System Description

Resource Limits per User

Each user has 100GB of persistent file storage. The following table outlines the resource allocations of each server option.

Resource Descriptions

Resource

Description

CPU 4CPU/16GB

You have access to:

  • Up to 4 CPUs when system demand allows (1 CPU guaranteed).

  • Up to 16GB of RAM when system demand allows (2GB guaranteed).

This option is for basic CPU needs.

A100 20GB VRAM GPU slice, 12CPU/72GB

You have access to:

  • A dedicated slice of one A100 GPU with 20GB of GPU memory (VRAM). Other users cannot use your slice.

  • Up to 12 CPUs when system demand allows (4 CPUs guaranteed).

  • Up to 72GB of RAM when system demand allows (16GB guaranteed).

H200 16GB VRAM GPU slice, 16CPU/32GB

You have access to:

  • A dedicated slice of one H200 GPU with 16GB of GPU memory (VRAM). Other users cannot use your slice.

  • Up to 16 CPUs when system demand allows (4 CPUs guaranteed).

  • Up to 32GB of RAM when system demand allows (16GB guaranteed).

GPU Slices

Each GPU option gives your session a dedicated slice of one GPU via NVIDIA Multi-Instance GPU (MIG). The slice’s GPU memory and compute are yours alone, so other users can’t slow down or crash your session. Because a slice is only part of a GPU, compute-heavy jobs (such as long model training runs) run slower than they would on a whole GPU.

Each GPU holds a fixed number of slices. If all slices are in use, see “0/<X> Nodes are Available” Error.

To check your slice’s GPU memory in PyTorch:

import torch
free, total = torch.cuda.mem_get_info()
print(f"GPU memory = {total/(1024**3):.1f}GB, free = {free/(1024**3):.1f}GB")

Requesting an Allocation Increase

You can request more cores and memory within reason. More storage space can be requested as well. Larger GPU slices (A100 40GB and H200 32GB) can also be requested. For more GPU than the largest slice, such as whole or multiple GPUs, see the GPU resources available through Illinois Computes, including the Campus Cluster and DeltaAI.

If the request is approved, the increase will last until the end of the current semester. Your allocation will show up as a new Resource choice at the bottom of the drop down menu on the Session Options screen.

Session Options screen with the Resource drop down menu showing the increased allocation as the last item.

In order to request an allocation increase submit a support request and provide a justification for the request.

Actual CPU Core and Memory Limits

Warning

/dev/shm uses your RAM allocation.

Because they can vary based on system demand, it is useful to know your actual CPU and memory limits. These actual limits are hard limits that you cannot exceed. If a process uses more than your actual memory limit, the process will be ended/canceled.

Use the following Python script to see your actual CPU core and memory limits:

import os
print(f"memory = {int(os.environ.get('MEM_LIMIT'))/(1024**3)}GB")
print(f"cores  = {os.environ.get('CPU_LIMIT')}")

# The output will look like this:
# memory = XX.0GB
# cores  = X.0

Notebook Duration Limits

Notebooks are temporary and intended to last a maximum of 24 hours. Notebooks older than 24 hours may be ended without warning.

Notebooks run live in a web browser, if you close your web browser or lose your internet connection for more than 1 hour, your notebook and running processes may end. It is expected that running processes will continue if your connection is interrupted for less than 1 hour.

User Directory

Files in your user directory are persistent.

Daily snapshots are run on the filesystem for the /home area. These snapshots allow you to go back to a point in time and retrieve data you may have accidentally modified, deleted, or overwritten. These snapshots are not backups and reside on the same hardware as the primary copy of the data.

To access snapshots for data in your /home directory, run cd ~/.snapshot/snapshots-daily-_YYYY-MM-DD_HH_mm_ss_UTC/ where “YYY-MM-DD_HH_mm_ss” is the timestamp of the snapshot you want to recover from. To list the available snapshots, run the command: ls ~/.snapshot/

Check this link for more details.

Image Options

Python

The Python image is a JupyterLab environment with Python and Conda installed. Python is a general-purpose programming language. Conda is an open source package management system.

PyTorch

The PyTorch image is a JupyterLab environment with PyTorch installed. The GPU resource is recommended for this image. NVIDIA provides a brief overview of GPU computing at the beginning of the Using GPUs with Python webinar hosted by NCSA.

R

The R image is a JupyterLab environment with R Kernel installed. This can be used for R development in the Jupyter interface. R is a programming language and software environment for statistical computing and graphics.

RStudio

The RStudio image has R, RStudio, and jupyter-rsession-proxy installed. Take a look at the RStudio Get Started Guide if you are new to using RStudio.