@underpostnet/underpost
Version:
66 lines (60 loc) • 1.9 kB
YAML
apiVersion: v1
kind: ConfigMap
metadata:
name: tf-gpu-test-script
namespace: default
data:
main_tf_gpu_test.py: |
import os
import tensorflow as tf
print("--- Starting GPU and Library Check ---")
gpus = tf.config.list_physical_devices("GPU")
if gpus:
try:
tf.config.set_visible_devices(gpus[0], "GPU")
logical_gpus = tf.config.list_logical_devices("GPU")
print(
f"TensorFlow detected {len(gpus)} Physical GPUs, {len(logical_gpus)} Logical GPUs. Using: {gpus[0].name}"
)
except RuntimeError as e:
print(f"RuntimeError during GPU configuration: {e}")
else:
print("TensorFlow did not detect any GPU devices. Running on CPU.")
print(f"XLA_FLAGS environment variable: {os.environ.get('XLA_FLAGS')}")
print(f"TF_XLA_FLAGS environment variable: {os.environ.get('TF_XLA_FLAGS')}")
print(f"TensorFlow version: {tf.__version__}")
print(f"Built with CUDA: {tf.test.is_built_with_cuda()}")
print(f"Is GPU available: {tf.config.list_physical_devices('GPU') != []}")
print("--- GPU and Library Check Complete ---")
apiVersion: v1
kind: Pod
metadata:
name: tf-gpu-test-pod
namespace: default
spec:
restartPolicy: Never
runtimeClassName: nvidia
containers:
- name: tensorflow-gpu-tester
image: nvcr.io/nvidia/tensorflow:24.04-tf2-py3
imagePullPolicy: IfNotPresent
command: ['python']
args: ['/app/main_tf_gpu_test.py']
resources:
limits:
nvidia.com/gpu: '1'
env:
- name: NVIDIA_VISIBLE_DEVICES
value: all
volumeMounts:
- name: tf-script-volume
mountPath: /app
volumes:
- name: tf-script-volume
configMap:
name: tf-gpu-test-script
items:
- key: main_tf_gpu_test.py
path: main_tf_gpu_test.py