Data Center / Cloud

Supercharging Live Media Workflows with NVIDIA NIM and NVIDIA Holoscan for Media

A picture of a person sitting in front of audiovisual equipment.

NVIDIA Holoscan for Media is an NVIDIA-accelerated platform designed for multi-vendor live production and AI. It will be showcased at GTC, highlighting NVIDIA NIM, AI SDKs, and microservices that enhance live production workflows. 

Built on Kubernetes, the container orchestration platform simplifies media timing, synchronization, and management through NVIDIA components such as the GPU and network operators, performance tuning for ST 2110, and monitoring stack. Reference applications use the NVIDIA Rivermax, NVIDIA NMOS, and NVIDIA DeepStream SDKs, to enable you to run real live media workflows from day one. 

Holoscan for Media accelerates the industry’s adoption of AI, by connecting next-generation models with uncompressed live media pipelines to deliver innovative viewer experiences and production processes.

Revolutionizing AI video and audio pipelines with NVIDIA NIM 

NIM provides reference AI models with optimal inference performance and reliability as downloadable microservices deployed on Kubernetes through Helm charts. Holoscan for Media was built on the exact same architecture to enable multi-vendor applications deployment and facilitate uncompressed media support through ST 2110 and NMOS. 

At GTC 2025, we are showcasing live how AI SDKs, microservices, and code samples, can power media production:

  • Reference code sample to connect ST 2110 audio streams to the NVIDIA Riva Automatic Speech Recognition (ASR) NIM microservice for live transcription and search.
  • Person detection and recognition, with next-generation upscale (super resolution, denoising, and deblurring) for virtual cameras from ST 2110 video.
The diagram shows the GTC 2025 demo for Holoscan for Media featuring AI models and NIM microservices, connected partner applications powered by NVIDIA technologies, inputs/outputs, and other resources all hosted on on-premises servers. 
Figure 1.  Live media workflow integrating AI models

NVIDIA is building the next stage of AI NIM microservices and SDKs for live media.

More for live media developers

A new release of Holoscan for Media platform (version 25.4) also provides new tools:

  • Improved automation for single-node OpenShift production environment and support for more networking variants
  • Full monitoring stack with platform and application metrics and dashboards that can run on Red Hat OpenShift production environments as well as open-source Kubernetes for developers with NVIDIA Cloud Native Stack.
  • Updated developer deployment guide
  • Red Hat OpenShift version 4.16 and NVIDIA driver R570 support
A screenshot shows some of the preconfigured dashboards provided by the Holoscan for Media platform.
Figure 2. Resource utilization during a 200 GbE stress test

We performed extended 200 GbE benchmarks, including stress tests and stream redundancy, reaching 17 streams of uncompressed 4k60 video and 71 streams of uncompressed 1080p60 per interface. All were fully compliant with ST 2110 specs, providing unparalleled density. 

Last but not least, you can validate your application integration on a dedicated Holoscan for Media production environment hosted on NVIDIA LaunchPad. For access to the LaunchPad lab, contact  Holoscan-Media-Support@nvidia.com.

Get started with Holoscan for Media  

Explore the online documentation and download Holoscan for Media from NGC.

Take a deep dive on Holoscan for Media at GTC 2025! Join the Connecting NIM Microservices to Uncompressed Video Pipelines with Holoscan for Media session and find the demo on the show floor.  

Discuss (0)

Tags