TensorFlow Release 21.03
The NVIDIA container image of TensorFlow, release 21.03, is available on NGC.
Contents of the TensorFlow container
This container image includes the complete source of the NVIDIA version of TensorFlow in /opt/tensorflow
. It is pre-built and installed as a system Python module.
To achieve optimum TensorFlow performance, for image based training, the container includes a sample script that demonstrates efficient training of convolutional neural networks (CNNs). The sample script may need to be modified to fit your application. The container also includes the following:
- Ubuntu 20.04
Note:
Container image
21.03-tf1-py3
and21.03-tf2-py3
contains Python 3.8 - NVIDIA CUDA 11.2.1 including cuBLAS 11.4.1.
- NVIDIA cuDNN 8.1.0
- NVIDIA NCCL 2.8.4 (optimized for NVLink™ )
- Horovod 0.21.3
- OpenMPI 4.0.5
- TensorBoard
-
21.03-tf1-py3
includes version 1.15.0+nv21.3 -
21.03-tf2-py3
includes version TensorBoard 2.4.1
-
- MLNX_OFED 5.1
- OpenSeq2Seq at commit 8f040a49
- Included only in
21.03-tf1-py3
- Included only in
- TensorRT 7.2.2.3
- DALI 0.31.0
- DLProf 1.0.0
- Included only in
21.03-tf1-py3
- Included only in
- Nsight Compute 2020.3.0.18
- Nsight Systems 2020.4.3.7
- XLA-Lite
- Tensor Core optimized examples: (Included only in
21.03-tf1-py3)
- JupyterLab 1.2.14 including Jupyter-TensorBoard
Driver Requirements
Release 21.03 is based on NVIDIA CUDA 11.2.1, which requires NVIDIA Driver release 460.32.03 or later. However, if you are running on Data Center GPUs (formerly Tesla), for example, T4, you may use NVIDIA driver release 418.40 (or later R418), 440.33 (or later R440), 450.51(or later R450). The CUDA driver's compatibility package only supports particular drivers. For a complete list of supported drivers, see the CUDA Application Compatibility topic. For more information, see CUDA Compatibility and Upgrades and NVIDIA CUDA and Drivers Support.
GPU Requirements
Release 21.03 supports CUDA compute capability 6.0 and higher. This corresponds to GPUs in the NVIDIA Pascal, Volta, Turing, and Ampere Architecture GPU families. Specifically, for a list of GPUs that this compute capability corresponds to, see CUDA GPUs. For additional support details, see Deep Learning Frameworks Support Matrix.
Key Features and Enhancements
This TensorFlow release includes the following key features and enhancements.
- TensorFlow container images version 21.03 are based on Tensorflow 1.15.5 and 2.4.0
- The latest version of NVIDIA CUDA 11.2.1 including cuBLAS 11.4.1.1026
- The latest version of NVIDIA cuDNN 8.1.1
- The latest version of Horovod 0.21.3
- The latest version of TensorBoard
-
21.03-tf1-py3
includes version 1.15.0+nv21.3 -
21.03-tf2-py3
includes version TensorBoard 2.4.1
-
- The latest version of TensorRT 7.2.2.3
- The latest version of DALI 0.31.0
- The latest version of DLProf 1.0.0
- Ubuntu 20.04 with February 2021 updates
- NVTX profiling annotation ranges more accurately report the execution of asynchronous operations. Note that when profiling NVTX ranges must now be explicitly enabled by setting the environment variable TF_ENABLE_NVTX_RANGES=1.
- The CUDNN backend API is now used for convolutional ops. This provides a significant performance benefit by reducing CPU overheads of convolutions.
-
The fused Conv+Bias+Relu op regression in CUDNN has been fixed and this op has been re-enabled in both XLA and the TF grappler optimizers. This improves performance particularly for inference in convolutional models.
-
Bugs relating to auto-graph in TensorFlow 1.15 with Python 3.8 were fixed.
Announcements
- Python 2.7 is no longer supported in this TensorFlow container release.
- The TF_ENABLE_AUTO_MIXED_PRECISION environment variables are no longer supported in the tf2 container because it is not possible to automatically enable loss scaling in many cases in the tf 2.x API. Instead tf.train.experimental.enable_mixed_precision_graph_rewrite() should be used to enable AMP.
- Deep learning framework containers 19.11 and later include experimental support for Singularity v3.0.
