Abstract

These support matrices provide a look into the supported platforms, features, and hardware capabilities of the TensorRT 6.0.1 APIs, parsers, and layers.

For previously released TensorRT documentation, see TensorRT Archives.

1. Features For Platforms And Software

Table 1. List of supported features per platform.
  Linux x86-64 Windows x64 Linux ppc64le Linux AArch64 QNX AArch64
Supported CUDA versions 10.2
Supported cuDNN versions 7.6.5 7.6.5 7.6.5 7.6.5 7.6.5
TensorRT Python API Yes No Yes Yes1 No
NvUffParser Yes Yes Yes Yes Yes
NvOnnxParser Yes Yes Yes Yes Yes
Note: Serialized engines are not portable across platforms or TensorRT versions.

2. Layers And Features

Table 2. List of supported features per TensorRT layer.
Layer Dimensions of input tensor Dimensions of output tensor Does the operation apply to only the innermost 3 dimensions? Supports broadcast (see Note 1) Supports broadcast across batch (see Note 2)
IActivationLayer 0-7 dimensions 0-7 dimensions No No No
IConcatenationLayer 1-7 dimensions 1-7 dimensions No No No
IConstantLayer 0-7 dimensions 0-7 dimensions No No Always
IConvolutionLayer > 2D Convolution 3 or more dimensions 3 or more dimensions Yes No No
IConvolutionLayer > 3D Convolution 4 or more dimensions 4 or more dimensions No No No
IDeconvolutionLayer > 2D Deconvolution 3 or more dimensions 3 or more dimensions Yes No No
IDeconvolutionLayer > 3D Deconvolution 4 or more dimensions 4 or more dimensions No No No
IElementWiseLayer 0-7 dimensions 0-7 dimensions No Yes Yes
IFullyConnectedLayer 3 or more dimensions 3 or more dimensions Yes No No
IGatherLayer
  • Input1: 1-7 dimensions
  • Input2: 0-7 dimensions
0-7 dimensions No No Yes
IIdentityLayer 0-7 dimensions 0-7 dimensions No No No
ILRNLayer 3 or more dimensions 3 or more dimensions Yes No No
IMatrixMultiplyLayer 2 or more dimensions 2 or more dimensions No Yes Yes
IPaddingLayer 3 or more dimensions 3 or more dimensions Yes No No
IPluginLayer User defined User defined User defined User defined User defined
IPluginV2Layer User defined User defined User defined User defined User defined
IPoolingLayer > 2D Pooling 3 or more dimensions 3 or more dimensions Yes Yes Yes
IPoolingLayer > 3D Pooling 4 or more dimensions 4 or more dimensions No Yes Yes
IRaggedSoftMaxLayer
  • Input: 2 dimensions
  • Bounds: 2 dimensions
2 or more dimensions No No Yes
IReduceLayer 1-7 dimensions 0-7 dimensions No No No
IResizeLayer 1-7 dimensions 1-7 dimensions No No No
IRNNLayer 3 dimensions 3 dimensions No No No
IRNNv2Layer
  • Data/Hidden/Cell: 2 or more dimensions
  • Seqlen: 0 or more dimensions
Data/Hidden/Cell: 2 or more dimensions No No No
IScaleLayer 3 or more dimensions 3 or more dimensions Yes No No
IShapeLayer 1 or more dimensions 1 dimension No No No
IShuffleLayer 0-7 dimensions 0-7 dimensions No No No
ISliceLayer 1-7 dimensions 1-7 dimensions No No Yes
ISoftMaxLayer 1-7 dimensions 1-7 dimensions No No Yes
ITopKLayer 1-7 dimensions
  • Output1: 1-7 dimensions
  • Output2: 1-7 dimensions
Yes No Yes
IUnaryLayer 0-7 dimensions 0-7 dimensions No No No
Note:
  1. Indicates support for broadcast in this layer. This layer allows its two input tensors to be of dimensions [1, 5, 4, 3] and [1, 5, 1, 1], and its output out be [1, 5, 4, 3]. Note: The second input tensor has been broadcast in the innermost 2 dimensions.
  2. Indicates support for broadcast across the batch dimension.
For more information about each of the TensorRT layers, see TensorRT Layers.

