Launch a GPU cluster from your dashboard or reserve 128, 256, 1,024 GPUs and beyond for training, fine-tuning and inference.
NVIDIA H200 available today, with more Instant Cluster GPU options being added.
Reserve hundreds of GPUs. Or thousands.
Need a different size? Tell us your target. We’ll confirm GPU availability and delivery dates for your deployment.
Two ways to get started
Choose self-service access to available GPUs or plan capacity around your project.
Self-service
Create a cluster directly in the dashboard using available GPU capacity.
Check the dashboard for the current GPU lineup and availability.
Planned capacity
Work with our team to arrange GPU capacity for your workload and schedule.
Bring your workload, target GPU count and timeline.
Reserved & bulk GPU options
Explore GPU and network configurations for reserved deployments and bulk capacity.
InfiniBand
Contact sales
InfiniBand
Contact sales
InfiniBand
Contact sales
InfiniBand
Contact sales
200–400 Gb/s · No InfiniBand
Contact sales
NVIDIA RTX Pro 6000 uses 200–400 Gb/s networking without InfiniBand. We’ll confirm the exact configuration in your quote.
Built for work across nodes
Our NVIDIA H100, NVIDIA H200, NVIDIA B200 and NVIDIA B300 clusters pair NVIDIA reference architecture with InfiniBand networking for work across nodes.
Choose the right network
For distributed workloads that need frequent communication between nodes, available with NVIDIA H100, NVIDIA H200, NVIDIA B200 and NVIDIA B300.
Bulk NVIDIA RTX Pro 6000 capacity uses 200–400 Gb/s networking without InfiniBand.
Before you start
Choosing hardware, checking capacity and getting started.
Need a single VM?Choose Instant Clusters to create a cluster yourself using the GPU types and capacity available in the dashboard. Choose Reserved Clusters when you need a different GPU, a larger deployment or capacity for a planned project. Our team will confirm the configuration, dates and pricing with you.
Yes. We support reserved deployments of 128, 256, 1,024 GPUs and beyond. Share your target GPU count and timeline so we can confirm availability. These sizes describe reserved deployments; Instant Cluster capacity is shown separately in the dashboard.
NVIDIA H200 is available today, with eight GPUs per node and InfiniBand networking. We’re adding more GPU options to Instant Clusters; check the dashboard for the current lineup and available capacity.
NVIDIA H100, NVIDIA H200, NVIDIA B200 and NVIDIA B300 clusters include InfiniBand. NVIDIA RTX Pro 6000 clusters use 200–400 Gb/s networking without InfiniBand. We’ll help you assess the network requirements for your workload before you reserve.
Contact our team with your preferred GPU, GPU count, start date and expected duration. We’ll confirm the available configuration, delivery schedule and commercial terms for your deployment.
A VM is a good starting point when your workload fits on one machine with up to eight GPUs. Choose a cluster when your workload needs multiple nodes, or when you need GPU capacity across several machines.
Choose your starting point
Create a cluster in the dashboard, or bring us the plan for your next large deployment.
NVIDIA, the NVIDIA Logo, and other NVIDIA marks are trademarks or registered trademarks of NVIDIA Corporation in the United States and other countries.