Skip to content

TIR Instance FAQ

View as Markdown

For step-by-step fixes, see Troubleshooting TIR Instances.

Should I use TIR or a MyAccount GPU node for AI and machine learning work?

Section titled “Should I use TIR or a MyAccount GPU node for AI and machine learning work?”

Use TIR. It is E2E’s platform for AI work, and a TIR Instance is ready for that work as soon as it starts. You pick a pre-built image for PyTorch, TensorFlow, Transformers, vLLM, and more, and you get JupyterLab, the TensorBoard and Gradio add-ons, and a workspace that keeps your files across restarts. TIR runs GPUs in Delhi and Chennai, including H200 and B200 plans; MyAccount GPU nodes are in Delhi only. For a full virtual machine with its own kernel, TIR also has Virtual Machines under Compute. A MyAccount GPU node suits a GPU that has to sit beside your other MyAccount resources, or a Windows GPU machine. See Create an instance.

What is the difference between a TIR Instance, a TIR Virtual Machine, and a MyAccount GPU node?

Section titled “What is the difference between a TIR Instance, a TIR Virtual Machine, and a MyAccount GPU node?”

A TIR Instance is a container with pre-built machine learning images and JupyterLab. Only /home/jovyan and your mounted storage keep their data across restarts, and it runs without systemd or a Docker daemon. A TIR Virtual Machine, under Compute > Virtual Machines, is a full Linux virtual machine with its own kernel and root access, and its disk keeps its files across restarts and stops. A MyAccount GPU node is a virtual machine in MyAccount, in Delhi only, where you choose the operating system. Its NVIDIA driver ships with the node’s image, and GPU images built on Ubuntu 22.04 also come with Docker and the NVIDIA Container Toolkit; on Ubuntu 24.04 you install those two yourself. TIR Instances run in Delhi and Chennai. See TIR Instance and GPU Node.

Yes. You are root inside the instance, so you can install any package you need. The instance is still a container, so it cannot run the Docker daemon or load kernel modules. See Can I run Docker inside an instance?

Does the instance keep its system disk after a restart?

Section titled “Does the instance keep its system disk after a restart?”

No. Only the workspace at /home/jovyan and your mounted datasets, SFS, and PFS keep their data. Everything else in the instance goes back to the image’s contents when it restarts. See Where your files live.

Which GPUs can I use, and in which regions?

Section titled “Which GPUs can I use, and in which regions?”

In Delhi, H200, H100, A100 40 GB, and L4 plans come with one, two, four, or eight GPUs. A100 80 GB, L40S, and V100 plans come with one, two, or four. In Chennai, B200 and H100 plans come with one, two, four, or eight GPUs, and the H200 plan comes with eight GPUs, on committed terms only. Spot plans are H200 and H100 in both regions. CPU plans are the C3 series in both regions. The list changes over time, and the create form shows what is on sale now. See Create an instance.

Can I run an H200 instance on an hourly plan in Chennai?

Section titled “Can I run an H200 instance on an hourly plan in Chennai?”

No. In Chennai, the H200 comes as one eight-GPU plan, sold on committed terms only. For an hourly GPU instance in Chennai, choose the H100 or the B200. In Delhi, the H200 is offered hourly with one, two, four, or eight GPUs. See Billing options.

Are my instance’s GPUs, vCPUs, and RAM shared with other customers?

Section titled “Are my instance’s GPUs, vCPUs, and RAM shared with other customers?”

No. They are your instance’s alone, and each GPU is a whole physical GPU.

Can I run more than one container on an instance?

Section titled “Can I run more than one container on an instance?”

No. Each instance runs one container from one image. To run another image, create another instance, or update this one’s image. See Update the image of an instance.

No. JupyterLab comes with the pre-built images that include it, and with custom images you launch with JupyterLab Support ticked. Tick JupyterLab Support only for an image built with TIR’s image builder or on a TIR pre-built image. The ComfyUI and FramePack pre-built images open their own app instead of JupyterLab, from the same Launch App button. On Base OS images, public images, and custom images launched with JupyterLab Support left unticked, you work over SSH. See Create an instance from your own image.