3. Layers And Precision

The following table lists the TensorRT layers and the precision modes that each layer supports. It also lists the ability of the layer to run on Deep Learning Accelerator (DLA). For more information about additional constraints, see DLA Supported Layers.

For more information about each of the TensorRT layers, see TensorRT Layers. To view a list of the specific attributes that are supported by each layer, refer to the TensorRT API documentation.

Table 3. List of supported precision mode per TensorRT layer.
Layer FP32 FP16 INT8 INT32 DLA FP16 DLA INT8
IActivationLayer Yes Yes Yes No Yes2 Yes3
IConcatenationLayer Yes Yes Yes Yes Yes4 Yes3
IConstantLayer Yes Yes Yes Yes No No
IConvolutionLayer > 2D Convolution Yes Yes Yes No Yes Yes5
IConvolutionLayer > 3D Convolution Yes Yes No No No No
IDeconvolutionLayer > 2D Deconvolution Yes Yes Yes No Yes Yes6
IDeconvolutionLayer > 3D Deconvolution Yes Yes No No No No
IElementWiseLayer Yes Yes No Yes Yes7 Yes8
IFullyConnectedLayer Yes Yes Yes No Yes Yes
IGatherLayer Yes Yes No Yes No No
IIdentityLayer Yes Yes Yes Yes No No
IPluginV2Layer Yes Yes Yes No No No
ILRNLayer Yes Yes Yes No Yes No
IMatrixMultiplyLayer Yes Yes No No No No
IPaddingLayer Yes Yes Yes No No No
IPluginLayer Yes Yes No No No No
IPoolingLayer > 2D Pooling Yes Yes Yes No Yes9 Yes8
IPoolingLayer > 3D Pooling Yes Yes No No No No
IRaggedSoftMaxLayer Yes No No No No No
IReduceLayer Yes Yes No No No No
IResizeLayer Yes Yes No No No No
IRNNLayer Yes Yes No No No No
IRNNv2Layer Yes Yes No No No No
IScaleLayer Yes Yes Yes No Yes10 Yes9
IShapeLayer11 Yes Yes Yes Yes No No
IShuffleLayer Yes Yes Yes Yes No No
ISliceLayer Yes Yes No12 Yes No No
ISoftMaxLayer Yes Yes No No No No
ITopKLayer Yes Yes No No No No
IUnaryLayer Yes Yes No No No No
Note: DLA with FP16/INT8 precision with some restrictions on layer parameters.

4. Hardware And Precision

The following table lists NVIDIA hardware and which precision modes each hardware supports. It also lists availability of Deep Learning Accelerator (DLA) on these hardware. TensorRT supports all NVIDIA hardware with capability SM 3.0 or higher.
Table 4. List of supported precision mode per hardware.
CUDA Compute Capability Example Device FP32 FP16 INT8 FP16 Tensor Cores INT8 Tensor Cores DLA
7.5 Tesla T4 Yes Yes Yes Yes Yes No
7.2 Jetson AGX Xavier Yes Yes Yes Yes Yes Yes
7.0 Tesla V100 Yes Yes Yes Yes No No
6.2 Jetson TX2 Yes Yes No No No No
6.1 Tesla P4 Yes No Yes No No No
6.0 Tesla P100 Yes Yes No No No No
5.3 Jetson TX1 Yes Yes No No No No
5.2 Tesla M4 Yes No No No No No
5.0 Quadro K2200 Yes No No No No No
3.7 Tesla K80 Yes No No No No No
3.5 Tesla K40 Yes No No No No No
3.0 Tesla K10 Yes No No No No No

5. Software Versions Per Platform

Table 5. List of supported platforms per software version.
  Compiler version Python version
Ubuntu 14.04 x86-64 gcc 4.8.4 2.7, 3.4
Ubuntu 16.04 x86-64 gcc 5.4.0 2.7, 3.5
Ubuntu 18.04 x86-64 gcc 7.4.0 2.7, 3.6
CentOS 7.5 x86-64 gcc 4.8.5 2.7, 3.6
Windows 10 x64
CUDA 10.0, 10.1
MSVC 2017u5
CUDA 9.0
MSVC 2017u3
 
