---
title: "Instances- Launch "
---

# Create Instance

![Create Instance Demo](../Nodes/images/Screen_Recording.gif)

## What are Instances (Nodes)
Instances are ready-to-use, collaborative workspaces for AI/ML development.  
Each Instances combines:
- A container
- JupyterLab (if included in selected image)
- Common ML libraries  

This allows your team to run Instances, Scripts, Datasets, and Model experiments **without manual environment setup**.

---

## Quickstart 

1. Log in to E2E networks.
2. Create a **Project** (projects live in your private workspace).  
3. Inside the project, go to the section **Instances (Nodes) → Create Instance**.  
4. Pick an **image** (TIR Pre-built, Base OS, Custom Images, or Public Images).  
For **Public Images**, go to the **Custom Images** tab, select **Image Type** as **Public**, and enter the **image URL**.
5. Next, choose a **CPU or GPU plan**, provide an **instance name**, set the required **configuration and storage**, select an **instance pricing plan**, and optionally enable **SSH** by choosing an SSH key.
6. Click **Create** and open the Instance via **JupyterLab** (or **SSH** if enabled).  


---

## Instance Images 

Instances run on container images that serve as blueprints, bundling the necessary libraries, tools, and dependencies for each instances.

- **Pre-built** — Ready to use images with ML libraries pre-installed  
- **Base OS** — Minimal OS, install everything manually
- **Custom Images** — Use your own custom-built image (via E2E Container Registry)
- **Public Images** - Ready-to-use images maintained by the community or vendors, available from public registries like Docker Hub


💡 *Start with a prebuilt image unless you require a custom environment.*

---

## Key Options You Should Know

- **Enable SSH** — add an SSH public key to access the Instance shell  

:::note
  To enable SSH, you must attach at least one Security Group with port 22 enabled.
:::

- **Workspace path** — the Instance’s workspace mounts at `/home/jovyan`.  
  Use this path for persistent files. Files outside this path may be lost on restart because Instances are container-native.

- **Disk size** — default **30 GB** free storage; expandable up to **5 TB**  
  _(raise a support ticket if you need >5 TB)._

<!-- - **Local NVMe** — available only on H100 plans at `/mnt/local`.  
  High-speed but temporary — move critical data to /home/jovyan or the EOS bucket before shutdown -->

- **Plan & billing** — choose **hourly** or **committed**.  
  Committed plans often give discounts and may unlock local NVMe for certain GPU 

---

## Instance Lifecycle & Status

- **Waiting** — deploying on hardware.  
- **Running** — active; access via JupyterLab or SSH.  
- **Stopped** — not assigned to a machine; workspace remains and may incur storage charges.  
- **Pending** — inventory unavailable; node stays pending for 48–72 hours.  
- **Expired** — request not fulfilled after 72 hours.  

---


### Spot Instance

A Spot Instance is a virtual instance that uses spare cloud compute capacity at significantly reduced costs compared to On-Demand pricing. Spot pricing fluctuates based on supply and demand for available capacity in each Availability Zone, making it a cost-effective option for flexible workloads.

![Select spot plans](../Nodes/spot_instance_images/spotinstancenewui.png)




## Important Warnings (must-read)

⚠️ **Persistence** — use `/home/jovyan` for saving files. Anything outside this path may be lost on container restarts  
⚠️ **Local NVMe is ephemeral** — move checkpoints/data off `/mnt/local` before shutdown  
⚠️ **SSH changes require a stop** — If you forgot to add a key during instance creation, simply enable SSH Access and upload the key.   To enable SSH, you must attach at least one Security Group.    
⚠️ **Deleting a node deletes its disk/workspace** — back up anything you want to keep  
⚠️ **Pending → Expired** — if inventory isn’t free, the request waits **48–72 hours** (Pending) and expires after **72 hours**  

---

## Quick Best Practices

- Start with a **prebuilt Jupyter image** for faster onboarding.  
- Keep **project datasets mounted and versioned** (move large data to EOS buckets).  
- Use **committed plans** for regular heavy workloads to optimised cost and access performance options.  
- Keep an **SSH key backed up** — it’s the recommended way to access the instance via shell.  


---