Is there a limit on how many instances I can run?

Section titled “Is there a limit on how many instances I can run?”

No. What limits a launch is free capacity for the plan and, on a prepaid account, your credit. See Create an instance.

How long does a new instance take to start?

Section titled “How long does a new instance take to start?”

About two minutes on a pre-built image. The instance shows Waiting while TIR finds a host, the host downloads the image, and the image starts up. Then it shows Running. A large custom image takes longer to download. Billing starts when it reaches Running. See Create an instance.

No. Every instance is launched on a paid plan. See Billing options.

By the minute, at the plan’s hourly rate, from when the instance reaches Running until you stop or delete it. For prices, see E2E Networks pricing, and for how each billing option works, see Billing options.

Billing for the plan stops. The workspace above 30 GB is still billed until you delete the instance, and datasets, SFS, PFS, and reserved IPs are billed as their own products. See Stop, start, and restart an instance.

Am I billed while my instance shows Waiting?

Section titled “Am I billed while my instance shows Waiting?”

Only for its workspace above 30 GB. The plan is billed while the instance is Running, so a new instance is first billed when it reaches Running. On an hourly plan, a running instance can drop back to Waiting, for example after it runs out of memory, and its plan is billed again once it is Running. See The instance stays on Waiting.

How much credit do I need to launch an instance?

Section titled “How much credit do I need to launch an instance?”

On a prepaid account, launching an On-Demand instance needs credit for about 24 hours of its hourly price, and a committed instance needs credit for the full committed price. Starting a stopped hourly instance needs credit for at least one hour of its plan. The check applies to prepaid accounts only. See Insufficient Credits! when you launch or start an instance.

What am I still billed for after I delete an instance?

Section titled “What am I still billed for after I delete an instance?”

Billing for the instance and its workspace stops when you delete it. Saved images in the Container Registry, datasets, SFS, PFS, and reserved IPs are kept and billed as their own products until you delete them. See Delete an instance.

Yes. A reserved IP is billed monthly, as a product of its own, until you delete it. For prices, see E2E Networks pricing, and to reserve, attach, or release one, see Manage IP addresses on an instance.

Do SSH keys, security groups, or start scripts cost anything?

Section titled “Do SSH keys, security groups, or start scripts cost anything?”

No. SSH keys, security groups, start scripts, add-ons, and opening the app are all free. See Billing options.

Is network bandwidth or data transfer billed?

Section titled “Is network bandwidth or data transfer billed?”

No. An instance has a 1 Gbps network connection, and data going in and out of it is free, as long as use stays fair.

The full price of the term is deducted when the instance first reaches Running, and that one payment covers the plan for the whole term. Terms differ by plan, from seven to 1,095 days. See Billing options.

No. A committed instance stays on through its term, so its menu offers a restart in place of Stop Instance and Start Instance. You can also delete it once you are done with it. See Stop, start, and restart an instance.

Can I delete a committed instance before its term ends?

Section titled “Can I delete a committed instance before its term ends?”

Yes. The term was paid upfront, so the days left are not refunded. See Delete an instance.

The renewal preference you chose takes effect. Auto renewal starts a new term of the same plan and deducts its price. Hourly billing keeps the instance running, billed On-Demand from the end of the term. Auto deletion deletes the instance and its workspace. TIR emails reminders two days and one day before the term ends, and seven days before on terms longer than seven days. It emails you again once the action is done. See Billing options.

Why did my committed instance move to hourly billing instead of renewing?

Section titled “Why did my committed instance move to hourly billing instead of renewing?”

Auto renewal renews the same committed plan. On a prepaid account with too little credit for the renewal, the instance moves to hourly billing instead. To commit again, add credit and convert the instance to committed. See Convert an instance to committed.

Can I convert an hourly instance to a committed plan?

Section titled “Can I convert an hourly instance to a committed plan?”

Yes, with Convert to Committed. A running instance keeps running on the same plan, now billed as committed, and the full commitment is charged upfront. On a prepaid account, you need credit for the full commitment. Spot Instances and instances on a private cluster cannot be converted. See Convert an instance to committed.

