From Sensors to Supercomputers Eric Van Hensbergen Principal - PowerPoint PPT Presentation

From Sensors to Supercomputers Eric Van Hensbergen Principal Engineer – HPC ARM Research Austin SYSTEMS RESEARCH & DEVELOPMENT

ARM Primer SYSTEMS RESEARCH & DEVELOPMENT

Terminology Architecture Processor Micro-Architecture Processor Hard-Macro “ARMv8” “Cortex-A57” Implementation SYSTEMS RESEARCH & DEVELOPMENT

ARM Business Model IP Processors licence GPUs Physical IP Interconnect chips royalty ARM invests in ecosystem Ecosystem provides value chain with support & products based on ARM technology SYSTEMS RESEARCH & DEVELOPMENT

Wide Choice of Partners SYSTEMS RESEARCH & DEVELOPMENT

Linaro – Distributed Software Organization T echnical Steering Committee (TSC) Working Platform Landing T eams Groups Engineering working under NDA Office of the CTO (OCTO) Kernel Consolidation LAVA Toolchain Validation & Benchmarking Graphics Builds & Baselines Power Management OE/Yocto, Android, Fedora, Ubuntu Segment Groups QA : LEG (15 companies) Continuous Integration for Linaro output LNG (12 companies) Optimization, consolidation Validation, benchmarking SoC support and and innovation and release management optimization SYSTEMS RESEARCH & DEVELOPMENT

ARM: 21 st Century Design § Focuses on Design & Licensing of IP Building Blocks for SoCs (=LEGO’s) § Fosters an eco-system of standard pieces, acting as COTS-on-Silicon § COTS-on-Silicon encourages multi-suppliers through the eco-system § Allowing the eco-system the ability to deliver cost effective solutions § Enforcing localized optimized end designs § Enabling circuit-boards to be miniaturized onto a single chip § While retaining the technology DNA of Energy-Efficiency SYSTEMS RESEARCH & DEVELOPMENT

Example SYSTEMS RESEARCH & DEVELOPMENT

ARM Spectrum Cortex-M series Cortex-R series Cortex-A series (Microcontroller) (Real Time) (Application) E 2+ GHz C N A M R O 64-bit F R E P A57 Fully OoO 1+ GHz x2 R7 MP 28HPM 800+ MHz +FP +DSP 40LP M4 0.12mW/MHz A5 0.68 mm 2 40LP 0.071mW/MHz R4 40G 0.18 mm 2 Y C N 4 uW/Mhz E I C I F F E R E W 0.01 mm 2 O P M0 12K Gates Note: the numbers may vary depending on specific implementation details SYSTEMS RESEARCH & DEVELOPMENT

Markets Mobile 3 % 4.6bn in 2011 H o m e Home 4 0% 0.4bn in 2011 Embedded 25 % 2.3 bn in 2011 Enterprise 10 % 1.4bn in 2011 SYSTEMS RESEARCH & DEVELOPMENT

10 Billion Cores Last Year, 50 Billion Overall SYSTEMS RESEARCH & DEVELOPMENT

Embedded HPC is not a new concept SYSTEMS RESEARCH & DEVELOPMENT

Top 500 Historical SYSTEMS RESEARCH & DEVELOPMENT

Why is ARM interested in HPC 1.E+18 1.E+17 1.E+16 1.E+15 Supercomputing - #1 system on Top 500 1.E+14 1.E+13 1.E+12 Nexus 9 1.E+11 1.E+10 1.E+09 iPad 2 == Cray 2* 1.E+08 1.E+07 1.E+06 1.E+05 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 * J. Dongarra & P . Luszczek HPEC 2012 SYSTEMS RESEARCH & DEVELOPMENT

Why ARM is interesting in HPC GFLOPS/Core 10 15 20 25 30 35 0 5 6/1/1993 11/1/1993 4/1/1994 9/1/1994 2/1/1995 7/1/1995 12/1/1995 5/1/1996 10/1/1996 3/1/1997 8/1/1997 SYSTEMS RESEARCH & DEVELOPMENT 1/1/1998 6/1/1998 11/1/1998 4/1/1999 9/1/1999 2/1/2000 7/1/2000 12/1/2000 5/1/2001 10/1/2001 #1 on 3/1/2002 8/1/2002 1/1/2003 T 6/1/2003 op 500 11/1/2003 4/1/2004 9/1/2004 2/1/2005 7/1/2005 12/1/2005 5/1/2006 10/1/2006 3/1/2007 8/1/2007 1/1/2008 6/1/2008 11/1/2008 4/1/2009 9/1/2009 2/1/2010 7/1/2010 12/1/2010 5/1/2011 10/1/2011 3/1/2012 8/1/2012 1/1/2013 6/1/2013 11/1/2013 4/1/2014

System Design Space Exploration RTL Simulation Co-Design Characterization Centers Design Workloads Sensitivity Studies HPC Disk Image HPC End Performance Customers Projection SYSTEMS RESEARCH & DEVELOPMENT

Research Themes § Co-Design § Workload optimizations and characterization for HPC & big data § Architectural & system design sensitivity sweeps § Simulation and modeling infrastructure § Software Ecosystem Enablement § Operating systems and runtimes targeted and optimized for ARM HPC § Math library enablement of ARM SIMD and architecture § Parallel and vector optimizing compilers and runtimes § Cross-stack optimizations for resiliency and energy efficiency § Acceleration § Compute Acceleration § IO Acceleration (Compute Near Memory, Compute Near Network, Compute Near Storage) SYSTEMS RESEARCH & DEVELOPMENT

Questions? SYSTEMS RESEARCH & DEVELOPMENT