Is this page helpful?

TensorRT 11.2.1 Release Notes#

These Release Notes apply to x86 Linux and Windows users, and ARM-based CPU cores for Server Base System Architecture (SBSA) users on Linux. This release includes several fixes from the previous TensorRT releases and additional changes.

Announcements#

Platform dependency upgrades: Updates internal build dependencies for TensorRT 11.2.1. TensorRT 11.2.1 packages are built against CUDA 13.3 update 1 (Debian, RPM, tar, and zip package filenames still use cuda-13.3). Refer to the TensorRT Support Matrix for per-platform CUDA version pinning and to Prerequisites for installer prerequisites.
GridSample 3D support: Extends GridSample from 2D-only to 3D (rank-5 input) with FP32, FP16, and BF16; linear, nearest, and cubic interpolation; align_corners 0/1; and padding modes zeros, border, and reflection.

Key Features and Enhancements#

ONNX DFT operator support: Adds a cuFFT-based plugin for the ONNX DFT operator, including forward and inverse C2C, R2C, and C2R transforms.
PluginV2 to PluginV3 migration sample: Adds a Python sample that demonstrates how to migrate a TensorRT PluginV2 to PluginV3, and the mappings from V2 to V3 plugin methods. Refer to Plugin API Description and the Sample Explorer.
Improved CMake support: Tar and Zip packages now include CMake configuration files under the cmake directory. Import TensorRT with find_package(TensorRT-<Product>), where <Product> is Enterprise, Automotive, RTX, or SafeInference. Prefer this import path over manually wiring include and library paths in CMake projects.

Breaking ABI Changes#

TensorRT 11.0 removed weak typing APIs along with several other deprecated APIs. Refer to the NVIDIA TensorRT Migration Guide for more information.

Breaking Behavioral Changes#

trtexec

The --useCudaGraph, --noDataTransfers, --useSpinWait, and --separateProfileRun flags were enabled by default and deprecated in TensorRT 11.0. Each flag is still accepted but has no effect. For replacement flags and details, refer to the trtexec migration page.
The --stronglyTyped flag has no effect in TensorRT 11.0.0+, since strongly typed networks are now the default. The flag is still accepted for backward compatibility.

Compatibility#

For comprehensive platform compatibility, hardware requirements, and feature availability information, refer to the TensorRT Support Matrix. The support matrix provides detailed information about supported operating systems, CUDA versions, GPU architectures, precision modes, compiler requirements, and ONNX operator support for TensorRT 11.2.1.

Limitations#

DLA is not supported in TensorRT 11.0, 11.1, or 11.2. TensorRT 10.7 was the last release that supported DLA. Refer to Working with DLA for guidance that applies to supported earlier releases.
NVIDIA JetPack is not supported in TensorRT 11.2.1. Jetson deployments must remain on a TensorRT 10.x release supported by their JetPack version.
The high-precision weights used in FP4 double quantization are not refittable.
Version Compatible engine builds from ONNX models with explicit quantization/dequantization (Q/DQ) nodes may fail during engine build if Q/DQ is per-channel scaling for convolution filter.
ISelectLayer must have data inputs (thenInput and elseInput) of the same datatype.
When implementing a custom layer using IPluginV3 plugin class where the custom layer has data-dependent shape (DDS), the size tensors must be of only INT64 type and not INT32 type, as the latter would result in a compilation failure. Related samples have been updated accordingly.

Deprecated API Lifetime#

APIs deprecated in TensorRT 11.2.1 will be retained until July 2027.
APIs deprecated in TensorRT 11.1.0 will be retained until June 2027.
APIs deprecated in TensorRT 11.0.0 will be retained until March 2027.
APIs deprecated in TensorRT 10.16 will be retained until March 2027.
APIs deprecated in TensorRT 10.15 will be retained until January 2027.
APIs deprecated in TensorRT 10.14 will be retained until October 2026.
APIs deprecated in TensorRT 10.13 will be retained until August 2026.

Deprecated and Removed Features#

For a complete list of deprecated C++ APIs, refer to the C++ API Deprecated List.
Windows 10 support is considered deprecated and Windows 10 will no longer be supported by TensorRT after October 2026. Windows 10 has been End-of-Life since October 2025. GeForce driver support for Windows 10 will also be reduced at that time. For more information, refer to GeForce Support Plan for Windows 10.
The TensorRT Python bindings for Python versions 3.8 and 3.9 are deprecated. These Python versions are considered End-of-Life by Python upstream and have not been supported by TensorRT’s samples for multiple releases. These bindings will be removed in a future release.