Can I move a committed instance back to hourly billing?

Section titled “Can I move a committed instance back to hourly billing?”

No. The instance stays committed until its term ends. To move to hourly billing then, select Update Plan on the instance’s page, choose Auto start hourly billing for this machine, at <price>/hr under Select Committed Policy, and select Update. See Change the plan of an instance.

A Spot Instance runs on spare GPU capacity at a lower hourly price, and TIR can stop it at any time to take the capacity back. The Spot rate is fixed for each plan, whatever the demand. Spot Instances are always hourly: billed by the minute at the Spot rate while they run. Spot plans are H200 and H100, in Delhi and Chennai. Choose Spot for work that can stop at any moment and pick up again. See Billing options.

Why is no Spot plan available when I create an instance?

Section titled “Why is no Spot plan available when I create an instance?”

Spot capacity is usually low. You can launch a Spot plan only while spare GPUs are free, and they often run out. When they do, the launch is refused; try again later, or choose a GPU plan. See Create an instance.

Can I use a Spot plan for a TIR Virtual Machine?

Section titled “Can I use a Spot plan for a TIR Virtual Machine?”

No. Spot plans are for TIR Instances only.

What happens when TIR takes back a Spot Instance?

Section titled “What happens when TIR takes back a Spot Instance?”

TIR stops the instance without warning and emails you afterwards. A start on a Spot Instance that TIR stopped is refused; see the message it shows. Its workspace is kept: to use the instance again, change its plan to a GPU or CPU plan, and it starts on that plan with its workspace. See Change the plan of an instance.

Yes, once it is stopped. A Spot Instance you stopped yourself can move to another Spot plan, or to a GPU or CPU plan. A Spot Instance that TIR took back can move only to a GPU or CPU plan. Only a Spot Instance can move to a Spot plan. See Change the plan of an instance.

In the workspace at /home/jovyan, or on a mounted dataset, SFS, or PFS. Files there survive restarts, stops, and starts. Everything else in the instance is lost when it restarts. See Where your files live.

Why do my installed packages disappear after a restart?

Section titled “Why do my installed packages disappear after a restart?”

A package installed with apt, or with a system-wide pip install, is written outside /home/jovyan, and a restart puts everything there back to the image’s contents. pip install --user and a virtual environment under /home/jovyan are kept, because they are inside the workspace. To keep an apt package or a system-wide pip install, install it from a start script, which runs each time the instance starts. You can also save the instance as an image, or use a custom image that already has the packages. See Run start scripts on an instance and Save an instance as an image.

It starts at 30 GB. You can set it up to 15,000 GB when you create the instance, or grow it later on Workspace Size. The limit for your account in a region is 15,000 GB unless E2E has set another. When you reach it, select Request to ask E2E to review a larger one. The FramePack image needs a workspace of at least 50 GB. See Increase the workspace size.

No. The workspace can only grow. To use less storage, create a new instance with a smaller workspace and move your files to it. See Increase the workspace size.

/mnt/local is fast local NVMe storage. It comes with plans that include it, shown as Local NVMe Storage in the plan’s details, and with a private cluster’s local volume. Local storage that comes with the plan survives a restart, and stopping the instance clears it. A private cluster’s local volume is kept when you stop the instance. See Where your files live.

Yes, when you create the instance. Workspace Encryption under Storage on the create form turns on AES-256 encryption of the workspace. You can set it only on the create form. See Create an instance.

If I delete an instance, are its datasets, SFS, and PFS deleted too?

Section titled “If I delete an instance, are its datasets, SFS, and PFS deleted too?”

No. Deleting an instance deletes the instance and its workspace. Its saved images, datasets, SFS, PFS, and reserved IPs stay, ready to use with another instance. See Delete an instance.

How do I move files in and out of an instance?

Section titled “How do I move files in and out of an instance?”

Copy them into /home/jovyan so they survive restarts. See Copy files to and from an instance for the ways to do it, and Mount datasets and file systems to bring in data you keep in a dataset or file system.

Can several instances share the same data?

Section titled “Can several instances share the same data?”

