Dark mode Light mode

Getting Started with Great Lakes

Content

• Available Resources
• Creating an ARC account
• Configuration
• Launching a Session
  • Launching an Interactive Session
  • Launching an SSH Session
  • Creating a Job Template
  • Connecting via SSH on VS Code
• Setting Up Your Own Base Conda
• Useful Resources

---

Available Resources

You must connect to MWireless or U-M VPN to access Great Lakes.

Compute

• Great Lakes
  • Shared requestable servers
  • ~4,000 CPU hours (~560 GPU hours) for the whole semester
  • Expect some wait time for GPUs during peak hours and around finals
  • The availability depends on the resource and the amount of time requested. If the wait time is too long, try reducing the requested time or switching to another partition.
  • All the GPUs on Great Lakes are fast enough for most cases.
  • We can only use up to five GPUs as a whole at the same time. Please terminate your session when you finish your work so that others can access the GPUs.

Storage

• Home directory
  • 80 GB limit per user
  • Available at /home/{UNIQNAME}
• Scratch space
  • 10TB limit as a whole
  • Data may be removed if not used in 30 days
  • Available at /scratch/pat464564w26_class_root/pat464564w26_class/{UNIQUENAME}

---

Creating an ARC Account

1. Request an HPC login here

   Please put “Accessing Great Lakes for the course “Generative AI for Music and Audio Creation” (PAT 464/564, Winter 2026)” in the request.

   

2. Log into your account in the web portal

   You must connect to MWireless or U-M VPN to access Great Lakes.

   

---

Configuration

Recommended Configuration for CPU-only Workloads

• Partition: standard
• Number of cores: 4 / 8 / 16 / 36
• Memory (GB): 28 / 56 / 112 / 180

Click to show the screenshot

Recommended configuration for GPU Workloads

There are three GPU-available partitions. spgpu is usually the one with the least wait time, which has 224 NVIDIA A40 GPUs (48GB VRAM). On spgpu, you can request 4 CPUs and 48 GB RAM without extra cost for each GPU requested on the spgpu partition, so the following is the most cost-effective option.

• Partition: spgpu
• Number of cores: 4
• Memory (GB): 48

Click to show the screenshot

Other configurations

For each GPU requested, Great Lakes provides some CPU cores and RAM memory without additional charge. Please use the following configurations for each partition to maximize the freely-provided CPUs and memory (as of Winter 2025; see the current configurations):

Partition	CPU cores	Memory (RAM)	GPU	GPU speed	GPU memory	#GPUs available
spgpu	4	48 GB	A40	faster	48 GB	224
gpu_mig40	8	124 GB	A100	fastest	40 GB	16
gpu	20	90 GB	V100	fast	16 GB	52
standard*	1	7 GB	-	-	-	-

Here is the cost for each partition (as of Winter 2025; see the current rates):

Partition	Hourly cost	CPU hours equivalent
spgpu	0.11	7.33
gpu_mig40	0.16	10.66
gpu	0.16	10.66
standard*	0.015	1

---

Launching a Session

Launching an Interactive Session

1. Navigate to “Interactive Apps” on the web portal

2. Select the application

   You’ll likely want to select “JupyterLab” for the assignments.

   

3. Specify the configuration of the instance (see above for suggestions)

   If you’re unsure, select “python3.11-anaconda/2024.02” as your Python distribution.

   Also, make sure to select “pat464564w26_class” as the slurm account so that the cost is charged to the right account.

   

4. Connect to the launched instance

   

5. Make sure to delete the job if your workload is finished earlier

Launching an SSH Session

This will create a persistent compute node that won’t end automatically. Please make sure to kill the job after your workload is finished so that the resources are released.

1. Navigate to “Job Composer” on the web portal for Great Lakes

   

2. Click “New Job > From Default Template”

   

3. Click “Open Editor”

   

4. Replace everything in the editor with the following and click “Save”

   This will request 1 GPU with 4 CPUs and 48 GB RAM for 2 hours on Great Lakes. Adjust the time limit (in the format of D-HH:MM:SS) if the wait time is too long. See above for configuration suggestions.