Fixed Issues#

Fixed execution-context GPU memory usage for some networks where TensorRT 11.1 used more GPU memory than TensorRT 11.0 without a corresponding inference speedup. This addresses the performance issue described in the TensorRT 11.1.0 release notes Known Issues.
Fixed elevated host (CPU) memory consumption during engine build for some ONNX networks where TensorRT 11.1 used substantially more memory than TensorRT 11.0. An increase of approximately 25–30% had been observed on NVIDIA H100 when building certain speech-synthesis models with TF32 enabled. This addresses the build-time memory issue described in the TensorRT 11.1.0 release notes Known Issues.
Fixed engine build failures for explicitly quantized FP8 convolution layers that use large spatial strides (such as vision-transformer patch-embedding layers) on NVIDIA Blackwell platforms. These layers previously failed during engine build with no available tactics. TensorRT 11.2.1 adds fallback support for such FP8 convolutions. This addresses the functional issue described in the TensorRT 11.1.0 release notes Known Issues.
Fixed incorrect all-NaN inference outputs for some explicitly quantized NVFP4 diffusion transformer models on NVIDIA Blackwell platforms. These models previously produced results that disagreed with the reference output. This addresses the functional issue described in the TensorRT 11.1.0 release notes Known Issues.
Fixed invalid memory access and IExecutionContext initialization failures on NVIDIA Blackwell B100 for some networks that use attention patterns. These networks previously failed during initialization with a CUDA driver error reporting CUDA_ERROR_INVALID_HANDLE (“Cannot pass CUkernel handle to this API”). The failure had been observed on B100 with Ubuntu 24.04 and CUDA 13.2.1. This addresses the functional issue described in the TensorRT 11.1.0 release notes Known Issues.
Fixed trtexec argument parsing for commands that specify the deprecated --stronglyTyped flag more than once. All occurrences are now accepted as deprecated no-ops instead of subsequent occurrences causing an [E] Unknown option: --stronglyTyped error. Strongly typed networks remain the default in TensorRT 11.0.0 and later.
Fixed errors reported by the ECCommunicatorAPITests.SetCommunicatorFailsWithoutSupportedLayer and ECCommunicatorAPITests.SetCommunicatorSucceedsWithDistCollective runtime tests under valgrind (host) and compute-sanitizer memcheck (GPU). The tests no longer report host-side memory leaks from NCCL allocations during ncclCommInitRank and ncclCommSplit on H100 and B100 or GPU-side NCCL kernel-probing errors on H100 (SM 90).
Fixed GPU time regressions for some strongly typed FP16 networks on RTX PRO 6000 Blackwell Max-Q (SM 120, Arm64) compared with TensorRT 10.16.1. The regressions, previously observed at up to approximately 45% on individual networks, affected fused-graph kernels, particularly fused Conv/BiasAdd patterns and mean-reduction kernels.
Added Python 3.13 and 3.14 support to the Python samples on platforms listed in the TensorRT Support Matrix. Windows Python 3.14 support remains preliminary. On Windows, applications that use NumPy or other native extensions must install compatible package builds; incompatible or experimental builds can crash during import before TensorRT is loaded. Refer to Method 1: Python Package Index (pip) for Python wheel installation instructions.

