ZenDNN library, which includes APIs for basic neural network building blocks optimized for AMD CPU architecture, targets deep learning application and framework developers with the goal of improving deep learning inference performance on AMD CPUs.
ZenDNN v3.3 Highlights Downloads Documentation
- Enabled, tuned, and optimized for inference on AMD 3rd Generation EPYCTM processors
- Integrated with TensorFlow v2.9 and PyTorch v1.11.0
- Features highly optimized primitives for AMD CPUs, targeting a variety of workloads, including computer vision, natural language processing, and recommender systems
This release of ZenDNN will work with AMD optimized models from Unified Inference Frontend (UIF), which are intended to provide significant performance improvement (in the range of 2-5x). More than 50 Xilinx optimized models will be available in this release. For more information on these models, refer to the “UIF Documentation” section.
- Tested with a variety of neural network models across three major workload types:
- Computer Vision: AlexNet, InceptionV3, InceptionV4, GoogLeNet, ResNet50, ResNet152, VGG16, VGG19
- Natural Language Processing: BERT
- Recommender Systems: DLRM, Wide & Deep
- Supported on Ubuntu 18.04, Ubuntu 20.04, RHEL 8.3, SLES15 SP3
Resources and Technical Support
Resources
Documentation
TensorFlow + ZenDNN User Guide
ZenDNN support matrix lists the details of ZenDNN releases and other relevant artifacts.
UIF Documentation
UIF documentation will be available on GitHub (https://github.com/amd/UIF).
This GitHub page describes how to incorporate the AMD optimized models into your workflow and provides guidance on the possible performance uplift.
Installation Tutorials
Technical Support
For support options, refer to Technical Support.
Downloads
TensorFlow and PyTorch are available with different Python versions.
For previous ZenDNN releases, refer to ZenDNN Archive.