My system has Tensorflow 2.2 (Python 3.8, anaconda). I read about the Tensorflow XLA Compiler and to my understanding it is an alternative way to precompiled GPU operations that Tensorflow ships with. Even when I run code without eager execution or experimental_compile=True I see XLA service 0x5644b1cec990 initialized for platform CUDA (this does not guarantee that XLA will be used) in the terminal.
Also if it is linked in some way, the Executing op Fill in device /job:localhost/replica:0/task:0/device:GPU:0 takes considerably long time (~2mins) before proceeding to the next op.
Asked
Active
Viewed 747 times
0
shivampotdar
- 1
- 1
1 Answers
-1
As far as the release notes go: Yes, XLA now builds with windows and is shipped with all prebuild packages
Zulu02
- 1
- 1
-
it is included yes. But is it the default compiler / executor framework now, even when not enabled explicitly? – shivampotdar Jul 16 '20 at 11:35