Ubuntu 18.04 ppc64le gcc 7.4.0 2.7, 3.6
CentOS 7.5 ppc64le gcc 4.8.5 2.7, 3.6
Ubuntu 18.04 AArch64 2.7, 3.6
QNX AArch64 gcc 5.4.0  

6. Supported Ops

The following lists describe the operations that are supported in a Caffe or TensorFlow framework and in the ONNX TensorRT parser:

Caffe

These are the operations that are supported in a Caffe framework:
  • BatchNormalization
  • BNLL
  • Clip13
  • Concatenation
  • Convolution
  • Crop
  • Deconvolution
  • Dropout
  • ElementWise
  • ELU
  • InnerProduct
  • Input
  • LeakyReLU
  • LRN
  • Permute
  • Pooling
  • Power
  • Reduction
  • ReLU, TanH, and Sigmoid
  • Reshape
  • SoftMax
  • Scale

TensorFlow

These are the operations that are supported in a TensorFlow framework:
  • Add, Sub, Mul, Div, Minimum and Maximum
  • ArgMax
  • ArgMin
  • AvgPool
  • BiasAdd
  • Clip
  • ConcatV2
  • Const
  • Conv2D
  • ConvTranspose2D
  • DepthwiseConv2dNative
  • Elu
  • ExpandDims
  • FusedBatchNorm
  • Identity
  • LeakyReLU
  • MaxPool
  • Mean
  • Negative, Abs, Sqrt, Recip, Rsqrt, Pow, Exp and Log
  • Pad is supported if followed by one of these TensorFlow layers: Conv2D, DepthwiseConv2dNative, MaxPool, and AvgPool.
  • Placeholder
  • ReLU, TanH, and Sigmoid
  • Relu6
  • Reshape
  • ResizeBilinear, ResizeNearestNeighbor
  • Sin, Cos, Tan, Asin, Acos, Atan, Sinh, Cosh, Asinh, Acosh, Atanh, Ceil and Floor
  • Selu
  • Slice
  • SoftMax
    Note: If the input to a TensorFlow SoftMax op is not NHWC, TensorFlow will automatically insert a transpose layer with a non-constant permutation, causing the UFF converter to fail. It is therefore advisable to manually transpose SoftMax inputs to NHWC using a constant permutation.
  • Softplus
  • Softsign
  • Transpose

ONNX

Since the ONNX parser is an open source project, the most up-to-date information regarding the supported operations can be found in GitHub: ONNX TensorRT.
These are the operations that are supported in the ONNX framework:
  • Abs
  • Add
  • ArgMax
  • ArgMin
  • AveragePool
  • BatchNormalization
  • Cast
  • Ceil
  • Clip
  • Concat
  • Constant
  • Conv
  • ConvTranspose
  • DepthToSpace
  • Div
  • Dropout
  • Elu
  • Exp
  • Flatten
  • Floor
  • Gather
  • Gemm
  • GlobalAveragePool
  • GlobalMaxPool
  • HardSigmoid
  • Identity
  • ImageScaler
  • InstanceNormalization
  • LRN
  • LeakyRelU
  • Log
  • LogSoftmax
  • MatMul
  • Max
  • MaxPool
  • Mean
  • Min
  • Mul
  • Neg
  • Pad
  • ParametricSoftplus
  • Pow
  • Reciprocal
  • ReduceL1
  • ReduceL2
  • ReduceLogSum
  • ReduceLogSumExp
  • ReduceMax
  • ReduceMean
  • ReduceMin
  • ReduceProd
  • ReduceSum
  • ReduceSumSquare
  • Relu
  • Reshape
  • Resize
  • ScaledTanh
  • Selu
  • Shape
  • Sigmoid
  • Sin, Cos, Tan, Asin, Acos, Atan, Sinh, Cosh, Asinh, Acosh, and Atanh
  • Size
  • Slice
  • Softmax
  • Softplus
  • Softsign
  • SpaceToDepth
  • Split
  • Squeeze
  • Sub
  • Sum
  • Tanh
  • ThresholdedRelu
  • TopK
  • Transpose
  • Unsqueeze
  • Upsample

Notices

Notice

THE INFORMATION IN THIS GUIDE AND ALL OTHER INFORMATION CONTAINED IN NVIDIA DOCUMENTATION REFERENCED IN THIS GUIDE IS PROVIDED “AS IS.” NVIDIA MAKES NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE INFORMATION FOR THE PRODUCT, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE. Notwithstanding any damages that customer might incur for any reason whatsoever, NVIDIA’s aggregate and cumulative liability towards customer for the product described in this guide shall be limited in accordance with the NVIDIA terms and conditions of sale for the product.

