search

openai / roboschool

DEPRECATED: Open-source software for robot simulation, integrated with OpenAI Gym.
Other
2.13k stars 488 forks source link

changing GPU device_count has an impact on result #185

Closed TheCrazyT closed 5 years ago

TheCrazyT commented 5 years ago

Not shure if that is a tensorflow-bug or something related to it. I installed tensorflow-gpu and changed the value for GPU's from 0 to 1. device_count = { "GPU": 1 } demo_race2.py now suddenly fails, the agents now forget how to walk and instantly fall. Sadly can't figure out the source of the problem.

Edit: Oh well ... guess my tensorflow-gpu installation is broken. Another gpu-example with tensorflow failed as well. I was expecting more serious errors with a wrong tensorflow-gpu installation instead of wrong results. (Note to myself, installing nvidia cuda 10.1 and simply renaming the dll's is NO sollution ... will try with cuda 10.0 next time)

TheCrazyT commented 5 years ago

yes ... that was it, cuda version 10.1 is just not compatible.