- OpenNebula lowers cost barrier to local AI.
- Users to get benefit of NVIDIA’s Spectrum-X Ethernet networking.
- Ideal for multi-tenancy and sandbox testing.
OpenNebula Systems has been approved by NVIDIA as an orchestration platform that integrates with NVIDIA’s Spectrum-X Ethernet networking system. With a growing focus on infrastructure designed specifically for large artificial intelligence workloads, the stamp of approval positions OpenNebula as a cloud management platform suited to so-called AI Factories.
AI factories could be described as environments built to support intensive training and inference for large language models, where each procoss typically requiring different infrastructure to get optimal performance as models are built, and in day-to-day use.
OpenNebula manages private, hybrid, and edge cloud environments, and has a strong presence in multi-tenant deployments, including research institutions, service providers, and public sector organisations. In recent years, the platform has extended its portfolio to offer large-scale GPU-accelerated infrastructure in response to rising demand for control and sovereignty of AI-based systems.
The NVIDIA Spectrum-X Ethernet platform targets a what the company states is a weakness in traditional data centre networking in the context of AI workloads. LLM training and inference can be affected by latency and congestion, where jitter or packet loss increases training and inference times. Spectrum-X Ethernet is designed to address such constraints by the addition of optical connections (faster and consuming less power than copper) and Remote Direct Memory Access (RDMA), which allows data transfers to bypass the kernel’s networking stack for direct-to-memory communication between parallel nodes.
Under the validation announced by OpenNebula Systems, its control plane will integrate with Spectrum-X Ethernet to let compute, GPU, and network resources be orchestrated in its software-defined environments. The integration will support multi-tenancy, so multiple users or organisations can share the Spectrum-X fabric. Direct pass-through of GPUs and NVIDIA SuperNICs form part of the design to help reduce cross-talk between tenants.
The integration should reduce the complexity associated with deploying and running AI infrastructure at scale, with the OpenNebula control plane automating tenant provisioning, network configuration, and device attachment inside Spectrum-X environments. Such automation is intended to lower the risk of configuration errors and to make it easier for operators to deliver environments suited to concurrent AI workloads.
The OpenNebula control plane runs on NVIDIA Air, NVIDIA’s cloud-hosted environment for testing and validation of AI infrastructure, so users are able to evaluate an OpenNebula and Spectrum-X integration without committing to physical hardware. OpenNebula describes such simulations as useful proofs-of-concept in large AI factory designs. While this does not replace production testing, it provides a lower-cost entry point for organisations assessing different architectural options.
OpenNebula’s leadership highlights support for NVIDIA Grace Blackwell and Grace Blackwell Ultra platforms using Spectrum-X networking, with its offered orchestration tooling as an alternative to loosely-integrated components. NVIDIA terms the accreditation as an example of cloud-native tooling adapting to the demands of AI infrastructure in multi-tenant settings.
The announcement is particularly relevant to the companies’ European clients, where enterprises and service providers face sovereignty concerns around AI infrastructure. Platforms running AI workloads on-premises or in private environments have gained attention recently as an alternative to managed public cloud services from US hyperscalers.
OpenNebula Systems has become one of the more common choices for end-users wishing to move away from VMware since the latter’s acquisition by Broadcom and the ensuing licensing changes and Broadcom’s stringent focus on revenue generation. OpenNebula reports more than 5,000 deployments worldwide, including environments with thousands of hosts, many with large GPU fleets.
(Image source: “Liège / Luik / Lüttich” by Bert Kaufmann is licensed under CC BY 2.0.)
Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.
TechHQ is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
