Skip to Main Content
Llama.swap Set the user name from the user email address

Installing Conda and PyTorch on NVIDIA DGX Spark

Conda PyTorch Python
When you install a GPU-enabled version of PyTorch via Conda, Conda manages the necessary NVIDIA CUDA toolkit and libraries internally, without requiring a complex, system-wide CUDA installation. This significantly simplifies the process of getting PyTorch to run on your GPU for faster computations.

Hardware 

  • GPU: GB10 Grace Blackwell Superchip 
  • CPU: ARM-based Grace (aarch64 architecture)
  • OS: Ubuntu 24.04
  • CUDA: 13.0.2 NVIDIA

Step 1 – Conda Installation

https://conda-forge.org/download/

sudo mkdir -p $HOME/codebase/conda 
sudo mkdir -p /usr/local/miniforge3
sudo chown $USER:$USER /usr/local/miniforge3

cd $HOME/codebase/conda 
curl -LO https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Linux-aarch64.sh  

bash Miniforge3-$(uname)-$(uname -m).sh -b -u -p /usr/local/miniforge3

Step 2 – Update your profile

vi $HOME/.profile

# Add to Extend PATH
PATH=$PATH:/usr/local/miniforge3/bin

Exit and log back in

Step 3 – Activate Conda

#                                                                                                                                                                                                                                                            
# To activate this environment, use                                                                                                                                                                                                                          
# conda activate ai
#
# To deactivate an active environment, use
# conda deactivate

# Run: 
conda init
conda create -n ai python=3.12
conda activate ai

 

Step 4 – Install PyTorch 

https://pytorch.org/get-started/locally/

 

# Instead of using the suggested pip instalaltion line I use:

python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

# In case of issues here is the un-install command:
# pip uninstall torch torchvision torchaudio -y

Step 5 – Validate the installation works

# Example One line validation
# Run the code:
python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"

# Example output:
# $ python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"
# $  2.11.0+cu130
# $ True

# Example Code validation

sudo mkdir -p $HOME/codebase/bin/

echo "
import torch
print(\"PyTorch version:\", torch.__version__)
if torch.cuda.is_available():
    print(\"GPU:\", torch.cuda.get_device_name(0))
else:
    print(\"No CUDA GPU detected\")
" > $HOME/codebase/bin/cuda.ok.py

# Run the code:
python $HOME/codebase/bin/cuda.ok.py

# Example output:
# $ python $HOME/codebase/bin/cuda.ok.py
# $ PyTorch version: 2.11.0+cu130
# $ GPU: NVIDIA GB10

 

 
www.1apx.com
_