Table of contents

Creating Pods

This guide walks through creating a new pod on Podstack with all available configuration options.

Quick Create

  1. Navigate to Compute > Pods
  2. Click Create Pod
  3. Fill in the required fields (name, image, resources)
  4. Click Create

Configuration Options

Basic Information

Pod Name A descriptive name for your pod. Must be unique within the project.

Project Select the project this pod belongs to. Resources are billed to the project owner.

Container Image The Docker image to run. Examples:

  • pytorch/pytorch:2.0.0-cuda11.7-cudnn8-runtime
  • tensorflow/tensorflow:latest-gpu
  • nvidia/cuda:12.0-base
  • jupyter/scipy-notebook

Private Registry (optional) For private images, provide:

  • Registry URL
  • Username
  • Password/Token

Resource Allocation

GPU Configuration

  • GPU Type: Select from available types (A100, H100, H200, V100, L40S, T4, etc.)
  • GPU Count: Number of GPUs (0 for CPU-only)
  • GPU Cores: Fractional GPU allocation (0.1 - 4.0 cores per GPU)
  • GPU Memory: Memory allocation per GPU

CPU and Memory

  • CPU Cores: 0.1-16 cores (fractional values supported, e.g., 0.5)
  • Memory: RAM in GB

Storage

  • Disk Size: Ephemeral storage for the container

Replicas

  • Replica Count: Number of identical pod instances to run
  • Each replica runs independently and is billed separately
  • Useful for load balancing or parallel processing

Billing Period

  • Select billing frequency (hourly, daily, monthly)
  • Monthly billing may offer better rates for long-running workloads

Networking

Exposed Ports Configure ports to expose for external access:

  • Container Port: Port inside the container
  • Protocol: TCP or UDP
  • Public: Whether to expose externally

Common ports:

ServicePort
SSH22
Jupyter8888
HTTP80/8080
TensorBoard6006

Access Configuration

SSH Key Select an SSH key for terminal access. Keys can be managed in SSH Keys.

Notebook Access (optional) Enable Jupyter notebook with:

  • Username
  • Password

Environment Variables

Add environment variables as key-value pairs:

CUDA_VISIBLE_DEVICES=0,1
MODEL_PATH=/data/models

Init Command (optional)

Command to run when the container starts:

pip install -r requirements.txt && python train.py

Volume Mounts

Mount persistent storage:

  • NFS Volumes: Select from your existing volumes
  • Mount Path: Where to mount inside the container (e.g., /data)

Cost Estimation

Before creating, review the estimated cost showing:

  • Hourly rate breakdown (GPU, CPU, memory, storage)
  • Total hourly cost

Ensure your wallet has sufficient balance.

Creating from Template

To use a saved configuration:

  1. Go to Templates
  2. Find your template
  3. Click Deploy
  4. Modify any settings if needed
  5. Click Create

After Creation

Once created, the pod will:

  1. Pull the container image
  2. Allocate resources
  3. Start the container
  4. Transition to Running status

This typically takes 30 seconds to a few minutes depending on image size.

GPU Availability and Waitlist

If your requested GPU type is not immediately available:

Check Availability

The pod creation form shows real-time GPU availability. Unavailable GPU types are marked accordingly.

Join Waitlist

When GPUs are unavailable:

  1. Select the desired GPU configuration
  2. Click Join Waitlist
  3. Enter your requirements (quantity, duration)
  4. Submit your request

Waitlist Notifications

When GPUs become available:

  • You’ll receive an email notification
  • Act quickly as inventory is first-come-first-served
  • Your waitlist position is preserved for a limited time

Managing Waitlist

View and manage your waitlist entries:

  • See your position in the queue
  • Cancel entries you no longer need
  • Update requirements if needed

Troubleshooting

Pod stuck in Pending

  • Check if requested GPU type is available
  • Verify wallet has sufficient balance
  • Review resource limits
  • Check for any account restrictions

Insufficient Balance

  • Top up your wallet before creating
  • The estimated cost is shown before creation
  • Consider reducing resources if budget is limited

Image pull failed

  • Verify image name and tag
  • Check private registry credentials
  • Ensure image exists and is accessible

GPU Unavailable

  • Join the waitlist for notification
  • Consider alternative GPU types
  • Check GPU Marketplace for baremetal options

Next Steps

Once your pod is running, learn how to connect to it.