Support Matrix

This support matrix provides filterable compatibility information. The software/system and feature explorers cover releases from 10.0.0 EA through 11.4.1. The GPU architecture and precision explorer reports TensorRT 11.4.1.

Software and System Requirements

Use this when you need to know what to install or what versions are required

Engine Portability

• Platform: Serialized engines are not portable across platforms, such as Linux and Windows.
• Version Compatibility: Engines built with the version-compatible flag can run with supported newer TensorRT runtimes, subject to the version-compatibility known issues in the release notes. Two combinations are supported across major versions: TensorRT 10 version-compatible engines can run with a TensorRT 11 runtime, and TensorRT 8 version-compatible engines can run with a TensorRT 10 runtime. No other combination across major versions is supported. For example, a TensorRT 8 engine cannot run with a TensorRT 11 runtime. The general runtime has no separate compatibility preflight API: deserialize during controlled startup, require a non-null engine, and inspect the TensorRT logger before accepting traffic. Refer to the TensorRT 11.4.1 known issues before deployment.
• Hardware Compatibility: Engines built with hardware compatibility mode can run on multiple GPU architectures, depending on the hardware compatibility level used. Without this mode, engines are not portable across different GPU architectures.

Find out what you need to install and run TensorRT. First select your TensorRT version to view software requirements by platform group (CUDA, NCCL, drivers, and more). Optionally select a platform in Step 2 to view compiler versions, Python requirements, compute capabilities, driver requirements, and framework compatibility.

Step 1: Select TensorRT Version -- Select a version -- 11.4.1 (Latest) 11.4.0 11.3.0 11.2.1 11.1.0 11.0.0 10.16.1 10.16.0 10.15.1 10.14.1 10.13.3 10.13.2 10.13.0 10.12.0 10.11.0 10.10.0 10.9.0 10.8.0 10.7.0 10.6.0 10.5.0 10.4.0 10.3.0 10.2.0 10.1.0 10.0.1 10.0.0 EA

Step 2: Select Platform -- Select a platform -- Ubuntu 22.04 x86-64 Ubuntu 24.04 x86-64 Ubuntu 26.04 x86-64 RHEL 8 x86-64 RHEL 9 x86-64 RHEL 10 x86-64 Rocky Linux 8 x86-64 Rocky Linux 9 x86-64 Rocky Linux 10 x86-64 Debian 12 x86-64 SLES 15 x86-64 Windows 10 x64 Windows 11 x64 Windows Server 2022 x64 Windows 11 Arm64 Ubuntu 24.04 SBSA Ubuntu 26.04 SBSA Debian 12 SBSA NVIDIA JetPack AArch64 Ubuntu 24.04 x86-64 DriveOS 7 Linux DriveOS 7 QNX DRIVE Thor (DriveOS 7.2.6+) DRIVE Orin (DriveOS 7.0) DRIVE Orin (DriveOS 7.3) Jetson AGX Orin (JetPack 7.2.2+) Jetson Thor (JetPack 7.2.2+) Jetson Orin/Thor (JetPack 7.x) Jetson Orin (JetPack 6.x) Ubuntu 24.04 SBSA Ubuntu 26.04 SBSA Debian 12 SBSA NVIDIA JetPack AArch64

Show Results Reset

Select a TensorRT version above to view compatibility information. Optionally select a platform to filter results.

GPU Architecture and Precision Support for TensorRT 11.4.1

Use this when you need to know hardware capabilities and precision support

TensorRT 11.4.1 results. Filter by GPU architecture (compute capability) or precision mode to see which hardware supports which precision types in this release. TensorRT 11.4.1 supports NVIDIA hardware with compute capability SM 7.5 or higher. Use the versioned explorers for older releases.

GPU Architecture -- Select compute capability -- All GPU Architectures 12.1 (DGX Spark) 12.1 (Windows on ARM DLA) 12.0 (NVIDIA RTX PRO 6000 Blackwell) 11.0 (DRIVE AGX Thor) 10.7 (NVIDIA Vera Rubin) 10.3 (B300) 10.0 (B200) 9.0 (H100, GH200) 8.9 (L40S) 8.7 (Jetson AGX Orin) 8.6 (A10) 8.0 (A100) 7.5 (T4)

Precision Mode -- Select precision mode -- All Precision Modes TF32 FP32 FP16 FP8 FP4 BF16 INT8 DLA

Show Results Reset

Select filters above to display GPU architecture and precision support.

