NX tensorrt FP16 engine generation error

I’m using jetpack 4.4

Building an engine from file ctdet_hardnet_68_480x480.onnx; this may take a while…
[TensorRT] ERROR: …/builder/cudnnBuilderUtils.cpp (427) - Cuda Error in findFastestTactic: 700 (an illegal memory access was encountered)
[TensorRT] ERROR: …/rtSafe/safeRuntime.cpp (32) - Cuda Error in free: 700 (an illegal memory access was encountered)
terminate called after throwing an instance of ‘nvinfer1::CudaError’
what(): std::exception
Aborted (core dumped)

You can find the onnx file here:

I have encountered a very similar error from a previous post.
From my experiments it seems to be from the groups parameter in the Conv2D.
When will the next version come out and what exactly is causing this?

Hi,

This is a known issue and is originally reported in Cuda Error in findFastestTactic: 700 when using tensorrt in fp16 mode on Xavier NX with Jetpack4.4.
The fix is integrated into TensorRT-7.2 and will be available for public in the the upcoming release.

Please pay attention to our announcement and update.
Thanks.

More, this issue only occurs on fp16 precision mode.
So maybe you can use fp32 mode as a temporal fix.

Thanks.

Hi,

We confirmed that his issue is fixed in our future JetPack release.
Please wait for our announcement for the new software.

Thanks.