trt8.0 in jestion convert success and trt8.6 in x86 convert failed！！！

[luoshiyong]

my onnx convert success in jeston NX with trt8.0 and infer good~

but when i use same file convert in 3090(x86_64) with trt8.6(ubuntu20.04) convert failed ,and i also try it with trt8.5 in win11, failed too. The error is

[luoshiyong]

i try to increase the opset verion and use onnx-simpy but no use

[lix19937]

You can try to use latest version of tensorrt.

[luoshiyong]

i want ask why low-version trt can export ,however high-version can not

[lix19937]

i want ask why low-version trt can export , however high-version can not

I means you can try to use latest version of tensorrt at jeston NX not x86 PC.
From your desc, you not do some experiment with high-version(trt8.6/8.5) on jeston NX.

NOTE: Different arch of OS and GPUs, the cuda-x libs are different.
Sometimes, high-version maybe introduced bug.

[luoshiyong]

usually trt on jeston NX is diffcult to upgrade because deepstream version. I mainly want to export trt on mypc(win11 trt8.5.1.7) and my server(ubuntu 20.04 trt8.6), can you give me some advice for this error? i read over 3 issues talk about this error but no standard response~thanks your response!

[zerollzeng]

but when i use same file convert in 3090(x86_64) with trt8.6(ubuntu20.04) convert failed ,and i also try it with trt8.5 in win11, failed too. The error is

Would be great if you can try TRT 10, if still fails, please provide a reproduce.