PyTorch on NVIDIA DGX Spark GB10 on CUDA 13 - ARM64

[obriensystems]

• https://forums.developer.nvidia.com/t/effective-pytorch-and-cuda/348230
• https://discuss.pytorch.org/t/nvidia-dgx-spark-support/223677

(venv-t214) michael@spark-7d19:~/wse_github/ObrienlabsDev/machine-learning$ pip3 install torch torchvision --index-url ``https://download.pytorch.org/whl/cu130``
Looking in indexes: https://download.pytorch.org/whl/cu130
Requirement already satisfied: torch in ./venv-t214/lib/python3.12/site-packages (2.9.0)
Collecting torchvision
  Downloading https://download.pytorch.org/whl/cu130/torchvision-0.24.0-cp312-cp312-manylinux_2_28_aarch64.whl.metadata (5.9 kB)
Requirement already satisfied: filelock in ./venv-t214/lib/python3.12/site-packages (from torch) (3.20.0)
Requirement already satisfied: typing-extensions>=4.10.0 in ./venv-t214/lib/python3.12/site-packages (from torch) (4.15.0)
Requirement already satisfied: setuptools in ./venv-t214/lib/python3.12/site-packages (from torch) (80.9.0)
Requirement already satisfied: sympy>=1.13.3 in ./venv-t214/lib/python3.12/site-packages (from torch) (1.14.0)
Requirement already satisfied: networkx>=2.5.1 in ./venv-t214/lib/python3.12/site-packages (from torch) (3.5)
Requirement already satisfied: jinja2 in ./venv-t214/lib/python3.12/site-packages (from torch) (3.1.6)
Requirement already satisfied: fsspec>=0.8.5 in ./venv-t214/lib/python3.12/site-packages (from torch) (2025.10.0)
Requirement already satisfied: numpy in ./venv-t214/lib/python3.12/site-packages (from torchvision) (2.3.4)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in ./venv-t214/lib/python3.12/site-packages (from torchvision) (12.0.0)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in ./venv-t214/lib/python3.12/site-packages (from sympy>=1.13.3->torch) (1.3.0)
Requirement already satisfied: MarkupSafe>=2.0 in ./venv-t214/lib/python3.12/site-packages (from jinja2->torch) (3.0.3)
Downloading https://download.pytorch.org/whl/cu130/torchvision-0.24.0-cp312-cp312-manylinux_2_28_aarch64.whl (7.9 MB)
   ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 7.9/7.9 MB 52.5 MB/s  0:00:00
Installing collected packages: torchvision
Successfully installed torchvision-0.24.0


michael@spark-7d19:~/wse_github/ObrienlabsDev/machine-learning$ cat pytorch.py 
# https://github.com/ObrienlabsDev/machine-learning/issues/49
import torch
import os

print("cpus: ", os.cpu_count())

print("Torch:", torch.__version__)
print("CUDA:", torch.version.cuda)
print("GPU available:", torch.cuda.is_available())
print("GPU:", torch.cuda.get_device_name(0))


not working for GB10
(venv-t214) michael@spark-7d19:~/wse_github/ObrienlabsDev/machine-learning$ python pytorch.py 
Torch: 2.9.0+cpu
CUDA: None
GPU available: False
Traceback (most recent call last):
  File "/home/michael/wse_github/ObrienlabsDev/machine-learning/pytorch.py", line 6, in <module>
    print("GPU:", torch.cuda.get_device_name(0))
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/wse_github/ObrienlabsDev/machine-learning/venv-t214/lib/python3.12/site-packages/torch/cuda/__init__.py", line 582, in get_device_name
    return get_device_properties(device).name
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/michael/wse_github/ObrienlabsDev/machine-learning/venv-t214/lib/python3.12/site-packages/torch/cuda/__init__.py", line 614, in get_device_properties
    _lazy_init()  # will define _get_device_properties
    ^^^^^^^^^^^^
  File "/home/michael/wse_github/ObrienlabsDev/machine-learning/venv-t214/lib/python3.12/site-packages/torch/cuda/__init__.py", line 403, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

[obriensystems]

working via docker

(venv-t214) michael@spark-7d19:~/wse_github/ObrienlabsDev/machine-learning$ sudo docker pull nvcr.io/nvidia/pytorch:25.09-py3
25.09-py3: Pulling from nvidia/pytorch
cc43ec4c1381: Pull complete 
...
216cdf20f351: Pull complete 
Digest: sha256:f3b4a33fd60d5e0358287bb44d57e91eda45e8f1681fef1bfd6c969371417130
Status: Downloaded newer image for nvcr.io/nvidia/pytorch:25.09-py3
nvcr.io/nvidia/pytorch:25.09-py3
(venv-t214) michael@spark-7d19:~/wse_github/ObrienlabsDev/machine-learning$ sudo docker run --gpus all -it --rm --ipc=host nvcr.io/nvidia/pytorch:25.09-py3
== PyTorch ==
NVIDIA Release 25.09 (build 210907838)
PyTorch Version 2.9.0a0+50eac81
Container image Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
Copyright (c) 2014-2024 Facebook Inc.

root@c8522e0e1502:/workspace# vi pytorch.py

import torch
import os

print("cpus: ", os.cpu_count())

print("Torch:", torch.__version__)
print("CUDA:", torch.version.cuda)
print("GPU available:", torch.cuda.is_available())
print("GPU:", torch.cuda.get_device_name(0))

root@c8522e0e1502:/workspace# python pytorch.py 
/usr/local/lib/python3.12/dist-packages/torch/cuda/__init__.py:63: FutureWarning: The pynvml package is deprecated. Please install nvidia-ml-py instead. If you did not install pynvml directly, please report this to the maintainers of the package that installed pynvml for you.
  import pynvml  # type: ignore[import]
cpus:  20
Torch: 2.9.0a0+50eac811a6.nv25.09
CUDA: 13.0
GPU available: True
GPU: NVIDIA GB10

[obriensystems]

CIFAR example
https://docs.pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html

or
https://build.nvidia.com/spark/pytorch-fine-tune
https://build.nvidia.com/spark/pytorch-fine-tune/instructions