NVIDIA TensorFlow Container Versions
The following table shows what versions of Ubuntu, CUDA, TensorFlow, and TensorRT are supported in each of the NVIDIA containers for TensorFlow. For older container versions, refer to the Frameworks Support Matrix.Container Version | Ubuntu | CUDA Toolkit | TensorFlow | TensorRT |
---|---|---|---|---|
21.03 | 20.04 | NVIDIA CUDA 11.2.1 | TensorRT 7.2.2.3 | |
21.02 | NVIDIA CUDA 11.2.0 | TensorRT 7.2.2.3+cuda11.1.0.024 | ||
20.12 | NVIDIA CUDA 11.1.1 | TensorRT 7.2.2 | ||
20.11 | 18.04 |
NVIDIA CUDA 11.1.0 | TensorRT 7.2.1 | |
20.10 | ||||
20.09 | NVIDIA CUDA 11.0.3 | TensorRT 7.1.3 | ||
20.08 | ||||
20.07 | NVIDIA CUDA 11.0.194 | |||
20.06 | NVIDIA CUDA 11.0.167 | |||
20.03 | NVIDIA CUDA 10.2.89 | TensorRT 7.0.0 | ||
TensorRT 6.0.1 | ||||
19.10 | NVIDIA CUDA 10.1.243 | 1.14.0 | ||
19.09 | ||||
19.08 | TensorRT 5.1.5 |
Tensor Core Examples
The tensor core examples provided in GitHub focus on achieving the best performance and convergence by using the latest deep learning example networks and model scripts for training. Each example model trains with mixed precision Tensor Cores on Volta, therefore you can get results much faster than training without tensor cores. This model is tested against each NGC monthly container release to ensure consistent accuracy and performance over time. This container includes the following tensor core examples.
- U-Net Medical model. The U-Net model is a convolutional neural network for 2D image segmentation. This repository contains a U-Net implementation as described in the paper U-Net: Convolutional Networks for Biomedical Image Segmentation, without any alteration. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- SSD320 v1.2 model. The SSD320 v1.2 model is based on the SSD: Single Shot MultiBox Detector paper, which describes an SSD as “a method for detecting objects in images using a single deep neural network”. Our implementation is based on the existing model from the TensorFlow models repository. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- Neural Collaborative Filtering (NCF) model. The NCF model is a neural network that provides collaborative filtering based on implicit feedback, specifically, it provides product recommendations based on user and item interactions. The training data for this model should contain a sequence of user ID, item ID pairs indicating that the specified user has interacted with, for example, was given a rating to or clicked on, the specified item. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- BERT model. BERT, or Bidirectional Encoder Representations from Transformers, is a new method of pre-training language representations which obtains state-of-the-art results on a wide array of Natural Language Processing (NLP) tasks. This model is based on BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding paper. NVIDIA's BERT is an optimized version of Google's official implementation, leveraging mixed precision arithmetic and Tensor Cores on V100 GPUS for faster training times while maintaining target accuracy. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- U-Net Industrial Defect Segmentation model. This U-Net model is adapted from the original version of the U-Net model which is a convolutional auto-encoder for 2D image segmentation. U-Net was first introduced by Olaf Ronneberger, Philip Fischer, and Thomas Brox in the paper: U-Net: Convolutional Networks for Biomedical Image Segmentation. This work proposes a modified version of U-Net, called TinyUNet which performs efficiently and with very high accuracy on the industrial anomaly dataset DAGM2007. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- GNMT v2 model. The GNMT v2 model is similar to the one discussed in the Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation paper. The most important difference between the two models is in the attention mechanism. In our model, the output from the first LSTM layer of the decoder goes into the attention module, then the re-weighted context is concatenated with inputs to all subsequent LSTM layers in the decoder at the current timestep. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
- ResNet-50 v1.5 model. The ResNet-50 v1.5 model is a modified version of the original ResNet-50 v1 model. The difference between v1 and v1.5 is in the bottleneck blocks which requires downsampling, for example, v1 has stride = 2 in the first 1x1 convolution, whereas v1.5 has stride = 2 in the 3x3 convolution. The following features were implemented in this model; data-parallel multi-GPU training with Horovod, Tensor Cores (mixed precision) training, and static loss scaling for Tensor Cores (mixed precision) training. This model script is available on GitHub as well as NVIDIA GPU Cloud (NGC).
Known Issues
If you encounter functional or performance issues when XLA is enabled, please refer to the XLA Best Practices document. It offers pointers on how to diagnose symptoms and possibly address them.
- Using XLA together with Horovod to parallelize training on a single node can result in out-of-memory errors. A workaround is to execute the job as follows. This will be fixed in a future release.
XLA_FLAGS=--xla_multiheap_size_constraint_per_heap=2000000000 TF_NUM_INTEROP_THREADS=1 horovodrun -np 8 bash -c 'CUDA_VISIBLE_DEVICES=$OMPI_COMM_WORLD_LOCAL_RANK python ...'
- There is a known performance regression of up to 30% when training SSD models with fp32 data type on T4 GPUs. This will be addressed in a future release.
- There is a known issue where attempting to convert some models using TF-TRT produces an error "Failed to import metagraph". This issue is still under investigation and will be resolved in a future release.
- Training the UNET3D models with a batch size of 1 can result in OOM (Out-Of-Memory) in the TensorFlow 1 container. This is caused by the map_and_batch_fusion optimizer from using the tf.datasets. One workaround solution is to add:
if self._batch_size == 1: options = dataset.options() options.experimental_optimization.map_and_batch_fusion = False dataset = dataset.with_options(options)