Migrating from TensorRT 8.x to 10.x#

This section explains how to migrate from NVIDIA TensorRT 8.x to 10.x.

If you are unfamiliar with these changes, refer to our sample code for clarification.

See also

C++ API Reference

Full C++ API documentation.

Python API Reference

Full Python API documentation.

Version Compatibility

Engine compatibility across TensorRT versions and hardware.

Troubleshooting

Common errors and debugging guidance.

How to Use This Guide#

Select the migration surface that matches what you ship:

Python Integration

Migrate Python inference and engine-build flows, and review Python APIs added and removed in 10.x.

C++ Integration

Migrate C++ inference and engine-build flows, including 64-bit dimension updates, and review C++ APIs and plugins removed with replacements.

trtexec Command-Line Usage

Migrate trtexec command lines, and review removed or deprecated flag replacements.

Safety Runtime

Migrate safety runtime code paths, and review removed safety C++ APIs with replacements.

Suggested Reading Order#

  1. Start with Python Integration or C++ Integration for application code paths (read both only if you maintain both stacks).

  2. Add trtexec Command-Line Usage when your workflow depends on CLI flags or scripts.

  3. Read Safety Runtime when you target the safety runtime instead of the standard runtime.

    Attention

    TensorRT 10.x safety proxy runtime support on QNX Standard (without QNX Safety support) is available in limited, alpha-quality capacity in this release. Full functional safety runtime support, including QNX Safety, is planned for a future sideband drop on top of NVIDIA DriveOS 7.2.5.