Description
DLProf release for 20.07, available in the NVIDIA TensorFlow 1.x, TensorFlow 2.x, and PyTorch NGC containers.
Driver Requirements
Release 20.07 is based on NVIDIA CUDA 11.0.167, which requires NVIDIA Driver release 450 or later. However, if you are running on Tesla (for example, T4 or any other Tesla board), you may use NVIDIA driver release 418.xx or 440.30. 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.
New Features
The key features of DLProf v0.13.0 / r20.07 are:
- Released in the TensorFlow 1.x 20.07, TensorFlow 2.x 20.07 and PyTorch 20.07 NGC container.
- Latest DLProf build is based on TensorFlow 1.15.2, TensorBoard 1.15.0, PyTorch 1.6.0, and Nsight Systems 2020.3.2.
- Support for Tensor core detection in the new A100 GPU architecture, new TF32 tensorcores are detected for Tensorflow and Pytorch model profiles.
- Improved PyTorch support:
- Detection of data loading bottlenecks by Expert systems.
- Detection of AMP usage by Expert Systems.
- Improved TensorCore utilization calculation.
- Tensorflow AMP enablement detection and recommendation in Expert Systems.
- Removed all dependencies on additional dynamic libraries.
- Added ability to specify name of model profiled:
- Model name is added to reports.
- Model name is propagated to the TensorBoard DLProf plugin.
Known Issues
- This software is only accessible in the NGC TensorFlow and PyTorch containers.
- This software is only supported for TensorFlow 1.15, and PyTorch 1.6 and TensorBoard 1.15.
- Partial simple mode profiling is supported for TensorFlow 2.
Resolved Issues
- None