Running tutorial_example_trainer (similar leak for my own omnihub scripts) results in ~70MB of memory being lost. free shows the memory is used, but ps shows that no process is using the memory. Nothing short of a reboot seems to free up the memory.
Reproduce with the following (After the ~10 minute run, you should be down ~7GB of memory)
for i in seq 1 100; do ./bazel-bin/tensorflow/cc/tutorials_example_trainer --use_gpu; done
Running tutorial_example_trainer (similar leak for my own omnihub scripts) results in ~70MB of memory being lost. free shows the memory is used, but ps shows that no process is using the memory. Nothing short of a reboot seems to free up the memory.
Reproduce with the following (After the ~10 minute run, you should be down ~7GB of memory) for i in seq 1 100; do ./bazel-bin/tensorflow/cc/tutorials_example_trainer --use_gpu; done
os: 14.04 gpu: Titan X (361.28) commit: 0927e5e (from yesterday) cuda: 7.5 cudnn: 4.0.7