ARM64 does not support NUMA - returning NUMA node zero

Spending many many hours, to get tensorflow 2.6 working on my Jetpack 4.6

Cuda compilation tools, release 10.2, V10.2.300
Build cuda_10.2_r440.TC440_70.29663091_0

STILL this old message basically says: We will use tensorflow without GPU acceleration???

tf.test.is_gpu_available(
… cuda_only=True,
… min_cuda_compute_capability=None
… )
WARNING:tensorflow:From :3: is_gpu_available (from tensorflow.python.framework.test_util) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.config.list_physical_devices('GPU') instead.
2021-10-29 18:31:51.165249: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:1017] ARM64 does not support NUMA - returning NUMA node zero
2021-10-29 18:31:51.165947: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1835] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at GPU support  |  TensorFlow for how to download and setup the required libraries for your platform.
Skipping registering GPU devices…
False

Hi,

This is a known issue.

TensorFlow v2.6.0+nv21.9 GPU mode is not working due to a CUDA library called cuTensor.so.
Our internal team is actively working on this.

Currently, please use the v2.5.0+nv21.8 package instead.
Thanks and sorry for the inconvience.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.