Train

TrainPods

Rent on-demand NVIDIA GPUs - from A100 and L40S to H100, H200 and B200 - billed by the hour in INR from your prepaid Podstack wallet. Instances boot in seconds and you connect over SSH straight from the podstack CLI, with your keys staying local.

Overview

TrainPods is PodStack's on-demand GPU cloud for training and fine-tuning. Launch an instant-boot NVIDIA GPU instance in seconds, pay by the hour in INR from your prepaid Podstack wallet, and connect securely with `podstack gpu instances ssh` - no browser terminal, and your SSH keys never leave your machine. The whole workflow is terminal-native: the podstack CLI copies your training data up, exposes ports for tools like Jupyter or TensorBoard, and tears everything down when you're done. GPUs are available across global regions in the US, Europe, Canada, India and Asia-Pacific.

Loading live data…

ISO 27001 certified
Zero egress fees
Per-minute billing
Instant provisioning

Why teams choose TrainPods

Built on PodStack's proprietary PodVirt layer — not vanilla Kubernetes or OpenStack — so you get fractional GPUs, sub-second scaling, and one bill.

On-demand NVIDIA GPUs: A100 40/80GB, H100 (SXM/PCIe/NVL), H200, B200, L40S, L40, A40, A30, L4 and RTX 40/30-series

Billed by the hour in INR from your prepaid Podstack wallet - no subscription

Instant-boot GPU instances, ready in seconds

Connect securely via the podstack CLI over SSH - keys stay local, no browser terminal

Global regions across the US, Europe, Canada, India and Asia-Pacific

Copy data, expose ports and tear down instances entirely from the CLI

podstack cli
# Install the CLI
curl -fsSL https://podstack.ai/install-cli.sh | sh

# Rent an H100 and SSH in
podstack gpu launch --gpu h100 --region in-blr-1
podstack gpu ssh <instance-id>

# Copy data, expose Jupyter, tear down
podstack gpu cp ./data <instance-id>:/workspace/data
podstack gpu expose <instance-id> 8888
podstack gpu rm <instance-id>

Terminal-native GPU training

TrainPods are built for engineers who live in the terminal. No browser SSH, no noisy neighbors — just bare-metal NVIDIA GPUs you rent by the hour and manage through the podstack CLI.

  • Rent A100, H100, H200, B200, L40S, L4 and RTX-class GPUs
  • SSH directly from your machine — keys never leave your laptop
  • Copy data, expose ports, and tear down entirely from the CLI
  • Billed by the hour in INR, no subscription

Use cases

Train and fine-tune on demand

Spin up an A100, H100 or H200 in seconds, run your job, and stop paying the moment you tear it down.

Terminal-native workflow

Provision, SSH, copy data and expose ports entirely from the podstack CLI - no console or browser terminal required.

Pay only for the hours you use

Hourly, on-demand pricing in INR from your prepaid Podstack wallet - no subscription or long-term commitment.

When to use TrainPods

  • You want to rent a specific NVIDIA GPU - like an H100, A100 or H200 - by the hour, on demand
  • You live in the terminal and want CLI provisioning, SSH access and data transfer
  • You want simple hourly billing in INR from a prepaid wallet, with no subscription

How to use

  1. 1

    Install the podstack CLI

    One binary, from the terminal you already use.

    trainpods
    curl -fsSL https://github.com/Podstack-ai/podstack-cli/releases/latest/download/install.sh | sh
  2. 2

    Authenticate

    Log in with an API key (psk_...).

    trainpods
    podstack auth login
  3. 3

    Launch a GPU

    Boot an instant-boot NVIDIA GPU instance from the Podstack console, billed by the hour in INR.

  4. 4

    Connect over SSH

    Drop into your instance securely - your keys stay local.

    trainpods
    podstack gpu instances ssh <instance-id>

Frequently asked questions

What is TrainPods?

TrainPods is PodStack’s on-demand GPU cloud for training and fine-tuning. You rent NVIDIA GPUs by the hour, billed in INR from your prepaid Podstack wallet, and connect to each instant-boot instance securely through the podstack CLI over SSH.

Which GPUs can I rent and how much do they cost?

TrainPods offers NVIDIA A100 (40 and 80GB), H100 (SXM, PCIe and NVL), H200, B200, L40S, L40, A40, A30, L4 and RTX 4090 and 3090-class GPUs. Pricing is on-demand and billed per hour in INR, starting from about ₹30 per hour; live per-GPU rates are shown on the pricing page.

How am I billed for TrainPods?

You pay by the hour in Indian rupees (INR), drawn from your prepaid Podstack wallet - no subscription or long-term commitment. You are billed only while an instance is running and stop paying the moment you tear it down.

How do I connect to a GPU instance?

Everything runs through the podstack CLI. Install it, authenticate with an API key (psk_…), then run `podstack gpu instances ssh <instance-id>` to open a secure SSH session. There is no browser terminal - your SSH keys stay on your own machine.

How do I move data and manage instances?

Use the podstack CLI: `podstack gpu instances cp` copies training data up, `podstack gpu instances expose` opens a port for tools like Jupyter or TensorBoard, and `podstack gpu instances rm` tears the instance down when you are done.

Which regions are available?

TrainPods runs in global regions across the United States, Europe, Canada, India and Asia-Pacific, so you can launch GPUs close to your data and users.

Ready to try TrainPods?