Deep Neural Network Energy and Latency Estimation Tool

Overview

This new version of the tool is based on Timeloop/Sparseloop and Accelergy. The former is responsible for searching the possible ways to map a particular workload (i.e., a DNN layer) onto a hardware accelerator design (in this specific case, Eyeriss) and the latter is responsible for generating the cost of hardware activities (in this particular case, based on 45nm technology).

To use the tool, please upload a description of the DNN layer (we provide examples in later sections) to describe the layer shape and density. We will include more detailed descriptions of the input sepcification soon.

Running the Estimation Model

1. Check the "I am not a robot" checkbox and complete the Google reCAPTCHA challenge. Help us prevent spam.
2. Click the "Choose File" button below to choose your text file from your computer.
3. Click the "Run Estimation Model" button below to upload your text file and run the estimation model.

The estimation model might take several minutes to run, please be patient and only click the run button once. Upon successful completion a download button will appear at the top of the page which you can click to download a zip file with the results to your computer, see below for more information on the output. Error messages will be displayed in red text below the buttons if something goes wrong.

Output

The tool generates energy and cycle counts for the Eyeriss architecture in 45nm technology

1. A stats txt file: Records the energy brakdown across the various components in the Eyeriss architecture based on a 45nm technology node. The number of cycles is also reported.
2. A mapping txt file: Describes the best data movement searched by Timeloop in the form of "forloops".

Example Input

Click the links below for examples of the network configuration file:

AlexNet Conv Layer 3

Additional Information

For more information on the architecture runtime modeling and energy estimation methodology, please visit: http://accelergy.mit.edu