Mirantis is positioning its k0rdent AI platform to capture the growing demand for multi-tenant AI infrastructure by securing certification through the NVIDIA-Certified Hypervisors program. This validation targets enterprise AI factories and cloud service providers who require high-performance GPU virtualization without the traditional overhead of software abstraction. By achieving this certification, Mirantis aims to reduce the engineering friction associated with designing and benchmarking production-ready environments. The move signals a strategic push to provide a unified control plane for both virtual machines and Kubernetes workloads, specifically optimized for NVIDIA-accelerated hardware architectures.
k0rdent AI Performance on NVIDIA Hardware
Mirantis is leveraging extensive benchmarking to demonstrate that k0rdent AI virtualization can achieve performance within 5% of comparable bare-metal deployments. This validation was conducted across specific NVIDIA hardware tracks, including the NVIDIA GB200 NVL72 and NVIDIA HGX-based systems. The company is marketing this capability as a way to provide predictable, near bare-metal performance while maintaining the operational flexibility necessary for large-scale AI operations. By utilizing a topology-aware scheduler, the k0rdent platform attempts to manage the physical connections between GPUs, CPUs, network interfaces, and storage. This approach is designed to place workloads intelligently across NVIDIA NVLink fabrics, NUMA domains, and NVIDIA InfiniBand/RoCE networking to prevent the communication bottlenecks that often degrade AI training and inference performance.
Scaling Multi-Tenant AI Factories
The certification addresses a critical requirement for modern "AI factories": the ability to securely partition expensive compute resources among multiple users or projects. Rather than dedicating entire GPU servers to single tasks, k0rdent AI enables flexible GPU allocation and secure tenant isolation through virtual machines. Mirantis is positioning this as a method to improve infrastructure efficiency and GPU utilization for cloud providers and large enterprises. The platform seeks to simplify the management of diverse workloads by allowing organizations to run AI models, traditional enterprise applications, and databases through a single management interface. Currently, k0rdent AI Virtual Machine as a Service (VMaaS) is available as a technical preview, functioning alongside the company's existing Bare Metal as a Service (BMaaS) and Kubernetes orchestration capabilities to support a broader AI lifecycle.
Key Takeaways
- k0rdent AI achieved performance within 5% of bare-metal deployments during NVIDIA GB200 NVL72 and NVIDIA HGX validation testing.
- Mirantis is one of the inaugural independent software vendors (ISVs) participating in the NVIDIA-Certified Hypervisors program.
- The k0rdent platform utilizes a topology-aware scheduler to manage workloads across NVIDIA NVLink fabrics and InfiniBand/RoCE networking.
TechInsyte's Take
In our view, Mirantis is making a calculated bet that the future of enterprise AI lies in "neocloud" models where high-density GPU sharing is mandatory for economic viability. By securing NVIDIA certification, they are attempting to strip away the skepticism often directed at virtualization in high-performance computing. If the 5% performance delta holds true in real-world production environments, k0rdent could become a vital layer for enterprises trying to balance the massive capital expenditure of NVIDIA hardware with the need for granular, multi-tenant resource control. This isn't just about virtualization; it is about providing the orchestration layer necessary to turn raw silicon into a scalable, multi-tenant service.
Questions & Answers
How does k0rdent AI minimize performance loss during GPU virtualization?
The platform employs a topology-aware scheduler that recognizes the physical layout of GPUs, CPUs, and networking. By aligning workloads with NVIDIA NVLink fabrics and NUMA domains, it aims to maintain high-speed communication and avoid the bottlenecks typically seen in standard virtualization.
What specific NVIDIA hardware was used for the performance validation?
Mirantis conducted its benchmarking and validation tests on NVIDIA GB200 NVL72 and NVIDIA HGX-based hardware tracks to ensure the k0rdent stack could meet high-performance requirements.
Can k0rdent AI manage both containers and virtual machines?
Yes, the platform is designed to manage both virtual machines and Kubernetes workloads through a unified control plane, allowing for the simultaneous operation of AI models and traditional enterprise applications.
Is the Virtual Machine as a Service (VMaaS) feature fully production-ready?
No, k0rdent AI Virtual Machine as a Service (VMaaS) is currently available only as a technical preview.
Source: Businesswire
