search

GreNait / technical-thesis

This thesis contains all the work related to it. Python code, models, documentation etc.
0 stars 0 forks source link

llimiting the gpu usage for better teach #17

Open GreNait opened 4 years ago

GreNait commented 4 years ago

I have been able to lower the gpu memory limit by:

tf.config.experimental.set_virtual_device_configuration(gpus[0], [tf.config.experimental.VirtualDeviceConfiguration(memory_limit=8000)])

I controled it via the nvidia-smi command. Still, I got the black out if I tried to train the model.

GreNait commented 4 years ago

Because I have no idea how to proceed, I asked a question on stack overflow: https://stackoverflow.com/questions/60307931/as-soon-as-my-tf2-model-starts-training-the-display-goes-black

So I can go on with image augmetnation

GreNait commented 4 years ago

Stack overflow didn't help. But I figured out, that I was using Cuda 10.2 and only tf 2.0 After updating with conda update --tensorflow-gpu to tf2.1 it works now (appearently). I tested it witha limit of 5000MB. I will try 10000 next time.