Petascale Delivered Whats Past is Prologue IBMs pNext ; The Next - PowerPoint PPT Presentation

IBM Deep Computing – SC08 Petascale Delivered – What’s Past is Prologue IBM’s pNext ; The Next Era of Computing Innovation

IBM Deep Computing A mix of market forces, technological innovations and business conditions have created a tipping point in HPC System-level System-level Performance & Reliability Performance & Reliability Ecosystems & Collaboration HPC Going “Mainstream” Ecosystems & Collaboration HPC Going “Mainstream” • Petascale computing enabling new • Petascale computing enabling new paradigm in complex systems analysis paradigm in complex systems analysis • Mix of traditional HPC and “edge” • Hyper-growth within Departmental • Mix of traditional HPC and “edge” • Hyper-growth within Departmental applications will require new skills and Divisional HPC buyers applications will require new skills and Divisional HPC buyers • x86-based clusters dominate – but are • x86-based clusters dominate – but are not meeting ever-increasing • Growing necessity for collaborative • Large influx of new buyers provides not meeting ever-increasing • Growing necessity for collaborative • Large influx of new buyers provides requirements for ultra-scale users requirements for ultra-scale users opportunity for new solution models opportunity for new solution models initiatives to enhance solution initiatives to enhance solution context and manage risk context and manage risk Software Environmentals Software Environmentals HPC at a • ISV business model challenged • Growing importance and • ISV business model challenged • Growing importance and by new systems by new systems limitations on power limitations on power tipping point consumption, datacenter space, consumption, datacenter space, • RYO / Open-source initiatives • RYO / Open-source initiatives cooling capabilities, noise cooling capabilities, noise gaining strong-hold beyond gaining strong-hold beyond traditional segments traditional segments • Likely governmental actions that • Likely governmental actions that will influence system designs will influence system designs How can we exploit these shifts to meet our Performance & Productivity requirements? IBM Deep Computing – SC08 Page 2

IBM Deep Computing Performance and Productivity Challenges require a Multi- Dimensional Approach Highly Highly Scalable Hybrid Systems Productive Systems Multi-core Systems POWER Comprehensive (Holistic) System Innovation & Optimization IBM Deep Computing – SC08 Page 3

Hybrid Systems IBM Deep Computing June 18, 2008 – Petascale Delivered! Roadrunner Breaks Petaflop Milestone Date added: 18 Jun 2008 Lead engineer Don Grice of I BM inspects the w orld's fastest com puter in the com pany's Poughkeepsie, NY plant. The com puter nicknam ed "Roadrunner" w as built for the Departm ent of Energy's National Nuclear Security Adm inistration and w ill be housed at Los Alam os National Laboratory in New Mexico. I BM engineers in Poughkeepsie, N.Y., Rochester, Minn., Austin, Texas and Yorktow n Heights, N.Y., w orked on the com puter, the first to break a m ilestone know n as a "petaflop" -- the ability to calculate 1 ,0 0 0 -trillion Groundbreaking system operations every second. The com puter will have a profound packs the pow er of 1 0 0 ,0 0 0 laptops -- a stack 1 .5 m iles high. Roadrunner w ill impact on science, prim arily be used to ensure national business security, but w ill also help scientists perform research into energy, and society astronom y, genetics and clim ate change. IBM Deep Computing – SC08 Page 4

Hybrid Systems IBM Deep Computing The making of Roadrunner – Key Building Blocks New Software New Software New Hardware Open Source Bring Up, Test and Benchmarking New Hardware Open Source Bring Up, Test and Benchmarking SDK 3.1 TriBlade 17 Connected Units – 278 Racks Linux Library DaCS ALF Other LS21 Expansion Misc / QS22 LS21 QS22 Systems . Systems Integration Integration Firmware Device Driver A New Programming Model extended from � Los Alamos National Lab standard, cluster computing ( Innovation derived through close collaboration with LANL ) Hybrid and Heterogeneous Hardware � Built around BladeCenter and Industry IB-DDR � Slide 5 IBM Deep Computing – SC08 Page 5

