Skip to main content

​
How to Reserve a Bare Metal Server?

1

Go to the Baremetal Page from the left menu

You will be able to check all the available Bare Metal server configurations along with pricing & details such as CPU, Memory, Boot Stoage, Main Storage & Network
2

Select your preferred Baremetal Server

This will open a dialog box which shows you a form that needs to be filled in to send in a request for the server
3

Select your Commitment Period

Normally Bare Metal Servers are available for a tenure starting at 1 Year. If you want a custom tenure, you can mention that in the Notes part
4

Enter the details required

This is the place to enter your email (same as the email which was used for logging in), name & phone number
5

Enter the Organization name

This step will help you unlock Organizational disounts. This is optional
If you want the server for a custom timeframe, or anything that will help us give you the best price, Notes is where you put those thoughts

​
Key Benefits

​
Full Hardware Access

  • No virtualization layer: direct access to CPU, GPU, and memory resources.
  • Better performance and lower latency compared to VMs.

​
Dedicated Resources

  • 100% of the server’s GPUs, CPUs, RAM, and disks are reserved for your workloads.
  • Eliminates “noisy neighbor” problems from shared cloud environments.

​
Enterprise-Grade Performance

  • Optimized for training large language models (LLMs), high-throughput inference, and large-scale distributed training.
  • Supports multi-GPU and NVLink/NVSwitch interconnects for advanced workloads.

​
Security & Compliance

  • Single-tenant isolation ensures data privacy and compliance for enterprise, healthcare, and finance workloads.
  • Option for on-premises or hybrid deployment.

​
Flexibility & Control

  • Choose from multiple configurations (GPU types, memory, storage).
  • Install custom drivers, frameworks, or OS images as needed.

​
When to Choose Baremetal

Baremetal servers are recommended when you need:
  • Large-Scale Training: Multi-billion parameter models requiring multiple GPUs.
  • High-Performance Inference: Large-scale production serving with minimal latency.
  • Long-Running Jobs: Persistent workloads that run for days/weeks without interruption.
  • Custom Environments: Full control over OS, drivers, and frameworks.
  • Regulatory Compliance: Environments where data must remain fully isolated.
For smaller, short-lived, or cost-sensitive workloads, consider GPU Instances instead.

​
Available Baremetal Servers

Below are the current enterprise-grade baremetal GPU servers available for reservation.
Each server provides dedicated, single-tenant access with 8 × NVIDIA GPUs, high-core CPUs, ECC memory, NVMe storage, and 10GbE networking.

​
NVIDIA H200 141GB 8-GPU SXM Server

  • Price: $210,240 / year
  • GPU Configuration: 8 × NVIDIA H200 HGX 141GB SXM GPU
  • CPU: 2 × 48 cores, 350W, 2.3GHz
  • Memory: 32 × 64GB ECC Registered DDR5-5600
  • Boot Storage: 2 × 960GB NVMe M.2 SSD
  • Main Storage: 8 × 3.8TB NVMe 2.5” SSD
  • Network: 1 × 2-port 10GbE NIC

​
NVIDIA A100 80GB 8-GPU SXM Server

  • Price: $89,500 / year
  • GPU Configuration: 8 × NVIDIA A100 80GB HGX SXM GPU
  • CPU: 2 × 32 cores, 2.0GHz
  • Memory: 16 × 64GB ECC Registered DDR4-3200
  • Boot Storage: 2 × 960GB NVMe M.2 SSD
  • Main Storage: 2 × 1.9TB NVMe 2.5” SSD
  • Network: 1 × 2-port 10GbE NIC

​
NVIDIA H100 80GB 8-GPU SXM Server

  • Price: $133,414.80 / year
  • GPU Configuration: 8 × NVIDIA H100 80GB HGX SXM GPU
  • CPU: 2 × 52 cores, 2.0GHz
  • Memory: 16 × 64GB ECC Registered DDR5-5600
  • Boot Storage: 2 × 480GB NVMe M.2 SSD
  • Main Storage: 6 × 3.84TB NVMe 2.5” SSD
  • Network: 1 × 2-port 10GbE NIC

​
NVIDIA B200 180GB 8-GPU SXM Server

  • Price: $297,840 / year
  • GPU Configuration: 8 × NVIDIA B200 180GB HGX SXM GPU
  • CPU: 2 × 48 cores, 2.3GHz
  • Memory: 32 × 96GB ECC Registered DDR4-5600
  • Boot Storage: 2 × 960GB NVMe M.2 SSD
  • Main Storage: 4 × 7.68TB NVMe 2.5” SSD
  • Network: 1 × 2-port 10GbE NIC
Each server comes with:
  • High-Core CPUs (Intel/AMD EPYC)
  • Large RAM options (up to 2 TB depending on config)
  • NVMe SSDs / High-speed storage
  • Optional networking upgrades (100 Gbps+ InfiniBand / Ethernet for distributed workloads)

​
Billing

  • Hourly or Monthly Plans available depending on your use case.
  • Storage billed separately (root disk + attached volumes).
  • Pause/Stop not available: Baremetal servers are always billed while provisioned.
Use baremetal for long-running or mission-critical jobs. Keep root disk minimal and store large datasets in external storage. Use job schedulers (Slurm, Kubernetes) for multi-GPU workloads. Enable monitoring via nvidia-smi or platform dashboards. Regularly rotate SSH/API credentials for security.