bring_your_compute.md

If you are provisioning this on a non-Brev cloud instance, workstation, or local machine, please follow the steps below:

1. Provision a System Meeting These Requirements

All provisioned systems need to be RAPIDS capable. Here's what is required:

GPU: NVIDIA Volta™ or higher with compute capability 7.0+

• For most of the notebooks, we recommend a GPU with 24 GB VRAM or more, due to the dataset size, such as the L40S, which can be quickly deployed above. Some other common GPU options found in your favorite cloud service providers, your workstations, or your PC are:

Cloud/DGX	Workstation GPU	Consumer GPU
A/H/B100	A4000 ADA or better	5090
L40s/L40	A5000 or better	4090
A10/A10s	RTX6000 or better	3090

The out-of-core_processing notebook requires a large multigpu system when using the 11 million cell dataset, but a 48GB GPU is recommended if using the 1 million cell dataset. The workflows are respectively similar.

OS:

• Linux distributions with glibc>=2.28 (released in August 2018), which include the following:
  • Arch Linux, minimum version 2018-08-02
  • Debian, minimum version 10.0
  • Fedora, minimum version 29
  • Linux Mint, minimum version 20
  • Rocky Linux / Alma Linux / RHEL, minimum version 8
  • Ubuntu, minimum version 24.04
• Windows 11 using a WSL2 specific install

CUDA 13 & latest NVIDIA Drivers: Install the latest drivers for your system HERE

Note: RAPIDS is tested with and officially supports the versions listed above. Newer CUDA and driver versions may also work with RAPIDS. See CUDA compatibility for details.

2. Clone This Repo:

git clone https://github.com/clara-parabricks-workflows/single-cell-analysis-blueprint.git

3. Install Packaging Environment Software:

RAPIDS can be installed using Pip, Conda, or Docker. To replicate the same experience as in the Launchable, it is recommended to use Docker for installation.
3.1 Get Docker: Use the code below to download and install Docker on your system

curl -fsSL https://get.docker.com -o get-docker.sh
sh get-docker.sh

3.2 Get RAPIDS: Use the code below to pull the same RAPIDS container used in the Launchable. Example for 26.02:

docker run --gpus all --pull always --rm -it \
    --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 \
    -p 8888:8888 -p 8787:8787 -p 8786:8786
    -v ~/single-cell-analysis-blueprint:/home/rapids/notebooks/single-cell-analysis-blueprint
    nvcr.io/nvidia/rapidsai/notebooks:26.02-cuda13-py3.12

Note: The volume is currently assuming that you cloned the single-cell-analysis-blueprint repository into your HOME folder (~/single-cell-analysis-blueprint). If you did not clone it there, please change the ~/single-cell-analysis-blueprint portion of the above to the correct path before running the command.

4. Install the Extra RAPIDS-singlecell Libraries into the Running Container

Once the container is running,

• if the provisioned system is an external cloud or workstation, please use a web browser to navigate to the system's IP address and it's port 8888. Example http://192.168.1.2:8888
• if the provisioned system is your sustem, then use http://127.0.0.1:8888.

Once the JupyterLab instance loads, open the terminal form the JupyterLab GUI, and run:

cd /home/rapids/notebooks/single-cell-analysis-blueprint
pip install -r requirements.txt

For additional information on RAPIDS-singlecell please visit the RAPIDS-singlecell Docs For alternate installation instructions, please refer to the RAPIDS-singlecell Install Guide to install using pip, Conda, or Docker

Tips and Tricks

1. If after clicking a "Healthy" Port 8888 link in Deploy with Brev: Step 4, JupyterLab does not start, or the notebooks don't show, please try again in a few seconds. There is a known issue where there system needs a minute or two Also, sometimes, the page needs to be refreshed to update the status.
2. Currently, restarting an instance, after stopping it, will start up far faster than starting a new instance.
3. If you Stop the instance, all the data on the main storage will be retained for the next time you start it.
4. To conserve GPU memory, please remember to shut down your completed notebook's kernel before starting a new notebook.
5. If your data download gets interrupted, please delete the files you intended to download and try again.
6. The Standard RSC Instance (L40s) has 128GB of space (~90GB user usable).
7. If using an ARM based system, please conda install compile before installing RAPIDS-singlecell so that you can build scikit-misc from source/