Nebius Compute

AI model by Nebius

High-performance GPU infrastructure for AI training and inference at scale.

Nebius Compute provides high-performance cloud infrastructure specifically engineered for large-scale AI training and inference. It offers virtual machines powered by the latest NVIDIA GPUs, including H100, H200, and the Blackwell architecture (B200, B300), connected via high-speed InfiniBand networking. The platform supports various orchestration methods, from managed Kubernetes to Slurm-based Soperator clusters, catering to both modern containerized apps and traditional HPC workloads. Users can choose between on-demand instances for flexibility or reserved capacity for long-term stability. W

Languages
English
GPU Models
H100, H200, B200, B300, RTX 6000, L40S
Networking
InfiniBand support
CPU Platforms
AMD EPYC Genoa, Intel Ice Lake

Overview

Nebius Compute provides high-performance cloud infrastructure specifically engineered for large-scale AI training and inference. It offers virtual machines powered by the latest NVIDIA GPUs, including H100, H200, and the Blackwell architecture (B200, B300), connected via high-speed InfiniBand networking. The platform supports various orchestration methods, from managed Kubernetes to Slurm-based Soperator clusters, catering to both modern containerized apps and traditional HPC workloads. Users can choose between on-demand instances for flexibility or reserved capacity for long-term stability. With features like non-virtualized GPUs and market-driven spot pricing for preemptible VMs, it delivers raw power and cost efficiency for AI builders.

Compute Infrastructure

  • NVIDIA GPU Support: Provides high-performance virtual machines equipped with NVIDIA H100, H200, B200, and B300 GPUs.
  • InfiniBand Networking: Uses InfiniBand networks for high-speed, low-latency distributed computing between nodes.
  • Non-virtualized GPUs: Offers non-virtualized GPUs to provide raw hardware power for demanding AI workloads.
  • Preemptible Instances: Includes preemptible VM options for cost-effective processing of interruptible workloads.

Orchestration and Storage

  • Soperator Clusters: Supports Slurm workload manager clusters for large-scale ML/AI experiments.
  • Managed Kubernetes: Provides managed Kubernetes clusters optimized for containerized AI applications.
  • AI-Optimized Storage: Features high-speed storage options including WEKA and shared filesystems for data-heavy tasks.

Pricing

NVIDIA HGX H100 On-demand
$4.50 per GPU-hour
NVIDIA HGX H200 On-demand
$5.40 per GPU-hour
NVIDIA HGX B200 On-demand
$8.50 per GPU-hour
NVIDIA HGX B300 On-demand
$9.50 per GPU-hour
NVIDIA RTX PRO 6000 On-demand
$1.80 per GPU-hour
Preemptible GPU Instances
From $0.74 per GPU-hour

Specifications

GPU Models
H100, H200, B200, B300, RTX 6000, L40S
Networking
InfiniBand support
CPU Platforms
AMD EPYC Genoa, Intel Ice Lake

Added by AntSpace from public sources (nebius.com).

Related

More AI models by Nebius