Known Issues#

Functional

On CUDA versions prior to 13.2, the compute-sanitizer initcheck tool may flag Uninitialized __global__ memory read errors when running TensorRT applications on NVIDIA Hopper GPUs. These errors are false positives and can be safely ignored. To suppress them, upgrade to CUDA 13.2 or later.
Running TensorRT applications under Valgrind memcheck on CUDA 13.3 may report host-side memory leaks attributed to NVRTC components used during engine build and runtime compilation. These reports are not known to affect normal inference behavior. This will be addressed in a future CUDA or TensorRT release.
Running TensorRT inference under compute-sanitizer memcheck may report device memory errors during execution or engine teardown. These reports are not known to affect normal inference behavior when sanitizers are not enabled. This will be addressed in a future release.
inplace_add mini-sample of the quickly_deployable_plugins Python sample may produce incorrect outputs on Windows.
TensorRT may exit if inputs with invalid values are provided to the RoiAlign plugin (ROIAlign_TRT), especially if there is inconsistency in the indices specified in the batch_indices input and the actual batch size used.
On Windows, the TensorRT 11.0.0 runtime can fail to deserialize a Version Compatible engine that was built with TensorRT 10.1.0. Deserialization fails inside the runtime dispatch VC plugin loader when loading nvinfer_vc_plugin.dll (Windows ERROR_MOD_NOT_FOUND / error 126), surfaces as TensorRT error code 6 (API Usage Error), and leaves the returned ICudaEngine pointer null. Later TensorRT 10.x builders are not affected; only engines produced by TensorRT 10.1.0 have been observed to trigger this failure. As a workaround, rebuild the affected VC engines with a newer TensorRT 10.x release (10.2 or later) before deserializing them with the TensorRT 11.0.0 runtime on Windows. This will be addressed in a future release.
On Windows, Version Compatible engines built with TensorRT 10.1 through 10.4 that use the RoiAlign plugin (ROIAlign_TRT) cannot be deserialized with the TensorRT 11.x runtime due to a bug in how those engines were produced. The bug was fixed in TensorRT 10.5. Engines built with TensorRT 10.5 or later are expected to deserialize correctly. As a workaround, rebuild affected engines with TensorRT 10.5 or later before loading them with TensorRT 11.x.
Inspecting very large engines with the Engine Inspector may fail on systems with limited host memory. The inspection process can exceed available memory and be terminated by the operating system without a TensorRT error message. This has been observed with very large diffusion-model engines. As a workaround, run engine inspection on a system with sufficient host memory. This will be addressed in a future release.
Default-constructed tensorrt.PluginField().data returns an empty NumPy ndarray rather than a null-like value. With NumPy 2.x (for example, NumPy 2.2.3), expressions such as assert not pfield.data or if not pfield.data raise ValueError: The truth value of an empty array is ambiguous. As a workaround, check emptiness explicitly (for example, pfield.data.size == 0) instead of using the array in a boolean context. This will be addressed in a future release.
Explicit quantization with NVIDIA TensorRT Model Optimizer ONNX quantization may fail during calibration with ValueError: Too many bins for data range. Cannot create 128 finite-sized bins. The failure occurs in the ModelOpt ONNX calibration histogram path (ONNX Runtime calibration calling np.histogram), not during TensorRT engine build or inference. It has been observed for some explicitly quantized ONNX models, including Blackwell-targeted networks and quantized ResNet-50 transformer configurations, on SBSA platforms with CUDA 13.3.
Building TensorRT OSS samples with Visual Studio / MSBuild on Windows may fail during the CMake configure step when CUDA_HOME and CUDA_PATH point to different CUDA Toolkit installations. The failure is environment-wide rather than sample-specific and can affect multiple sample targets and trtexec. As a workaround, ensure CUDA_HOME and CUDA_PATH refer to the same CUDA Toolkit, or unset one of the variables so CMake discovers a single consistent toolkit. This will be addressed in a future release.
Quickly Deployable Plugins (QDP) registered with @trtp.register may raise TypeError: issubclass() arg 1 must be a class during plugin validation on Python 3.14. The failure occurs in the Python plugin annotation checks used by the QDP path and has been observed for shape and serialization QDP tests in TensorRT 11.2.x. As a workaround, use Python 3.13 or earlier for QDP plugin development and testing. This will be addressed in a future release.

Performance

A non-zero tilingOptimizationLevel might introduce engine build failures for some networks on L4 GPUs.
The kREFIT and kREFIT_IDENTICAL have performance regressions compared with non-refit engines where convolution layers are present within a branch or loop, and the precision is FP16/INT8. This issue will be addressed in future releases.
Some transformer models that use multi-head attention, including vision-transformer architectures, may run slower on GPU in TensorRT 11.x than in TensorRT 10.16 due to additional layout-related overhead during inference. This will be addressed in a future release.
Some Maxine AR Facial Landmark Detection models (landmark_quality_68_dynamic_strongtype_fp16 and landmark_quality_126_dynamic_strongtype_fp16) may exhibit approximately 7–9% higher GPU inference time in TensorRT 11.2.x than in TensorRT 11.1.0 on RTX PRO 6000 Blackwell Max-Q. It is for functional fixing, absolute impact is typically about 0.1 ms on workloads of roughly 1.7–1.9 ms.