LuoweiZhou / anet2016-cuhk-feature

Feature Extraction Toolbox from CUHK&ETHZ&SIAT submission to ActivityNet 2016

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Changing GPU device id fails.

hollowgalaxy opened this issue · comments

I was attempting to have one of the models on 1 gpu, while the other model on another gpu, however, I get a crash. I was setting to dev_ids 0, 1, and I verified nvidia-smi shows multiple gpus. Also verified that gpus work with tensorflow.
This happens after forward is called in action_caffe.py on line

out = self._net.forward(blobs=[score_name,], data=data)

Error:

 F0630 03:45:26.391547     1 cudnn_conv_layer.cu:34] Check failed: status == CUDNN_STATUS_SUCCESS (8 vs. 0)  CUDNN_STATUS_EXECUTION_FAILED

I do not think the code supports multi-GPU inference. You can refer to the original repo for details.

Thank you