THE NVIDIA PRODUCT DESCRIBED IN THIS GUIDE IS NOT FAULT TOLERANT AND IS NOT DESIGNED, MANUFACTURED OR INTENDED FOR USE IN CONNECTION WITH THE DESIGN, CONSTRUCTION, MAINTENANCE, AND/OR OPERATION OF ANY SYSTEM WHERE THE USE OR A FAILURE OF SUCH SYSTEM COULD RESULT IN A SITUATION THAT THREATENS THE SAFETY OF HUMAN LIFE OR SEVERE PHYSICAL HARM OR PROPERTY DAMAGE (INCLUDING, FOR EXAMPLE, USE IN CONNECTION WITH ANY NUCLEAR, AVIONICS, LIFE SUPPORT OR OTHER LIFE CRITICAL APPLICATION). NVIDIA EXPRESSLY DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY OF FITNESS FOR SUCH HIGH RISK USES. NVIDIA SHALL NOT BE LIABLE TO CUSTOMER OR ANY THIRD PARTY, IN WHOLE OR IN PART, FOR ANY CLAIMS OR DAMAGES ARISING FROM SUCH HIGH RISK USES.

NVIDIA makes no representation or warranty that the product described in this guide will be suitable for any specified use without further testing or modification. Testing of all parameters of each product is not necessarily performed by NVIDIA. It is customer’s sole responsibility to ensure the product is suitable and fit for the application planned by customer and to do the necessary testing for the application in order to avoid a default of the application or the product. Weaknesses in customer’s product designs may affect the quality and reliability of the NVIDIA product and may result in additional or different conditions and/or requirements beyond those contained in this guide. NVIDIA does not accept any liability related to any default, damage, costs or problem which may be based on or attributable to: (i) the use of the NVIDIA product in any manner that is contrary to this guide, or (ii) customer product designs.

Other than the right for customer to use the information in this guide with the product, no other license, either expressed or implied, is hereby granted by NVIDIA under this guide. Reproduction of information in this guide is permissible only if reproduction is approved by NVIDIA in writing, is reproduced without alteration, and is accompanied by all associated conditions, limitations, and notices.

Trademarks

NVIDIA, the NVIDIA logo, and cuBLAS, CUDA, cuDNN, DALI, DIGITS, DGX, DGX-1, DGX-2, DGX Station, DLProf, Jetson, Kepler, Maxwell, NCCL, Nsight Compute, Nsight Systems, NvCaffe, PerfWorks, Pascal, SDK Manager, Tegra, TensorRT, TensorRT Inference Server, Tesla, TF-TRT, and Volta are trademarks and/or registered trademarks of NVIDIA Corporation in the United States and other countries. Other company and product names may be trademarks of the respective companies with which they are associated.

1 Python is not supported on automotive platforms.
2 Partial support. Yes for ReLU, sigmoid and TanH activation types only.
3 Partial support. Yes for ReLU activation type only.
4 Partial support. Yes for concatenation across C dimension only.
5 Partial support. Yes for ungrouped convolutions and No for grouped.
6 Partial support. Yes for ungrouped deconvolutions and No for grouped.
7 Partial support. Yes for sum, sub, prod, min and max elementwise operations only.
8 Partial support. Yes for sum elementwise operation only.
9 Partial support. Yes for max and average pooling type only.
10 Partial support. DLA does not support power on scale layer.
11 Output is always INT32.
12 Partial support. Yes for unstrided Slice and No for strided.
13 When using the Clip operation, Caffe users must serialize their layers using ditcaffe.pb.h instead of caffe.pb.h in order to import the layer into TensorRT.