Max-Planck-Institut für Informatik
max planck institut
mpii logo Minerva of the Max Planck Society

MPI-INF or MPI-SWS or Local Campus Event Calendar

<< Previous Entry Next Entry >> New Event Entry Edit this Entry Login to DB (to update, delete)
What and Who
Title:Learning-Based Hardware/Software Power and Performance Prediction
Speaker:Andreas Gerstlauer
coming from:University of Texas at Austin
Speakers Bio:Andreas Gerstlauer is an Associate Professor in Electrical and Computer Engineering at The University of Texas at Austin. He received his Ph.D. in Information and Computer Science from the University of California, Irvine (UCI) in 2004. His research interests include system-level design automation, system modeling, design languages and methodologies, and embedded hardware and software synthesis.
Event Type:SWS Colloquium
Visibility:D1, D2, D3, D4, D5, SWS, RG1, MMCI
We use this to send out email in the morning.
Level:AG Audience
Date, Time and Location
Date:Monday, 11 June 2018
Duration:60 Minutes
Building:E1 5
Next to performance, early power and energy estimation is a key challenge in the design of computer systems. Traditional simulation-based methods are often too slow while existing analytical models are often not sufficiently accurate. In this talk, I will present our work on bridging this gap by providing fast yet accurate alternatives for power and performance modeling of software and hardware. In the past, we have pioneered so-called source-level and host-compiled simulation techniques that are based on back-annotation of source code with semi-analytically estimated target metrics. More recently, we have studied alternative approaches in which we employ advanced machine learning techniques to synthesize analytical proxy models that can extract latent correlations and accurately predict time-varying power and performance of an application running on a target platform purely from data obtained while executing the application natively on a completely different host machine. We have developed such learning-based approaches for both hardware and software. On the hardware side, learning-based models for white-box and black-box hardware accelerators reach simulation speeds of 1 Mcycles/s at 97% accuracy. On the software side, depending on the granularity at which prediction is performed, cross-platform prediction can achieve more than 95% accuracy at more than 3 GIPS ofequivalent simulation throughput.
Name(s):Susanne Girard
Video Broadcast
Video Broadcast:YesTo Location:Kaiserslautern
To Building:G26To Room:111
Meeting ID:
Tags, Category, Keywords and additional notes
Attachments, File(s):
Susanne Girard/MPI-SWS, 06/08/2018 12:38 PM
Last modified:
Uwe Brahm/MPII/DE, 06/11/2018 07:01 AM
  • Susanne Girard, 06/08/2018 12:42 PM -- Created document.