'Running the same detection model on different GPUs
I recently ran in to a bit of a glitch where my detection model running on two different GPUs (a Quadro RTX4000 and RTX A4000) on two different systems utilize the GPU differently. The model uses only 0.2% of GPU on the Quadro system and uses anywhere from 50 to 70% on the A4000 machine. I am curious about why this is happening. The rest of the hardware on both the machines are the same.
Additional information: The model uses a 3D convolution and is built on tensorflow.
Solution 1:[1]
Looks like the Quadro RTX4000 does not use GPU.
The method tf.test.is_gpu_available() is deprecated and can still return True although the GPU is not used.
The correct way to verify the usage of the GPU availability + usage is to check the output of the snippet:
tf.config.list_physical_devices('GPU')
On the Quadro machine you should also run (in terminal):
watch -n 1 nvidia-smi
to see real-time the amount of GPU memory used.
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|---|
| Solution 1 |