All running on a managed infrastructure, laboratories, universities, and private entities can now load and run their HPC jobs on our HPC Job Scheduler Clusters.
We offer a validated solution tailored for AI and ML tasks. This architecture supports scaling from one DGX to nine DGXs within a single rack.
Featuring up to 72 NVIDIA GPUs and NVMe storage back-end, it delivers over 90 GB per second data throughput in a single 2u form factor.
Reach out to us to discover more or Test Drive it firsthand
With tens of DGXs and linear scalability, a successfully testing and integrating with the world’s flagship clustered-parallel filesystem, Weka FS, storage stack, the NVIDIA DGX SuperPOD provides a robust solution for scalable AI development and deep learning tasks. Key features include:
SUNK unifies security, scalability, performance, and observability to deliver a high-performance Slurm experience purpose-built for AI research clusters on CoreWeave’s optimized infrastructure.
SUNK User Provisioning automatically synchronizes POSIX and Slurm users with CoreWeave IAM or any supported Identity Provider, such as Okta or Google Workspace. User and group updates propagate instantly, eliminating manual configuration, reducing operational risk, and accelerating secure researcher access to compute.
Workloads and data move fast and friction-free across clouds and regions without lock-in, giving you the flexibility to choose and the performance to stay.
The SUNK Scheduler runs training, inference, and reinforcement learning workloads on the same cluster to maximize efficiency. Topology-aware scheduling and optimized job requeue improve performance and resource utilization across every phase of research.
Quickly troubleshoot and optimize performance using Grafana dashboards purpose-built for SUNK. Access rich visibility into Slurm job metrics, hardware, networking, and storage layers, all tightly integrated with CoreWeave’s observability stack for end-to-end infrastructure insight.
SUNK runs on infrastructure services that provide the ideal combination of ease of use, workload fungibility, performance, and scale.
Get the latest GPU compute you need for your AI workloads through a Kubernetes-native environment
Flexible, purpose-built, high-performance storage solutions tailored for AI
High-performance networking for optimal cluster scale-out and connectivity
With massive megaclusters, CoreWeave GPU clusters help support multi-trillion parameter model training.