Yes. Mount the same SFS on each instance to share files read-write, or the same EOS dataset to share them read-only. A disk dataset can be mounted on one instance at a time. See Mount datasets and file systems.

root, with the command ssh root@<ip-address>, where <ip-address> is the address in SSH Command on the instance’s page. See Connect to an instance over SSH.

Can I log in to an instance with a password?

Section titled “Can I log in to an instance with a password?”

No. You sign in with an SSH key, which is the only way into an instance. See Connect to an instance over SSH.

TIR keeps its own SSH keys for each user, under SSH Keys in the account menu, separate from MyAccount’s. Add the key again in TIR. See My SSH key is missing or is not accepted.

Can a teammate log in to my instance over SSH?

Section titled “Can a teammate log in to my instance over SSH?”

Yes, with their own SSH key. Each person attaches keys from their own SSH Keys list, so your teammate attaches their own key to the instance. To do that, they need permission to update instances. See Connect to an instance over SSH.

How do I give a teammate access to one instance only?

Section titled “How do I give a teammate access to one instance only?”

Access is set per project, so put that instance in a project of its own and add the teammate to that project only. An OWNER, ADMIN, TEAM LEAD, PROJECT MANAGER, or PROJECT LEAD can work with every instance in the project. A MEMBER can do what their policies allow, on every instance in it. For the message a teammate sees when they lack access, see Troubleshooting TIR Instances.

Add the port to an inbound rule (Custom TCP) of a security group attached to the instance. The portal takes TCP ports 22, 80, 443, and 1024 to 65535, one at a time. Make the app listen on 0.0.0.0, which means all addresses, instead of localhost. The port is public: anyone who can reach the IP and port reaches your app. See Open a port on an instance.

Can I open a UDP port or a range of ports?

Section titled “Can I open a UDP port or a range of ports?”

No. Security group rules are TCP only and take one port at a time: 22, 80, 443, or any port from 1024 to 65535. Add each port you need as its own entry. See Open a port on an instance.

Does the instance’s public IP change after a stop?

Section titled “Does the instance’s public IP change after a stop?”

Yes, it can. An instance gives back its public IP when it stops, and on start it gets an address from E2E’s pool, which can be a different one. To keep the same address, use a reserved IP. See Manage IP addresses on an instance.

Five in each region, counted across all your account’s projects, unless E2E has raised the limit for your account. When you reach it, a further reservation is refused with You have reached the maximum limit for reserving IPs., and the reserve dialog offers Request, where you ask E2E for more, up to 10. The same limit applies when you keep an instance’s IP as you delete it. See Manage IP addresses on an instance.

Which address do outside servers see when my instance connects out?

Section titled “Which address do outside servers see when my instance connects out?”

The Egress IP, shown on the instance’s Network & Security tab under Network. It is the address for traffic going out from the instance to the internet, and the one to add to another service’s allow list. It is released when you stop the instance, and a start can give you a different one. See Manage IP addresses on an instance.

No. The VPC must be an Active MyAccount VPC in the instance’s region, with an address range that E2E assigned. A VPC with a custom CIDR is refused with VPC with custom CIDR not allowed. Reserving a VPC IP takes two addresses from the VPC. See Manage IP addresses on an instance.

Can my instance reach my office network privately?

Section titled “Can my instance reach my office network privately?”

Yes. Attach a VPC IP from a MyAccount VPC to the instance, then connect that VPC to your office with a site-to-site VPN on a Firewall.

Can I upgrade the NVIDIA driver on my instance?

Section titled “Can I upgrade the NVIDIA driver on my instance?”

No. The NVIDIA driver is part of the GPU host, outside the image, so your instance always uses the version the host runs. See My code needs a newer NVIDIA driver or CUDA version.

Can I use a newer CUDA version than my image has?

Section titled “Can I use a newer CUDA version than my image has?”

Yes, up to the version the host’s driver supports. The CUDA toolkit comes with the image, and you can install another version inside the instance. Run nvidia-smi and read CUDA Version: that is the highest CUDA that works on the instance. See My code needs a newer NVIDIA driver or CUDA version.

