2

With GPU isolation : multiple processes use the single GPU will be serialized, cause OOM easily link https://www.slideshare.net/mobile/Hadoop_Summit/deep-learning-on-yarn-running-distributed-tensorflow-etc-on-hadoop-cluster-v3 slide number 17

THEGreatGatsby
  • 121
  • 2
  • 2
    Can you give more context to your question? Is this a quote from some documentation? Can you link where you read it? – Nathan Reed Dec 17 '18 at 23:25

1 Answers1

4

It's saying that if you're using a single GPU, the multiple processes attempting to use the GPU will be queued up and run one at a time rather than some or all being run in parallel. If you have multiple GPUs some of the processes can be run at the same time.

user1118321
  • 3,401
  • 11
  • 14