TensorRT Release Properties

The following table describes the release properties and software versions.

TensorRT Release Properties

Property	Linux AArch64
Supported NVIDIA CUDA versions	12.8
TensorRT Python API	Yes
NvOnnxParser	Yes

Serialized engines are not generally portable across platforms or TensorRT versions. In the standard runtime, version numbers must match (in major, minor, patch, and build) for the previously generated serialized engine to be compatible.

Hardware and Precision

The following table lists NVIDIA hardware and the precision modes each hardware supports. For standard runtime, TensorRT supports SM 10.1.

For more information, refer to the FAQ section in the NVIDIA TensorRT 10.10.10 Developer Guide for DRIVE OS.

Hardware and Precision Support for TensorRT

CUDA Compute Capability	Example Device	TF32	FP32	FP16	FP8	BF16	INT8	FP16 Tensor Cores	INT8 Tensor Cores	DLA
10.1	NVIDIA Thor	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	No

Software Versions per Platform

Software Versions per Platform for TensorRT

Platform	Compiler Version	Python Version
Ubuntu 24.04 AArch64	gcc 13.2.0	3.12

Compatibility

TensorRT 10.10.10 has been tested with the following:

• CUDA 12.8

• TensorFlow 2.13.1

• PyTorch >= 2.0

• ONNX 1.16.0 and opset 20