Not able to fully use K 80 GPU on GCP

[Raja Raghudeep Emani]

Hi  ,
A typical resnet, model forward is taking 31ms for 0.2 data points in my home 6gb pc, but the same model forward is taking 51 ms for the same data on GCP k80 12GB. I am confused, missing something.

These are the stats while model is training on k80 gpu;

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 375.51                 Driver Version: 375.51                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla K80           On   | 0000:00:04.0     Off |                    0 |
| N/A   40C    P0   107W / 149W |    644MiB / 11439MiB |     85%      Default |
+-------------------------------+----------------------+----------------------+
                                                                              
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID  Type  Process name                               Usage      |
|=============================================================================|
|    0      3957    C   /usr/local/torch/install/bin/luajit            642MiB |
+-----------------------------------------------------------------------------+

It's hardly using any memory. Any thoughts?

Thanks

[Shivam(Google Cloud Support)]

This post is more suitable for GCE discussion forum. I would recommend to post it there.