Explore the cutting-edge world of processor simulations with SimEng in this NHR PerfLab Seminar talk. Delve into the challenges facing microarchitecture researchers and discover how SimEng, a new processor simulation framework, enables fast, accurate, and user-friendly cycle-level modeling of contemporary microarchitectures. Learn about its applications in understanding advanced processors like Apple's M1 and Fujitsu's A64fx, as well as its potential for exploring future designs. Gain insights from Professor Simon McIntosh-Smith's extensive experience in microprocessor architecture and HPC as he discusses the Asimov project, ideal processor design, design space exploration, and various models including Marvel Thunder X2 and Arm V1. Examine the implementation of HPC benchmarks, the Instruction Trace Viewer, and vector designs, while considering future plans and funding opportunities in this comprehensive exploration of next-generation processor simulations.
Enabling Next-Generation Processor Simulations with SimEng - NHR PerfLab Seminar
Overview
Syllabus
Introduction
Steves background
SimEng advantages
Asimov project
Ideal processor
Design space exploration
Design goals
Marvel Thunder X2
YAML files
Other models
Current work
HPC Benchmarks
Instruction Trace Viewer
Lark
Vector designs
Vector widths
Linear algebra kernels
DGM
Arm V1
Scalable Matrix Extension
A64FX Extension
Resnet
Future plans
Conclusions
Funding
Questions
Template magic
Simplicity
Configuration
Floating Point Operations
SimEng Code
Memory Traffic
Manual Work
Documentation
Reverse Engineering
Usability
Ease of use
Build a static binary
Taught by
NHR@FAU