   #!/bin/bash
   #SBATCH --account=pat464564w26_class
   #SBATCH --partition=spgpu
   #SBATCH --cpus-per-task=4
   #SBATCH --mem=48G
   #SBATCH --gres=gpu:1
   #SBATCH --time=2:00:00
   echo $SLURMD_NODENAME
   tail -f /dev/null
   

   

5. In the “Jobs” tab, select the created job and click “Submit”

   

6. At the right panel, click the “slurm-{ID}.out” file

   

7. Copy the node name

   

8. To access the compute node, you will need to first SSH into the login node

   ssh {UNIQNAME}@greatlakes.arc-ts.umich.edu
   

   And then you can SSH into the compute node

   ssh {NODENAME}.arc-ts.umich.edu
   

   Click to show how to set up the SSH configuration file

   Add the following lines to the SSH configuration file (usually at ~/.ssh/config):

   For Lighthouse:

   Host lighthouse-login
      HostName lighthouse.arc-ts.umich.edu
      User {UNIQNAME}
   
   Host lighthouse
      HostName lh2300.arc-ts.umich.edu
      User {UNIQNAME}
      ProxyJump lighthouse-login
   

   For Great Lakes:

   Host greatlakes-login
      HostName greatlakes.arc-ts.umich.edu
      User {UNIQNAME}
   
   Host gl*.arc-ts.umich.edu
      User {UNIQNAME}
      ProxyJump greatlakes-login
   

9. Make sure to delete the job after your workload is finished

   

Creating a Job Template

You can create a template to save some time from editing the job script every time.

1. Navigate to the “Templates” tab

2. Click “New Template”

   

3. Enter the template name and click “Save”

   

4. Select the created template and click “View Files”

   

5. Click the triple dots next to “main_job.sh” and click “Edit”

   

6. Replace everything in the editor with the following and click “Save”

   This will request 1 GPU with 4 CPUs and 48 GB RAM for 2 hours on Great Lakes. Adjust the time limit (in the format of D-HH:MM:SS) if the wait time is too long. See above for configuration suggestions.

   #!/bin/bash
   #SBATCH --account=pat464564w26_class
   #SBATCH --partition=spgpu
   #SBATCH --cpus-per-task=4
   #SBATCH --mem=48G
   #SBATCH --gres=gpu:1
   #SBATCH --time=2:00:00
   echo $SLURMD_NODENAME
   tail -f /dev/null
   

   

7. Return to the previous page and click “Create New Job”

   

Connecting via SSH on VS Code

1. Add the following lines to the SSH configuration file (usually at ~/.ssh/config):

   Host greatlakes-login
      HostName greatlakes.arc-ts.umich.edu
      User {UNIQNAME}
   
   Host gl*.arc-ts.umich.edu
      User {UNIQNAME}
      ProxyJump greatlakes-login
   

   On Great Lakes, the node name will change when you launch a new job as you’ll be assigned to one of the many compute nodes on the cluster.

2. Click “Open Remote Window” at the bottom left of the VS Code window

   

3. Click “Connect to Host”

   

4. Click “greatlakes” or “lighthouse”

   

5. Click the highlighted “details” link in the pop-up box at the bottom right

   Ignore the pop-up password prompt at the top

   

6. Enter your password at the terminal tab and hit “Enter”

   Ignore the pop-up password prompt at the top

   

7. Authenticate through Duo for two-factor authentication

   

8. Click the pop-up password prompt at the top and hit “Enter” 2-3 times until the pop-up prompt box at the bottom right changes from “Initializing VS Code Server” to “Setting up SSH tunnel”

9. Click the highlighted “details” link in the pop-up box at the bottom right

   Ignore the pop-up password prompt at the top

   

10. Enter your password at the terminal tab and hit “Enter”

    Ignore the pop-up password prompt at the top

    

11. Authenticate through Duo for two-factor authentication

    

12. It should say “Connected to SSH Host - Please do not close this terminal”

    

13. Click the pop-up password prompt at the top and hit “Enter” 2-3 times until the pop-up box at the bottom right disappears (you should also see “SSH: greatlakes” at the bottom left)

    

14. Now, you may start coding in VS Code! You can click the “+” sign at the right to open a terminal to verify the connection.

    

    Once you click it, you should see that you’re connected to the compute node:

    

---

Setting Up Your Own Base Conda

Follow the instructions below if you want to set up your own base Conda other than the default one installed on the cluster.

1. Follow the official instructions to install Conda

2. Update the Slurm script as follows:

   #!/bin/bash
   #SBATCH --account=aimusic_project
   #SBATCH --partition=aimusic_project
   #SBATCH --cpus-per-task=8
   #SBATCH --mem=62G
   #SBATCH --gres=gpu:1
   
   CONDA_EXE=/home/{UNIQNAME}/miniconda3/bin/conda
   CONDA_PREFIX=/home/{UNIQNAME}/miniconda3
   CONDA_PYTHON_EXE=/home/{UNIQNAME}/miniconda3/bin/python
   export PATH="/home/{UNIQNAME}/miniconda3/bin:/home/{UNIQNAME}/miniconda3/condabin:$PATH"
   
   echo $SLURMD_NODENAME
   tail -f /dev/null
   

---

Useful Resources

• Guide to Using Great Lakes
• Great Lakes Cheat Sheet
• Great Lakes User Guide

---