Create a cluster

Procedure

1. Go to the Slurm clusters page. In the sidebar, it is Slurm Cluster, under Compute.
2. Click Create Slurm cluster.
3. Fill in the sections below, from top to bottom.
4. Read the price in the Configuration card.
5. Click Create. The button shows the price per hour.

The console opens the cluster's page. The cluster status is Creating.

Cluster name

Use 3 to 31 lowercase letters, digits or -. Start with a letter. An example is team-pretrain.

You cannot have two clusters with the same name in one region.

Select resource

The page has no region field. Slurm clusters are available in India-02 only.

Field	What to do
Compute node type	Select a card. Each card shows the GPU type, the GPUs per compute node, InfiniBand, and the number of free compute nodes. The type today is H200, with 8 GPUs per compute node.
Compute nodes	Enter a number. The default is 1. The maximum is 128, or the number of free compute nodes if that is lower. The page shows the maximum next to the box.

The create page shows these messages if too few compute nodes are free:

All compute nodes of this type are in use right now. Pick another type or try again later.

Only 1 free compute node of this type right now. Choose fewer compute nodes.

Disk

Per compute node sets the disk size of each compute node. Use 500 GB to 4000 GB. All compute nodes of a cluster have the same disk size, and new compute nodes use it too.

Jobs can use this disk through /tmp and /var/tmp. Container images also use it. Refer to Storage.

Caution:

You cannot change this disk size after you create the cluster.

Login VM

Field	What to do
Size	Select a CPU and memory size. The default is 4 vCPU and 16 GB RAM. The console marks a size as unavailable if the region cannot provide it.
Disk	Use 100 GB to 2000 GB.

All people share the login VM, and nothing limits one person's use. Real work runs on compute nodes, so the login VM can be small.

Scheduler VM

Select one size: 4 vCPU and 16 GB RAM (the default), 8 vCPU and 32 GB RAM, or 16 vCPU and 64 GB RAM.

This VM runs the Slurm scheduler and job history. Jarvislabs manages it, and you cannot log in to it. Its disk is 100 GB and you cannot change it.

Shared filesystem

Select a filesystem from the list. The cluster mounts it at /home/jl_fs on the login VM and every compute node. Each person's home folder is /home/jl_fs/<login-name>.

If the list shows None in this region, click Create one. If another cluster uses your filesystem, the console refuses the create:

This filesystem is already used by Slurm cluster 'team-a'. Choose another filesystem, or delete that cluster first.

Networking

Private network (VPC) shows your default VPC in the region. Select another VPC if you need to. Refer to Before you start.

People

The first person has the Admin badge. You cannot change that person's role on this page. The console fills in a login name from your email address, and your saved SSH keys. Change them if you need to.

Click Add person to add more people. You can also add people later on the People tab. For each person, set these fields:

Field	Rule
Login name	Up to 32 lowercase letters, digits, - or _. Start with a letter or _. You cannot change it later. Each login name is once per cluster.
Role	Admin (sudo) or User. The first person has no choice: that person is an admin.
SSH public keys	At least one key. Select Add a saved key, or click Paste a key. Keys apply to this cluster only.

Jarvislabs reserves some names, such as root, ubuntu, admin, slurm, docker and users. The console tells you if you pick one.

If someone used this login name before on this filesystem, this person gets that home folder and its files. If the folder belongs to someone else, this person cannot use it. Pick another name.

These limits apply:

Limit	Value
People	50
SSH keys per person	50
Size of all keys of one person	58 KB
Size of all people, SSH keys and Slurm settings	The form shows the size when you get near the limit.

Near the limit, the form shows Settings size: with the used and total KB. Over the limit, the form shows this message and Create stays off. Add fewer people or keys, or shorten the Slurm settings.

The people, SSH keys and Slurm settings are too large: 70 of 63 KB used. Add fewer people or keys, or shorten the Slurm settings.

Note:

50 people fit only with short keys. Long keys, such as RSA 4096, reach the limit sooner. Use ssh-ed25519 keys if you have a large team.

Slurm settings (optional)

Click Slurm settings (optional) to open this section. You can leave it empty and set these values later on the Slurm settings tab. The fields and rules are the same there. Refer to Slurm settings.

Configuration card

The Configuration card on the right shows what you selected, and the price of the compute nodes, the login VM and the scheduler VM. It also shows the total per hour and per day.

The price includes each VM's disk. The shared filesystem has a separate bill. Refer to Billing.

Messages at create

The page shows most messages in a box at the field they are about. Balance and settings size messages, and messages after you click Create, show above Create.

Message	What to do
Your balance doesn’t cover one hour of this cluster. Add balance and try again.	Add balance.
Choose a filesystem.	Select a filesystem.
Number of compute nodes must be between 1 and 128.	Enter a number from 1 to 128.
Choose a login VM size.	Select a size that is available.
All login VM sizes are in use in this region right now. Try again later.	Try again later.
All scheduler VM sizes are in use in this region right now. Try again later.	Try again later.
No VPC in this region. Create one on the VPC page first.	Create a VPC.
Not enough free H200 compute nodes on one InfiniBand fabric in ...	Choose fewer compute nodes, or try later.
You already have a Slurm cluster named 'team-a' in ...	Choose another name.
We couldn’t confirm this. Check your Slurm clusters before trying again.	The connection stopped. Open the Slurm clusters page. The cluster may be there.