SCALE-Sim Tutorial - ASPLOS 2021

Date: April 16, 2021



SCALE-SIM is a cycle-accurate CNN accelerator simulator that provides timing, power/energy, memory bandwidth and memory access trace results for a specified accelerator configuration and neural network architecture. It is based on the systolic array architecture, used in various accelerators like Google’s TPU, Xilinx XDNN etc.

SCALE-SIM enables research into DNN accelerator architectures and is also suitable for system-level studies. Designing efficient DNN accelerator is a difficult problem which requires searching in an intricate trade-off space with large number of architectural parameters. Moreover, recent DNN workloads are increasingly becoming memory bound due to increase in model sizes. A simulation infrastructure like SCALE-Sim which can provide cycle accurate estimates of performance, memory accesses, and other design metrics is therefore a vital tool to enable fast and reliable design cycles. Unlike related infrastructure, which rely on analytical models to estimate the performance and operating cost of accelerator designs, SCALE-Sim lets designers to capture the behavior of a simulator at each cycle of operation. The tool reads workload configurations as layer hyperparameters, and architectural configurations as inputs, and then generates cycle accurate multilevel memory traces and lumped performance metrics as outputs

Schedule (April 16, 2021; Eastern Time)

Remark: You will get a link to the live Zoom-Session on ASPLOS' clowdr webpage.

Time Agenda Presenter Resources
10:10-10:45 Introduction to DNNs and Accelerator Design Tushar, Paul [Slides], [Youtube]
10:45-11:15 Overview of SCALE-Sim Paul, Anand, Moritz [Slides], [Youtube], [Colab Notebook]
11:15-11:50 Tutorial 1: Design Space Exploration using SCALE-Sim Anand [Slides], [Youtube]
noon-12:40 Tutorial 2: Modifying SCALE-Sim to add custom features Moritz [Slides], [Youtube]
12:45-1:30 Tutorial 3: Using SCALE-Sim to build larger simulators Anand [Slides], [Youtube]
1:30-2:00 Plenum: Discussion on future roadmap, planned features, and a ideas from the community Yuhao [Slides]


Relevant Papers