No. An instance is itself a container. To run your own Docker image, build it and push it to your Container Registry from another machine, then launch an instance from it as a custom image. To keep packages you added inside an instance, use Save Image. See Create an instance from your own image.

Why does systemctl not work in my instance?

Section titled “Why does systemctl not work in my instance?”

An instance runs without systemd, so systemctl and services that need systemd do not work. Start the program in the background from a start script instead. See Run start scripts on an instance.

No. A saved image holds the container outside /home/jovyan: apt packages, system-wide pip packages, and other files there. Your workspace files, and packages installed with pip install --user, stay out of it. Restoring the image on the same instance keeps its workspace; a new instance from the image starts with an empty one. See Save an instance as an image.

Saving a small image takes about 12 seconds, and a larger image takes longer. The instance shows Saving Image until the save finishes, then Running again on its own, and TIR emails you when it succeeds or fails. See Save an instance as an image.

Yes, with Update Image, and the instance can stay running while you do it. Changing the image is free, and the plan, its price, the workspace size, and other settings stay the same. A running instance restarts on the new image: files in /home/jovyan stay, and you install your own packages again. See Update the image of an instance.

Can I change the GPU or CPU of an instance?

Section titled “Can I change the GPU or CPU of an instance?”

Yes, for an hourly instance, with Update Plan. Stop the instance first. You can pick any GPU or CPU plan that supports the instance’s image version, larger or smaller. Only a Spot Instance can move to a Spot plan. The workspace, its files, and the image stay the same. See Change the plan of an instance.

Can I change the plan of a committed instance?

Section titled “Can I change the plan of a committed instance?”

No. The plan is fixed for the term; you can change only its renewal preference, with Update Committed Plan in the list’s menu or Update Plan on the instance’s page, in any status. See Change the plan of an instance.

Pending means the instance is waiting for capacity. It is rare: TIR refuses a create or a start when all capacity is in use, so an instance shows Pending only in cases such as capacity running out while its plan changes. A Spot Instance waiting the same way shows Queuing. While an instance is Pending, its menu offers Revoke Request, which stops it if it has run before and deletes it otherwise. An instance that stays Pending or Queuing for three days is stopped and marked Expired, and TIR emails you. See Delete an instance.

What is the difference between Restart and Stop?

Section titled “What is the difference between Restart and Stop?”

Restart Instance starts the instance’s container again: the workspace and mounted storage stay, and you install your own packages again. Stop releases the compute and stops compute billing until you start the instance again. See Stop, start, and restart an instance.

Yes. You can delete an instance in any status except Saving Image. Deleting the instance and its workspace is permanent, so copy what you need first. See Delete an instance.

Does TIR stop an instance that stays on Waiting?

Section titled “Does TIR stop an instance that stays on Waiting?”

No. An instance can stay on Waiting indefinitely; TIR leaves it where it is. You can restart it, or stop it if it is hourly, and contact support with the instance ID if it stays on Waiting. See The instance stays on Waiting.

Can I stop an instance automatically at a set time?

Section titled “Can I stop an instance automatically at a set time?”

No. Stop the instance yourself, or call the stop action of the TIR API from your own scheduler. See Stop, start, and restart an instance.

Check three causes: the instance ran out of memory, it filled its ephemeral storage, or someone in the project changed its mounts, plan, or image, which restarts a running instance. Ephemeral storage is the space outside /home/jovyan and mounted storage, and its contents are lost on restart. See The instance restarts on its own.

What happens to my instances if the project is deleted?

Section titled “What happens to my instances if the project is deleted?”

They are deleted with it, in both regions, with their workspaces, and their billing stops. The deletion is permanent. The project’s reserved IPs are released. Copy what you need to keep before you delete a project.

What happens to my instances if my account is deprovisioned?

Section titled “What happens to my instances if my account is deprovisioned?”

All the account’s projects are deleted, with every instance and workspace in them.

What happens to my instances if my account is suspended?

Section titled “What happens to my instances if my account is suspended?”

TIR leaves them as they are, but refuses every request, including listing your instances, with Customer is suspended!. The instances become reachable at their public or reserved IPs again once the suspension is lifted.