Vultr is thrilled to announce major updates to Vultr Clusters, our self-service solution for deploying multinode clusters without manual requests or reservations. Vultr Clusters now supports Vultr Bare Metal CPU servers and Vultr Cloud Compute instances, making Vultr the first cloud provider to offer on-demand, self-service GPU or CPU clusters harnessing the latest GPUs and CPUs. Vultr Clusters also now supports GPU cluster head nodes for simplified control and management, and streamlined controls for adding Vultr File System storage instances.
These developments enable Vultr customers to deploy full-stack compute infrastructure for a wide range of workloads, from artificial intelligence and machine learning to data analysis and scientific modeling.
Architect infrastructure with GPU or CPU clusters
The addition of CPU support for Vultr Clusters provides customers with the ability to leverage flexible configurations powered by Vultr’s latest-generation Vultr Bare Metal or Vultr Cloud Compute infrastructure. Whenever compute requirements scale beyond a single server or virtual machine, Vultr Clusters enables quick provisioning of highly performant compute solutions for a broad set of workloads.
Like with Vultr Cloud GPU instances or GPU Vultr Bare Metal servers, setting up a cluster of Vultr CPU instances is a straightforward process. Simply provision a cluster in one of Vultr’s 33 global data center regions, select the type of Vultr Bare Metal servers or Vultr cloud instances to be included, then add new or existing instances of that type to the cluster (they must be in the same Vultr cloud data center region). Vultr’s powerful networking automation will take care of the rest, provisioning high-performance networking and creating a ready-to-go cluster in just seconds. For private workloads, clusters can be added to a Vultr Virtual Private Cloud (VPC), or a new one can be created in a few easy steps.
Management is likewise a simple process, and compute instances can be added or removed from the cluster at will. Cluster resource usage is logged and can be viewed through the Vultr Console, with CPU usage and memory usage monitored and visualized. (Further metrics are made available for GPUs, including GPU-specific GPU, memory, and power utilization; fabric congestion; and more!)
Simplified creation and management for GPU clusters with head nodes – and Slurm or Kubernetes
The addition of cluster head nodes streamlines control and management for Vultr GPU clusters by providing a preinstalled cluster workload scheduler to handle these tasks. Cluster administrators simply select the cluster scheduler (Slurm or Kubernetes), the head node size and type, the type of GPU for the cluster, and the number of GPUs in the cluster. Vultr’s platform automation then handles the rest, including provisioning high-performance RoCEv2 or NVIDIA InfiniBand GPU fabric (as applicable) to the cluster nodes. By controlling the cluster through the head node, cluster admins do not need to manually provision or manage cluster nodes.
Cluster head nodes include Prometheus and Grafana preinstalled for cluster visibility, logging the results of GPU node health checks.
Cluster-level storage made simple
Vultr File System, Vultr’s managed shared file storage system, was engineered to meet the needs of clusters, with multiple nodes able to read and write to files simultaneously. It is now possible to straightforwardly attach a Vultr File System instance to all cluster nodes directly through the Vultr Console or API with just a few clicks.
A powerful solution for compute needs
With the addition of CPU support, GPU cluster head nodes, and a simple process for attaching Vultr File System, Vultr Clusters offers even greater flexibility and options to suit cloud computing needs. With no manual requests or reservations required, Vultr Clusters makes it possible to create and manage high-performance compute clusters on demand and scale them as needed.
Ready to experience the enterprise-grade infrastructure delivered by Vultr Clusters? Get started deploying self-service multinode clusters of Vultr's CPU and GPU offerings.