIDIA Cloud Functions (NVCF) 1.0.3, un-gates the Low Latency Streaming (LLS) components and documentation to make them publicly available, and replaces GXCache with the Omniverse Derived Cache (OVDC) as the GPU shader cache. The Agentic AI Skills also gain on-premises deployment and in-place upgrade workflows.
Added#
On-premises deployment: The Reference Architecture and the Agentic AI Skills now support customer-managed, upstream Kubernetes clusters in your own data center, with the same TCP TURN fallback for clients on networks that block UDP as on AWS and Azure. See the new On-Premises Deployment Guide and the NVIDIA-Omniverse/streaming-self-hosted repository.
In-place upgrades through the Agentic AI Skills: The skills gain an upgrade workflow that plans, applies, and validates an in-place upgrade of a running stack, including the NVCF control plane and compute plane, without redeploying from scratch, and recommends backups for stateful components before each apply step. See the NVIDIA-Omniverse/streaming-self-hosted repository and the Upgrading section below for the NVCF upgrade path.
LLS deployment guide: The Low Latency Streaming section now has an Overview and a Deployment guide that covers mirroring the LLS artifacts, the UDP load balancer, Helm values for the streaming proxy and the optional TCP or TLS TURN relay, installation, and verification.
Changed#
NVCF platform 1.0.3: (Omniverse Streaming on NVCF 26.10.0 uses NVCF 1.0.3)
Self-Hosted stack 1.0.1
Compute Plane stack 1.0.0
Observability stack 1.0.0
Documentation: Starting with NVCF 1.0.1, the three Helm stacks are versioned and released independently, and the NVCF documentation has been updated to reflect this:
Overview (Shared guides)
Self-Hosted Stack (Control Plane installation and operations)
Compute Plane Stack (GPU cluster setup)
Important
The NVCF documentation links in 26.10.0 point towards the 1.0.3 documentation. For the platform changes since 0.6.1, see the NVCF release notes; releases 1.0.2 and 1.0.3 are documentation-only patches, and the release 1.0.1 notes describe the stack changes.
LLS components are publicly available: The Streaming Proxy, TURN service, and the LLS Helm chart are un-gated and publicly published on NGC; see the Artifact Manifest for versions and locations.
Omniverse Derived Cache (OVDC) 6.0.1: The Derived Data Cache Service (DDCS) is renamed to Omniverse Derived Cache and updated to 6.0.1. OVDC is the recommended shader cache. See NVCF and Omniverse Caches and the OVDC documentation.
Omniverse Content Cache (OVCC) 3.0.2: The USD Content Cache Service (UCC) is renamed to OVCC and the container is updated to 3.0.2, which carries security fixes with Helm chart 4.0.4. The chart still defaults to container 3.0.1, so set
image.tag: "3.0.2"in your Helm values. See the Artifact Manifest.GXCache is deprecated: GXCache, the distributed shader cache used by earlier releases, is deprecated and removed from the Artifact Manifest. Starting with 26.10.0, use the Omniverse Derived Cache (OVDC) for shader caching; new deployments should not install GXCache. See NVCF and Omniverse Caches for additional information.
Agentic AI Skills: The skills and their component manifest are updated to deploy this release, including the on-premises provider and the upgrade workflows. See Agentic AI Skills and the NVIDIA-Omniverse/streaming-self-hosted repository.
Storage APIs: Storage Service 1.2.0 and Discovery Service 2.3.11.
Known Issues#
Function autoscaler: The NVCF function autoscaler is not deployed by the Agentic AI Skills in this release, so a streaming function will not automatically scale beyond its minimum instance count. Information on manual deployment can be found here: NVCF function autoscaler.
Chart image defaults: Some Helm charts in this release have image defaults that are older than the versions listed on the Artifact Manifest. Deployments made with the Agentic AI Skills use the manifest versions. When installing the self-hosted stack or individual charts directly, apply the image overrides documented in the Artifact Manifest, then verify the running images against the manifest.
Security#
Component updates in this release remediate known security vulnerabilities in the Omniverse Streaming components and the NVCF platform. Deploy the versions listed in the Artifact Manifest to receive the fixes.
Upgrading#
To upgrade an existing 26.08.0 deployment (using NVCF 0.6.1), first complete the NVCF 0.6.1 to 0.6.2 upgrade, then follow the NVCF 0.6.2 to 1.0.1 upgrade guide, which covers the ordered Cassandra, OpenBao, and service migrations and the compute-cluster re-registration. Then update the Reference Architecture components to the versions in the Artifact Manifest. The Omniverse Derived Cache replaces the DDCS chart with the ovderivedcache chart, so update the Kit parameters in the Deploying guide to the new service endpoint. Existing GXCache deployments can keep running, but migrating GPU shader caching to OVDC is recommended; a function should use one or the other. Use the in-place upgrade skills in the NVIDIA-Omniverse/streaming-self-hosted repository to apply the upgrade.
Component Versions#
The NVCF Compatibility Matrix lists the stack versions qualified to run together. For the Reference Architecture components, see the Artifact Manifest. For the NVCF platform artifacts, see the NVCF Artifact Manifest.