0

I am running ANN models, using tensorflow GPU and python (Windows 10). I never see the GPU usage goes above %20 regardless of the model/complexity... I recently upgraded to RTX 2070 hoping to achieve results faster, but because of the GPU utilisation not much changed.

Tried to re-install CudaNN, and tensorflow GPU no change. I have 32GB Ram, core i5 7400 3Ghz, Graphic card GiGabyte RTX 2070, Windows 10, Python, tensorflow cuda etc.. are latest version. There is no issue running the codes.

I was hoping to see nearly 100% GPU usage, but what happens is CPU usage goes to 100% and GPU <%20... Maybe its a hardware mismatch? (Core i5 for RTX 2070?)Performance screenshot

M Student
  • 29
  • 3
  • 4
    This question is not answerable without details about your network, input data and pipeline, etc. Gven the fact that the CPU is at 100%, all I can say is: your bottleneck is there. Probably in the data preprocessing, but it could b anywhere, really. Have a look at how to profile TF network executions and try to identify your bottlenecks – GPhilo Aug 01 '19 at 11:53
  • Please [check](https://stackoverflow.com/a/38019608/9291522) if Tensorflow even recognizes the GPU. – georg-un Aug 01 '19 at 12:01
  • Thanks for the feedback, the code to check the GPU runs OK, I can see the results and no error – M Student Aug 01 '19 at 12:06
  • In [51]: from tensorflow.python.client import device_lib ...: print(device_lib.list_local_devices()) [name: "/device:CPU:0" device_type: "CPU" memory_limit: 268435456 locality { } incarnation: 13946072752778883436 , name: "/device:GPU:0" device_type: "GPU" memory_limit: 6630704415 locality { bus_id: 1 links { } } incarnation: 6820793874645943545 physical_device_desc: "device: 0, name: GeForce RTX 2070, pci bus id: 0000:01:00.0, compute capability: 7.5" – M Student Aug 01 '19 at 12:08

0 Answers0