Start a Server

When you spawn a new JupyterHub server, you configure three components: the software image (your working environment), storage (where your files live), and hardware resources (CPU, RAM, GPU). This page walks through each choice.

Choose Your ImageCopy Anchor Link

Select a pre-configured environment tailored to your scientific domain:

Image cards in the spawn form carry badges such as AI, SSH, VSC, GPU, or CPU, and each card shows the image release date — pick the most recent tag for a category.

Using a Custom Docker Image: You can supply your own public Docker image in the format repo/imagename:tag (e.g., hosted on Docker Hub or the CERIT-SC Harbor registry). Always build on top of an existing Jupyter Docker Stack image to ensure all required JupyterHub startup scripts are present. For building images from scratch, see Custom Environments.

Configure StorageCopy Anchor Link

• Default Persistent Storage (/home/jovyan): Every notebook automatically mounts a persistent disk at /home/jovyan (or /home/rstudio in RStudio). Always save your scripts and results here so they persist when your server stops. You can also reuse and share this storage across different notebook instances. When creating a server you choose between a New home directory, reusing an Existing one, or Erase if home exists.
• MetaCentrum Home (/home/meta/<username>): You can mount one MetaCentrum storage home or project directory per notebook to access your existing grid files directly.
• S3 Object Storage: You can mount S3-compatible object storage; it appears inside the notebook at /storage/s3.

Allocate Hardware Resources (CPU, RAM, GPU)Copy Anchor Link

Request only what your current workflow requires:

• CPU (Guaranteed: 1 | Max: 96): Soft-handled. If your code exceeds your CPU limit, your notebook will not crash—the computation will simply be throttled (run slower).
• RAM / Memory (Guaranteed: 4 GB | Max: 1024 GB): Hard limit. Warning: If your code consumes more RAM than allocated, the kernel/process will be abruptly terminated (Out-Of-Memory error), though your notebook interface will stay open.
• GPU (Whole GPU or NVIDIA MIG Slices):
  • Whole GPU: Only select a GPU if you are also selecting a GPU-enabled image (marked with ...with GPU...), which includes the necessary CUDA drivers. Depending on availability, dedicated whole GPUs include NVIDIA A10 (24 GB), A40 (48 GB), A100 (80 GB), H100 (80/94 GB), and RTX PRO 6000 Blackwell; the currently free capacity of each type is shown in the spawn form.
  • NVIDIA MIG (MultiInstance GPU): Ideal for smaller models or testing. You can request isolated hardware slices of an NVIDIA A100 (80 GB) card—up to 4 slices of MIG (10GB) or MIG (20GB). Each slice acts as an independent GPU with dedicated memory. (Note: Using more than 1 MIG slice requires multi-GPU support in your code).

Once you have chosen image, storage, and resources, click Start and your server will spawn within roughly half a minute.