Download presentation
Presentation is loading. Please wait.
Published byKathleen Potter Modified over 8 years ago
1
Oct 31 st 2007University of Utah1 Multi-Cores: Architecture/VLSI Perspective The Hardware-Software Relationship: Date or Dump?
2
University of Utah Embedded Applications --Spencer From discreet cochlear implants to high-end biomedical imaging! Multi-cores speed up performance by 50x! Creating new application domains!
3
University of Utah How to use “multiple” cores? Oct 31 st 2007 3 Parallel programming Synchronization Deadlock Livelock Memory management
4
University of Utah Oct 31 st 2007 4 How to use “multiple” cores? Program = Communication + Computation Global restructuring and parallelization
5
University of Utah Oct 31 st 2007 5 Structured “Communication” Lang: StreamIt, MPI Compilers: RAW, CoGenE Architecture: TRIPS, HWRT Key: Help other levels and leverage communication
6
University of Utah Another Constraint? Oct 31 st 2007 6 Parallel programming Synchronization Deadlock Livelock Memory management Hey.. Surprise!!! Communication Scheduling
7
University of Utah Another Constraint? Oct 31 st 2007 7 Oh God!!! Communication Scheduling
8
University of Utah Focus of Architecture Research Reduce the load of programmers Hardware transactional memory Aggressive pre-fetching Dynamic reconfiguration at every possible level Keep the architectural innovations transparent to compilers or programmers Learn from the mistakes of ITANIUM ! Remember the success of OOO execution Oct 31 st 2007 8
9
University of Utah 9 Reliability Issues --Niti Shrinking transistor sizes & lower voltages Increased transient faults, process variations – leakage power and frequency variations, hard errors, interconnect noise Many-core – “Many cores” may not work reliably Some cores will end up providing redundancy Heterogeneous cores may be able to help Simple in-order cores can provide redundancy at low cost The compute power gain of many-core can get offset by reliability requirements of the system
10
University of Utah Oct 31 st 2007 10 On-Chip Sensor Networks --Nathaniel, Amlan Analog sensors everywhere! Need to monitor power, voltage droop, variation, critical paths, delays, slew rates, etc. Control system to react to changes. In multi-core, sensor network will only grow. Xeon and Itanium processors JSSC Jan 06 & 07 On-chip sensors
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.