// GPU CLOUD
Compute
Virtual machines and block storage for any AI and ML workloads.
Flexible capacity planning
Get access to the latest NVIDIA GPU platforms or CPU-only servers and balance reserved and on-demand pricing models aligned with your needs.
AI performance without penalty
Receive bare-metal-level performance from dedicated hosts — we do not virtualize and share GPU and network cards.
InfiniBand-powered AI clusters
Create multi-host clusters with non-blocking NVIDIA Quantum InfiniBand fabric. 3.2 Tbit/s throughput per 8-GPU host with direct GPU-to-GPU communication.
AI-ready operating system
Save time with an AI/ML-ready image containing pre-installed GPU and network drivers to start a GPU-accelerated environment quickly.
Network storage volumes
Reduce cluster recovery time by leveraging network disks mounted to every virtual instance. Cloud-native elasticity with quick VM restart on failure.
Integrated monitoring
Detailed cluster and VM performance metrics — from GPU utilization to InfiniBand network parameters — on web UI or Grafana dashboards.
GPU host configurations
NVIDIA B300
Available- 3.2 Tbit/s InfiniBand
- 1792 GB DDR5
- 128x vCPU
- 8x B300 GPU
NVIDIA B200
Available- 3.2 Tbit/s InfiniBand
- 224 or 1792 GB DDR5
- 20x or 160x vCPU Intel Emerald Rapids
- 1x or 8x B200 GPU 180GB SXM
NVIDIA H200
Available- 3.2 Tbit/s InfiniBand
- 200 or 1600 GB DDR5
- 16x or 128x vCPU Intel Sapphire Rapids
- 1x or 8x H200 GPU 141GB SXM
NVIDIA H100
Available- 3.2 Tbit/s InfiniBand
- 200 or 1600 GB DDR5
- 16x or 128x vCPU Intel Sapphire Rapids
- 1x or 8x H100 GPU 80GB SXM
NVIDIA A100 80GB
Available- 96 or 1152 GB DDR5
- 12x or 96x vCPU
- 8x A100 GPU 80GB SXM
NVIDIA L40S
Available- 32 or 160 GB DDR5
- 8x or 40x vCPU Intel Xeon Gold
- 1x L40S GPU 48GB PCIe
Try self-service console
Up to 32 NVIDIA GPUs are available immediately via web console.
Block network storage
Choose from three options of network disks that differ by performance, reliability and pricing:
- SSDs with data mirroring — highest reliability
- SSDs with erasure coding — balanced performance and reliability
- SSDs with no data replication — maximum performance, lowest cost
Observability and monitoring
Control the cluster state and detect performance issues early using our integrated monitoring capabilities. We display a wide range of performance metrics, from GPU utilization to InfiniBand network parameters, on web UI dashboards or as pre-assembled Grafana dashboards.
Read the article →Relevant products
Managed Service for Kubernetes
A fully managed container orchestrator that is optimized for modern AI workloads.
Learn more →Slurm Operator
A Slurm-based workload manager for ML and HPC clusters with a modern, simplified user experience.
Learn more →