Does anyone have instructions for how to use Tensorflow on the GPU on a DGX Spark? Is this possible yet?
NVIDIA optimized Tensorflow has reached EOL: End of Life Notices — NVIDIA AI Enterprise Notices
Try this by creating a virtual environment. It worked for me.
pip install nvidia-tensorflow[horovod] --extra-index-url=https://pypi.ngc.nvidia.com/
1 Like
Yes: I had to use this install method as well, but once done, TensorFlow worked fine:
pip install nvidia-tensorflow[horovod] --extra-index-url=https://pypi.ngc.nvidia.com/
Thanks to Santosh!
Hi, I am facing tensorflow issue too.
I running a container base on nvcr.io/nvidia/pytorch:25.10-py3, then installed tensorflow by pip install nvidia-tensorflow[horovod] --extra-index-url=https://pypi.ngc.nvidia.com/, package installed but error as below:
root@7ec1533a5c9f:/workspace# python3
Python 3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
2025-11-05 14:25:10.339985: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:485] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2025-11-05 14:25:10.347213: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:8473] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2025-11-05 14:25:10.350579: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1471] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered
2025-11-05 14:25:10.776737: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT: INTERNAL: Cannot dlopen all TensorRT libraries: FAILED_PRECONDITION: Could not load dynamic library 'libnvinfer.so.10.8.0'; dlerror: libnvinfer.so.10.8.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda/compat/lib.real:/usr/local/lib/python3.12/dist-packages/torch/lib:/usr/local/lib/python3.12/dist-packages/torch_tensorrt/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
>>> print(tf.config.list_physical_devices('GPU'))
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1762352719.666273 260 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1762352719.707974 260 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
I0000 00:00:1762352719.710736 260 cuda_executor.cc:1015] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
>>>
I also tried to run a program using tensorflow, more error:
2025-11-05 14:22:57.495784: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:388] MLIR V1 optimization pass is not enabled
2025-11-05 14:22:57.735448: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:531] Loaded cuDNN version 91400
W0000 00:00:1762352577.804734 240 gpu_timer.cc:114] Skipping the delay kernel, measurement accuracy will be reduced
W0000 00:00:1762352577.811749 240 gpu_timer.cc:114] Skipping the delay kernel, measurement accuracy will be reduced
Program finally complete and output correct result. However, it is far slower than tensorflow cpu version.