'How to interpret Tensorboard kernel_stats and tensorflow_stats?
I'm new to tensorboard and got pretty confused with tables presented in profile section. Can you please help me with several questions?
- As far as I understand, kernel_stats subsection shows how much time was spent on different Tensorflow-GPU operations (such as convolutions, activations, etc.). And tensorflow_stats shows operations for CPU, GPU and transfers between them.
Here is my kernel_stats
And here is my tensorflow_stats for device
How do these operations match with each other? I can see that total tensorflow_stats time except "Send" and "Unknown" operations is equal to total kernel_stats time (~24ms) and some operations from the first table match some operations in the second table but the global correspondence is not obvious. Do these tables show the same operations or I got it wrong?
What exactly "Send" operation means?
Is there a way to get inside "Unknown" operation and figure out what happens during this operation?
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|