defaultAllocator.cpp::nvinfer1::internal::DefaultAllocator::allocate::63] Error Code 1: Cuda Runtime (out of memory)

[geoexploring]

trafficstars

Change the input image size of ONNX model from 1 * 3 * 1024 * 1024 to 1 * 3 * 2000 * 2000, get an error when converting ONNX to trtmodel:

 defaultAllocator.cpp::nvinfer1::internal::DefaultAllocator::allocate::63] Error Code 1: Cuda Runtime (out of memory).

Environment

cudnn  cudnn8.4.1.50
cuda  11.2
TensorRT 8.4.2.4

Modifying the maxWorkspaceSize, either larger or smaller will not solve the problem. Is there any solution?

thanks

[zerollzeng]

This looks like a bug, can you share the onnx here?

[geoexploring]

@zerollzeng ，thank you for your attention，here's the model: random_param_4_nvidia_test_2e3_2e3.onnx, please download.

thanks again

[zerollzeng]

Thanks for reporting this, I've filed a internal bug to track it.

[zerollzeng]

this will be fixed in TRT 8.5.