TensorRT 6.0 Float32 engine will use MaxWorkspace


I need convert caffe model to TRT with float32.

When GPU has 4G of free memory,

There will have error: out of memory if MaxWorkspace is set to 6G,

but it is worked fine with MaxWorkspace 2G.

Did I miss any config?


MaxWorkspaceSize is the maximum GPU temporary memory which the TensorRT engine can use at execution time.

Please refer below troubleshooting link on how do I choose the optimal workspace size?