Your GPUs can now communicate seamlessly across machines, eliminating the old per-node limits and unlocking true multi-node clusters. Spin up as many GPUs as you need, pay only by the second, and stop anytime - no commitments, just on-demand scalability like regular GPU VMs.

​
How to request for a cluster?

1

Head over to the GPU Clusters tab from the left menu

This will open up the UI required for configuring your cluster
2

Select your GPU Model

Select from a pool of B200, H200, H100, A100, L40S, to start with the clusters
3

Select the type of Clusters

When creating a GPU Cluster, you can choose between Slurm Cluster or Kubernetes Cluster, depending on your workload and orchestration needs.
4

Enter the quantity of Nodes

Each node consists of 8 GPUs. For example, requesting 8 nodes will give you access to 64 GPUs.
5

Select the location for your Data Center

We have Data Centers in North America, Europe & Asia. These are SOC Type II & HIPAA complaint.
6

Select the Start Date

This will help us understand how soon do you want to start.
7

Provide Personal & Cluster Details

Enter your Organization name & the duration for which you would like the Clusters for (In Month)
8

Enter the remaining details

Help us know the primary reason for the cluster, primary usecase & any other details you would want us to know.
9

Click on the Submit button

Once done, our platform team will review your requirements and contact you.

​
Type of Cluster

When creating a GPU Cluster, you can choose between Slurm Cluster or Kubernetes Cluster, depending on your workload and orchestration needs.

​
Slurm Cluster

  • Best for: HPC (High-Performance Computing), AI/ML training, research workloads.
  • Why: Slurm is an industry-standard open-source job scheduler widely used in supercomputing.
  • Features:
    • Efficient scheduling for batch jobs and distributed training.
    • Optimized for large-scale GPU utilization.
    • Simple job submission (sbatch, srun).

​
Kubernetes Cluster

  • Best for: Enterprise AI services, production deployments, containerized workloads.
  • Why: Kubernetes is the standard for container orchestration and microservices.
  • Features:
    • Run distributed AI inference services at scale.
    • Manage workloads with pods, deployments, and autoscaling.
    • Seamless integration with CI/CD pipelines and cloud-native tools.
Choose Slurm Cluster: If you are doing training, HPC jobs, or large-scale distributed ML experiments. Choose Kubernetes Cluster: If you are running production inference, containerized services, or multi-team workloads.

​
Key Features

  • Multi-Node Scaling: Run workloads that span multiple nodes, each with 8 GPUs.
  • No Commitments: Spin up a cluster whenever you need it and stop anytime.
  • Flexible GPU Options: Choose from B200, H200, H100, A100, and L40S GPUs.
  • Pay-As-You-Go: Billed per second, just like individual GPU instances.
  • Cluster Management: Launch using a scheduler such as Slurm for distributed job orchestration.

​
Example: Cluster Setup

  • GPU Model: B200 (180 GB VRAM)
  • Cluster Type: Slurm
  • Node Quantity: 2 (16 GPUs total)
  • Data Center: North America
  • Start Date: 27-09-2025
  • Total VRAM: 2256 GB

​
Use Cases

  • Large-Scale LLM Training: Train multi-billion parameter models across 100+ GPUs.
  • Scientific & HPC Workloads: Molecular dynamics, physics simulations, and genomics.
  • Enterprise AI: Multi-tenant, distributed inference services.
  • Research Collaboration: Share clusters among teams for parallel experimentation.

​
Billing

  • Per-Second Billing: Pay only for the runtime of your cluster.
  • No Idle Compute Costs: Stop clusters when not in use.
  • Storage Billing: Root disk and attached volumes are billed separately.
Use Auto Stop to avoid idle compute charges. Store large datasets in external object storage to reduce root disk usage. Use Slurm job scheduling to efficiently distribute workloads across nodes. Start small (1–2 nodes) and scale up as needed.