Upgrade tensorRT from 7.1.3 to 7.2.0+

[JeremyYuan]

Hi, how can I update update tensorRT from 7.1.3 to 7.2.0 or newer without JP for JP just provides ver 7.1.3.
since this page suggest upgrade it.
https://github.com/onnx/onnx-tensorrt/issues/374

[dusty_nv]

Hi @JeremyYuan, the version of TensorRT can’t be upgraded independently of the JetPack version, as there are dependencies on the version of CUDA, cuDNN, and kernel/drivers in JetPack. The next release of JetPack later this summer will come with upgraded TensorRT version.

For now, my recommendation is to try and export your ONNX model without using dynamic shapes, so that you may work around this issue.

[msripooja]

Hi @dusty_nv ,

The information you shared is very clear, but I have a requirement in which I have to work with dynamic shapes only. I verified the algorithm that I converted to ONNX using static/fixed shape inputs and there are absolutely no issues in TRT engine creation on Jetson AGX Xavier.

As I mentioned here:

TRT engine created on Tesla T4, Quadro P1000 but engine creation fails on Jetson AGX Xavier

Description I have an ONNX model created successfully from source algorithm, there are no errors in ONNX creation, I verified the ONNX model using ONNX checker. Using this ONNX model I am able to create TRT engine and run inference successfully on Tesla T4, Quadro P1000 but, when I try to create the TRT engine on Jetson AGX Xavier board it fails with the error: [TensorRT] ERROR: …/builder/cudnnBuilderBlockChooser.cpp (117) - Assertion Error in buildMemGraph: 0 (mg.nodes[mg.regionIndices[outp…

TRT engine is created with dynamic shape inputs on Tesla T4, Quadro P1000 with TensorRT version 7.2.3.4 but engine creation fails only on Jetson AGX Xavier (Jetpack 4.5.1 installed).

My questions:

• Can I build TensorRT 7.2.3.4 from source on Jetson AGX Xavier without breaking any other dependencies?
• When (month and year) will the Jetpack for Jetson AGX Xavier be released with updated TensorRT version?

Thank you.

[dusty_nv]

@msripooja

Can I build TensorRT 7.2.3.4 from source on Jetson AGX Xavier without breaking any other dependencies?

Although you can build https://github.com/nvidia/tensorrt from source, that is just the plugins/parsers/examples, and not the core TensorRT libraries. The actual TensorRT libraries will need to be provided from JetPack.

@msripooja

When (month and year) will the Jetpack for Jetson AGX Xavier be released with updated TensorRT version?

The next JetPack with updated TensorRT version is tentatively supposed to be released around the end of July.

[msripooja]

Hi @dusty_nv ,

Can this issue TRT engine created on Tesla T4, Quadro P1000 but engine creation fails on Jetson AGX Xavier be attributed to TensorRT 7.1.3.0 available in Jetpack 4.5.1.

At this point I am not certain if it is the underlying hardware architecture of Jetson AGX Xavier that is causing the error or is it the TensorRT version on Jetson Xavier that is causing the error?

Thank you.

[dusty_nv]

Hi @msripooja, I am unfamiliar with that particular issue, so you may want to continue following up about it in that topic.