i used a tesla T4 GPU and achieved about 100ms per point cloud (for 1024 points and the 8x8x8 gaussian grid), how does the diff happened？ GPU performence？cuda version？
my env is ubuntu14 + cuda10 + cudnn7 + T4 GPU + tensorflow1.13.1
and i found net structure is related to batch_size, in order to evaluate time consume
i trained with batch size = 1
sitzikbs
commented
5 years ago
i used a tesla T4 GPU and achieved about 100ms per point cloud (for 1024 points and the 8x8x8 gaussian grid), how does the diff happened？ GPU performence？cuda version？ my env is ubuntu14 + cuda10 + cudnn7 + T4 GPU + tensorflow1.13.1 and i found net structure is related to batch_size, in order to evaluate time consume i trained with batch size = 1