DPUCAHX8H for Convolutional Neural Networks

PublishTime：2022-05-23

The Xilinx® Deep Learning Processor Unit (DPU) is a series of soft IP dedicated for convolutional neural networks acceleration. The DPUCAHX8H is a high throughput CNN inference IP for Alveo™ cards with high bandwidth memory (HBM). It runs with a set of efficiently optimized instructions and it can support most convolutional neural networks, such as VGG, ResNet,GoogLeNet, YOLO, SSD, MobileNet, FPN, etc.
●Features:
■ Supports one AXI slave interface for accessing configuration and status registers.
■ Supports one AXI master interface for loading instructions.
■ Supports two AXI master interfaces for loading model parameters.
■ Supports one to five AXI master interfaces for accessing input/output/intermediate featuremap stored in the HBM.
■ Supports all AXI master interfaces with 256-bit width.
■ DPU functionality includes the following:
▲Configurable number of processing engines (PE)
▲Convolution and deconvolution
▲Depthwise convolution
▲Maximum and average pooling
▲ReLU, ReLU6, and Leaky ReLU
▲Data split
▲Concat
▲Elementwise-sum
▲Dilation
▲Data reorganization
▲Fully connected layer
▲Batch normalization

Brand of Product：	Xilinx
Part#：	DPUCAHX8H More Part#
Product Category：	Deep Learning Processor Unit 、 DPU 、 programmable DPU More
Application：	More
Data Type：	User's Guide More
Safety & Environment Standards：	More
Operating Temperature：	Please see the document for details
Qualification Grade：	More
Qualification Approval：	More
Package：
Language：	English Chinese Chinese and English Japanese
Date of Creation：	January 20, 2022
Version Number：	v1.0
Data Encoding：	PG367
Data Size：	561 KB

Data section preview(2pages)

- The full preview is over. If you want to read the whole 29 page document,please Sign in/Register -

Download

+1 Like
Add to Favorites

Recommend

Technical Resources

More>

All reproduced articles on this site are for the purpose of conveying more information and clearly indicate the source. If media or individuals who do not want to be reproduced can contact us, which will be deleted.

Integrated Circuits