FAQ: L40s on Crusoe Managed Slurm

Introduction

Crusoe Managed Slurm is a fully managed HPC orchestration layer purpose-built for large-scale, multi-node GPU training on InfiniBand-connected SXM GPUs such as H100 and B200. It is not a general-purpose GPU compute platform — it is optimized for tightly coupled distributed workloads that require high-bandwidth interconnects and coordinated job scheduling across many nodes.

This FAQ covers GPU instance type availability on Crusoe Managed Slurm and explains where L40S workloads fit in the broader Crusoe product lineup — including which products are the right fit for inference, fine-tuning, and graphical workloads that don't benefit from InfiniBand.

Question 1: Can I use L40S GPUs on Crusoe Managed Slurm?

L40S instances are currently outside the scope of Managed Slurm, which is optimized for tightly coupled distributed training on InfiniBand-connected SXM GPUs. L40S GPUs are a strong fit for inference, fine-tuning, image generation, and professional graphics workloads, and are available across other Crusoe products best suited for those use cases.

Question 2: Where can I run L40S workloads on Crusoe?

L40S instances are available on the following Crusoe products:

• Crusoe Managed Kubernetes (CMK) — recommended for teams wanting flexible, Kubernetes-native orchestration for L40S workloads.
• On-demand VMs — available from l40s-48gb.1x (single GPU) to l40s-48gb.10x (10 GPUs) for teams that prefer direct VM access.

Question 3: What are the supported GPU SKUs on Managed Slurm?

Crusoe Managed Slurm is purpose-built for large-scale, multi-node GPU training workloads requiring high-bandwidth InfiniBand interconnects. The GPU types currently available are:

Additional GPU types may be supported in the future. Check the Supported GPU Types page for the latest supported SKUs.

Additional Resources

• Supported GPU SKUs on Managed Slurm
• Crusoe Managed Slurm Overview
• Crusoe Managed Kubernetes