I'm trying to optimize my models' performance using tensorboard profiling. I'm using M1 Max MBP, and the profiler shows that I'm using the CPU (see the picture below). However, when I run
device_name = tf.test.gpu_device_name()
if not device_name:
raise SystemError('GPU device not found')
print('Found GPU at: {}'.format(device_name))
it returns Found GPU at: /device:GPU:0Metal device set to: Apple M1 Max.
Furthermore, the profiler shows that I'm only using the host:
TF Op Placementinfo
generally desired to have more ops on device
Host: 100.0%
Device: 0.0%
Is TensorFlow not using the GPU during the training even if it can detect it? How can I fix this issue and use the GPU?