Feature and Component Support

Use this when you need to know which TensorRT features are available in which versions

Find out which TensorRT features and components are supported in which versions. Filter by TensorRT version and component to see feature availability across releases. Use this to determine if a specific feature is available in your target TensorRT version.

Step 1: Select TensorRT Version -- Select a version -- All Versions 11.4.1 (Latest) 11.4.0 11.3.0 11.2.1 11.1.0 11.0.0 10.16.1 10.16.0 10.15.1 10.14.1 10.13.3 10.13.2 10.13.0 10.12.0 10.11.0 10.10.0 10.9.0 10.8.0 10.7.0 10.6.0 10.5.0 10.4.0 10.3.0 10.2.0 10.1.0 10.0.1 10.0.0 EA

Step 2: Select Component/Feature -- Select a component -- All Components TensorRT Python API ONNX Parser Loops Conditionals Dynamic Shapes DLA (Deep Learning Accelerator) Weight Streaming INT4 Weight Only Quantization FP4 Data Type API Capture and Replay MoE (Mixture of Experts) Multi-Device Inference Global Performance Tuner Green Contexts

Show Results Reset

Select a TensorRT version above to view feature support information. Optionally select a component to filter results.

Footnotes

• [#f1] Built with CUDA Toolkit 12.9. Compatible with CUDA 12.x versions only.
• [#f3] Supported in hardware emulation mode (hardware does not accelerate FP4 linear operations).
• [#f4] Requires CUDA Toolkit 12.8 or newer.
• [#f5] Requires CUDA Toolkit 12.9 or newer.
• [#f6] Python Version Support: Debian/RPM packages support the Python version listed in the table for each platform. Wheel packages (tar/zip) support Python 3.10, 3.11, 3.12, 3.13, and 3.14 across all platforms. Python 3.8 and 3.9 wheels may still be available on some platforms, but those bindings are deprecated in TensorRT 11.x and will be removed in a future release.
• [#f7] For detailed driver compatibility information, refer to the NVIDIA CUDA Release Notes and NVIDIA Driver Downloads.
• [#f8] DLA availability on NVIDIA Orin varies by release. TensorRT 11.4.1 supports DLA on Jetson AGX Orin with NVIDIA JetPack 7.2.2 or later, installed from the Linux SBSA package; the SBSA build does not provide DLA outside NVIDIA JetPack. DLA on NVIDIA JetPack is generally available, and GPU fallback behaves as it did in TensorRT 10.x. These builds use strongly typed networks, so precision is determined by network tensor types instead of the former per-precision builder flags. Jetson Thor has no DLA, and DLA is not supported on NVIDIA DriveOS in TensorRT 11.4.1. TensorRT 11.4.0 supports DLA only on Windows on ARM, as a beta feature, and not for Linux SBSA, NVIDIA JetPack, or NVIDIA DriveOS deployments. DLA is not supported in TensorRT 11.3.0 or in TensorRT 11.3.1 for DriveOS. TensorRT 11.2.1 and earlier support only Orin iGPU.
• [#f9] Requires CUDA Toolkit 13.4 or newer.
• [#f10] Built with CUDA Toolkit 13.4. Compatible with CUDA 13.x versions only.
• [#f11] Built with CUDA Toolkit 13.4 update 1. Debian, RPM, tar, and zip packages require CUDA 13.x. Matching ``tensorrt-cu12`` pip wheels support CUDA 12.x.
• [#f12] Requires CUDA Toolkit 13.4 update 1 or newer.
• [#f13] DLA support on Windows on ARM is in beta in TensorRT 11.4.0 and TensorRT 11.4.1, and covers a restricted set of layers. There is no GPU fallback, so every layer must be DLA-capable, and batch size is 1. These restrictions are specific to Windows on ARM and do not describe DLA on other platforms. DLA engines built for Windows on ARM must be rebuilt once DLA support on that platform reaches general availability, so keep your source models.