Solution Architecture

Train faster, scale autonomously

Scale without overhead. With Crusoe Managed Kubernetes and Managed Slurm, we handle the provisioning and failover so you can focus on the model, not the machinery.

Model training on Crusoe Cloud

Crusoe Cloud provides managed infrastructure services specifically engineered to satisfy the stringent requirements of contemporary AI workloads. We integrate substantial computational capacity with adaptable orchestration capabilities to guarantee the efficient and dependable completion of your training runs.

Managed Orchestration with Crusoe Managed Kubernetes

Rapid cluster provisioning
Rapid cluster provisioning

Deploy high-performance GPU clusters via Crusoe Managed Kubernetes (CMK) or Crusoe Managed Slurm (CMS) in minutes. Built-in NVIDIA drivers with Quantum-2 InfiniBand networking and full AMD GPU support via ROCm ensure immediate compatibility across hardware configurations with no manual environment setup required.

Topology-aware orchestration
Topology-aware orchestration

The scheduler automatically maps jobs to the optimal node topology across your cluster. Native support for PyTorch, Kubeflow, and Ray means distributed training jobs are placed and balanced efficiently, with automated re-queuing maintaining workload continuity across interruptions.

Fault-tolerant scaling
Fault-tolerant scaling

Scale to thousands of GPUs with near-linear efficiency. AutoClusters automates node recovery and job resumption so workloads keep running without manual intervention, while Crusoe Command Center provides real-time cluster health visibility across the full fleet.

Ready to
run Slurm
Ready to
run Slurm

Crusoe Managed Slurm clusters come pre-configured with Slurm workload manager, NVIDIA drivers, and job scheduler, so your team can submit jobs immediately without standing up or tuning the environment. Whether you're migrating existing HPC workloads or starting fresh, CMS drops into familiar Slurm workflows with no retraining required.

Crusoe benefits for model training

Crusoe Cloud provides a fully managed environment that allows engineers to focus on code and model development, turning infrastructure into a competitive advantage.

1

Never lose a job to hardware failure

AutoClusters continuously monitors your cluster health, automatically detecting hardware failures and replacing nodes with hot spares in real-time, so training runs recover without engineer intervention.
2

Deploy and iterate faster

Spin up reproducible, NVIDIA NGC™-ready or AMD ROCm™ software environments in minutes using the Crusoe CLI or Terraform providers for Infrastructure as Code, so your team ships experiments, not tickets.
3

Govern and manage your training jobs and tasks

Ingest massive datasets into S3-compatible object storage or high-performance block storage over our global private backbone, so your data is ready the moment your next run starts. Then monitor cluster health, track job status, and maintain full operational visibility across your fleet all from Crusoe Command Center.
4

Train at scale without compromise

Built on NVIDIA and AMD reference architectures, Crusoe's infrastructure delivers the reliability and peak performance demanding AI initiatives require with InfiniBand and Infinity Fabric supplying the bandwidth large-scale distributed training demands.

Crusoe Managed solutions at a glance

Feature
Using Crusoe Managed Kubernetes or Crusoe Managed Slurm
Who it’s for
App Developers, Startups, Enterprise API Consumers
Setup time
Instant
Scaling
Automatic with AutoScaler enabled
Self-healing
Automatic with AutoCluster enabled
You can also build your own clusters with Crusoe VMs. Learn even more about that
here