onnx convert to TensorRT engin Error[1]: Unexpected exception bad allocation

Description

Can't convert an onnx model to TensorRT engine under the following condition. It seems to be an error of memory allocation.
But when I change the GPU from 3070 to 2070 or use 3070 without --fp16, it works.

Environment

TensorRT Version: 8.2.0.6
NVIDIA GPU: RTX 3070
NVIDIA Driver Version: 469.49
CUDA Version: 11.4
CUDNN Version: 8.2.1
Operating System: windows 10
Python Version (if applicable):
Tensorflow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if so, version):

Relevant Files

model.onnx

Steps To Reproduce

trtexec.exe --onnx=model.onnx --saveEngine=model.engine --buildOnly --workspace=4096 --fp16
and I got the following error:

[11/09/2021-17:05:46] [I] Finish parsing network model
[11/09/2021-17:05:46] [I] [TRT] [MemUsageSnapshot] Builder begin: CPU 4097 MiB, GPU 2099 MiB
[11/09/2021-17:05:47] [I] [TRT] [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +882, GPU +336, now: CPU 4994, GPU 2477 (MiB)
[11/09/2021-17:05:47] [I] [TRT] [MemUsageChange] Init cuDNN: CPU +0, GPU +11, now: CPU 4994, GPU 2488 (MiB)
[11/09/2021-17:05:47] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored.
[11/09/2021-17:12:13] [E] Error[1]: Unexpected exception bad allocation
[11/09/2021-17:12:13] [E] Error[2]: [builder.cpp::nvinfer1::builder::Builder::buildSerializedNetwork::561] Error Code 2: Internal Error (Assertion enginePtr != nullptr failed. )
[11/09/2021-17:12:13] [E] Engine could not be created from network
[11/09/2021-17:12:13] [E] Building engine failed
[11/09/2021-17:12:13] [E] Failed to create engine from model.
[11/09/2021-17:12:13] [E] Engine set up failed

No source files or tests are named. Start by reproducing the exact trtexec command with model.onnx, then compare the RTX 3070 --fp16 failure with the working RTX 2070 or non-FP16 runs and inspect the reported allocation and engine-creation errors. Done means the supplied model builds successfully under the affected configuration or the failure is diagnosed with actionable evidence.