What's New
-
We added support to save all tensors in the analysis mode.
-
We added support for using nested templates to construct hierarchical network graphs.
-
We significantly improved the performance of the type-checking process in the Editor.
-
We fixed a bug that prevented PyTorch exports on Linux from succeeding.
-
We removed clamping behavior from the affine layer. It no longer restricts the values of its scale and offset parameters. The options_on parameter has been deprecated; if you wish to hide interactive controls for this layer during analysis, you should set the new include_ui parameter to false.
- We fixed a bug that blocked FP16 inference when fusing 7x7 convolutions with batch normalizations.
NvNeural Changes
-
We added a new analysis layers: Signal Injector.
-
We added a new scalar-only Constant Input layer.
-
We optimized the performance of the BatchNorm layer.
-
We optimized the performance of the Upscale layer.
-
We added support for downscaling and fixed-size scaling to the Upscale layer.
-
The NvRTC wrapper in nvneural::ICudaRuntimeCompiler has been replaced with a stub when type-checking networks from the GUI. Plugins that rely on the ability to execute generated kernel code during initialization or nvneural::ILayer::reshape should call NvRTC directly, but for performance reasons we do not recommend this approach.
-
The forward() function in the exported PyTorch class now takes in keyword-only arguments. You should explicitly name the input paramters while calling the model/function.
-
The INetwork::inferenceSubgraph method now applies queued reshape operations. Queued reshapes are not cleared upon failure and will continue to block inference and inferenceSubgraph calls until they succeed.
Known Issues
-
When you have a high resolution monitor and you set the DPI scaling at larger than 100%, you may see rendering corruptions inside the editor.
-
The DirectML exporter does not currently honor the apply_bias layer parameter during code generation.
-
Command lines containing spaces are not quoted correctly for copy/paste in dialog boxes.
-
Custom layers using PrimaryInfinite inputs should provide no more than one Primary input.
-
The PyTorch exporter does not support connections to secondary layer inputs. Layer weights are defined exclusively using the weights system.
Platform Support
Windows 10: 20H1 or newer
Linux: Ubuntu 18.04 LTS or newer
Recommended Display Driver
You must have a recent NVIDIA display driver installed on your system to run NVIDIA Nsight Deep Learning Designer. The following display drivers are recommended:
Windows: Release 511.23 or newer
Linux: Release 510.39.01 or newer
Notices
Notice
ALL NVIDIA DESIGN SPECIFICATIONS, REFERENCE BOARDS, FILES, DRAWINGS, DIAGNOSTICS, LISTS, AND OTHER DOCUMENTS (TOGETHER AND SEPARATELY, "MATERIALS") ARE BEING PROVIDED "AS IS." NVIDIA MAKES NO WARRANTIES, EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR PURPOSE.
Information furnished is believed to be accurate and reliable. However, NVIDIA Corporation assumes no responsibility for the consequences of use of such information or for any infringement of patents or other rights of third parties that may result from its use. No license is granted by implication of otherwise under any patent rights of NVIDIA Corporation. Specifications mentioned in this publication are subject to change without notice. This publication supersedes and replaces all other information previously supplied. NVIDIA Corporation products are not authorized as critical components in life support devices or systems without express written approval of NVIDIA Corporation.