Crusoe deepens NVIDIA partnership with Run:ai open-source release
Crusoe welcomes the open-source release of NVIDIA Run:ai. Crusoe Managed Kubernetes (CMK) is a partner-compatible distribution: Run:ai schedules and shares GPU capacity across teams, while CMK manages the Kubernetes and GPU infrastructure underneath.
.avif)
Crusoe today welcomed NVIDIA’s open-source release of NVIDIA Run:ai. The release supports greater transparency, extensibility, and community participation in the software customers use to schedule AI workloads and allocate GPU resources.
Crusoe Managed Kubernetes (CMK) is already listed as a partner-compatible Kubernetes distribution for NVIDIA Run:ai. Building on their existing collaboration, Crusoe and NVIDIA provide customers with a validated deployment path that combines NVIDIA Run:ai’s workload orchestration capabilities with CMK’s managed Kubernetes and GPU infrastructure.
The open-source release gives customers new opportunities to inspect, extend, and contribute to NVIDIA Run:ai while continuing to use the CMK integration available today.
“Customers want to get more work out of their GPUs without taking on more infrastructure to manage. NVIDIA Run:ai gives teams control over how GPU capacity is shared and scheduled, while CMK handles the underlying cluster operations. We support making NVIDIA Run:ai open source because customers should be able to understand, extend, and help improve the software that determines how their compute is used.”
- Avinash Mandava, Product Manager at Crusoe
Managed GPU orchestration, built for AI training
GPU clusters often serve multiple teams with different workloads and competing demands for compute. Assigning fixed GPU capacity to each team can leave resources idle in one part of an organization while jobs wait elsewhere.
NVIDIA Run:ai provides workload scheduling and resource management on Kubernetes. Platform teams can establish project quotas and priorities, share available capacity across teams, and support training, inference, and interactive workloads. GPU sharing allows suitable workloads to use a fraction of a GPU, while larger workloads can use multiple GPUs across nodes.
CMK provides the managed Kubernetes foundation for these workloads. Crusoe provisions and manages Kubernetes control planes and GPU node pools, integrates GPU, networking, storage, and autoscaling components, and provides infrastructure health monitoring and remediation. By managing components like the NVIDIA GPU Operator, NVIDIA Network Operator, and Crusoe CSI storage driver, CMK reduces the cluster setup and maintenance work customers do before deploying NVIDIA Run:ai.
Crusoe and NVIDIA validated the deployment and representative workloads, including cluster registration, distributed multi-node PyTorch training, fractional GPU sharing, and inference serving. Customers can configure NVIDIA Run:ai for their teams and workloads without assembling and validating the underlying Kubernetes infrastructure themselves.
Building an open GPU ecosystem together
The open-source release of NVIDIA Run:ai gives customers visibility into the software responsible for scheduling their workloads and allocating GPU resources. Organizations can inspect its behavior, extend it for their requirements, integrate it with existing tools, and contribute improvements to the broader community.
Crusoe supports this approach through contributions to open-source GPU infrastructure projects, including NVSentinel and the Crusoe topology provider in NVIDIA topograph. Contributing upstream makes these improvements available beyond individual customer deployments and allows infrastructure providers, developers, and users to improve shared capabilities together.
For Crusoe, open source and managed infrastructure are complementary. Customers should be able to benefit from open software without having to operate every layer of the infrastructure themselves. NVIDIA Run:ai on CMK brings those benefits together: visibility and flexibility in workload orchestration, with managed operations for the Kubernetes and GPU infrastructure beneath it.
Deployment and ongoing compatibility
Get up and running fast: Crusoe's step-by-step guide walks through installing and configuring NVIDIA Run:ai on new or existing CMK clusters. It covers storage, networking, TLS, load balancing, and inference-serving configuration.
As CMK and NVIDIA Run:ai evolve, so does our partnership. Crusoe and NVIDIA are committed to keeping compatibility validation and deployment guidance current.
Ready to bring NVIDIA Run:ai to your teams? Get started on Crusoe Cloud and create a CMK cluster.
Resources
Frequently
asked questions
What is NVIDIA Run:ai, and why did NVIDIA make it open source?
NVIDIA Run:ai is workload orchestration and GPU scheduling software for Kubernetes. It lets platform teams set project quotas and priorities, share GPU capacity across teams, and support training, inference, and interactive workloads. NVIDIA open-sourced it to give customers greater transparency, extensibility, and the ability to contribute to the software that manages their compute.
How does NVIDIA Run:ai work with Crusoe Managed Kubernetes (CMK)?
CMK is a partner-compatible Kubernetes distribution for NVIDIA Run:ai. Run:ai handles GPU sharing, quotas, and workload scheduling, while CMK manages the underlying Kubernetes control planes, GPU node pools, NVIDIA GPU Operator, and NVIDIA Network Operator.
What workloads has Crusoe validated for NVIDIA Run:ai on CMK?
Crusoe and NVIDIA validated cluster registration, distributed multi-node PyTorch training, fractional GPU sharing, and inference serving, so platform teams can deploy without assembling and validating the Kubernetes infrastructure themselves.
Does the open-source release change how customers use the existing CMK integration?
No. Customers can continue using the CMK integration available today. The open-source release adds the ability to inspect, extend, and contribute to NVIDIA Run:ai on top of that existing deployment path.
How do I get started deploying NVIDIA Run:ai on Crusoe?
Crusoe provides a step-by-step guide covering installation and configuration on new or existing CMK clusters, including storage, networking, TLS, load balancing, and inference-serving setup.
Does Crusoe contribute to open-source GPU infrastructure beyond NVIDIA Run:ai?
Yes. Crusoe contributes to open-source projects including fastokens, Kubernetes, NVSentinel and the Crusoe topology provider in NVIDIA topograph, making improvements available beyond individual customer deployments.
