lambda-labs-gpu-cloud

To launch on-demand GPU cloud instances for ML training, use the Lambda Cloud API to provision scalable single-node or multi-node clusters. You receive dedicated resources with pre-installed ML software and SSH access.

Yes, you can run distributed training using optional 1-Click Clusters that are Slurm-ready. This allows you to scale multi-node training workloads across regions with pre-configured environments.

Yes, Lambda Labs GPU cloud supports persistent storage across sessions. Your ML training datasets and inference models remain intact on the filesystem even when instances are stopped or restarted.

No, you do not need to install your own ML frameworks because instances come with the pre-installed Lambda Stack. This provides the necessary software for ML training and inference immediately upon connection.

The best way to connect to a GPU cloud instance for inference workloads is through simple SSH access. After launching from the console, you can securely connect to your dedicated resources.

You should use 1-Click Clusters instead of single-node GPU instances when handling large workloads that require distributed training. Single-node instances are sufficient for smaller ML training or inference tasks.