Highly Scalable Multi-Core Systems IBM Deep Computing Role of Highly Scalable Multi-core Systems System continues to Evolve 72 Racks Cabled 8x8x16 Rack Blue Gene/P - a hallmark system for 32 Node Cards leadership performance in a space- 1024 chips, 4096 procs saving, power-efficient package for the most demanding and scalable high-performance computing applications Node Card 1 PF/s (32 chips 4x4x2) 144 TB 32 compute, 0-1 IO cards 14 TF/s 2 TB Compute Card 1 chip, 20 DRAMs 435 GF/s 64 GB Chip 4 processors 13.6 GF/s Front End Node / Service Node 2.0 (or 4.0) GB DDR POWER-based 13.6 GF/s Supports 4-way SMP blade & rack-based offerings 8 MB EDRAM IBM Deep Computing – SC08 Page 6

Highly Productive Systems IBM Deep Computing Highly Productive Computing Systems Next-Generation High-speed Scalable, Balanced Simple, Efficient Superb Reliability, POWER processor Network System Design Systems Mgmt. Availability, Serviceability Enhanced POWER Advanced, Eco-Friendly Footprint, Architecturally Robust Instruction Set Architecture Packaging, Power & Cooling Hardware & Software Design DARPA HPCS Performance Goals: Performance dimension Performance dimension Improvement Improvement 10X 10X Sustained execution speed Sustained execution speed Data processing rate Data processing rate up to 50X up to 50X up to 2,000X up to 2,000X Memory access Memory access Minimum bandwidth Minimum bandwidth up to 3,000X up to 3,000X connecting system connecting system components components IBM Deep Computing – SC08 Page 7

IBM Deep Computing Five Decades of Eco-friendly System Innovations Airgap Modular refrigeration S/360 Model 67 High-k cooling technology first virtualized Thermal conduction metal gates CMOS cooling technology processors machine POWER6 IBM Energy Efficiency Institute, Austin, TX 1960s -1970s mid-1990s 2000s late-1990s 1980s Cell BE processor VM virtualization Copper chip Air / liquid hybrid eDRAM cooling 3D chip Flat plate conduction technology stacking cooling technology IBM Deep Computing – SC08 Page 8

Holistic System Innovation IBM Deep Computing Holistic Design as Applied to Petascale Systems I nnovation from Atoms to Software I nnovation from Atoms to Software � The simultaneous optimization of � The simultaneous optimization of: Materials, Devices, Circuits, Cores, Chips, System Architecture, System Assets and System Software � Provides the most effective means to optimize the value of IT offerings to the end user � Catalyst for innovations in HPC: Systems, Algorithms, Software Tools, Business Models Note: Execution relies upon the seamless integration of skills from across the spectrum IBM Deep Computing – SC08 Page 9

IBM Deep Computing Market Dynamics and Design Principles; Petascale and Beyond � Existing and emerging workloads with highly specialized (heterogeneous) requirements � Increased societal awareness and regulatory mandate on environmentals/carbon footprint (power consumption, noise, space efficiency, resource lifecycle) � Open, adaptive software ecosystem that supports legacy applications and hardware architectures while enabling new, advanced capabilities � Robust, modular hardware architecture framework that enables new and existing processor types to be applied for workloads acceleration � Reliability and Scalability for 10 15 and beyond.... IBM Deep Computing – SC08 Page 10

IBM Deep Computing Next Era of Innovation – Hybrid Computing The Next Bold Step in Innovation & Integration Symmetric Multiprocessing Era Hybrid Computing Era Today pNext 1.0 pNext 2.0 p6 p7 Traditional Cell Throughput Computational BlueGene Technology Out Market In Driven by cores/threads Driven by workload consolidation IBM Deep Computing – SC08 Page 11 All statements regarding IBM future directions and intent are subject to change or withdrawal without notice and represent goals and objectives only. Any reliance on these Statements of General Direction is at the relying party's sole risk and will not create liability or obligation for IBM.

IBM Deep Computing HPC Cluster Directions ExaScale ExaF Accelerators Targeted Configurability Performance BG/Q Accelerators Accelerators PF Roadrunner PERCS Systems Extended Configurability BG/P Capability Machines Accelerators Power, AIX/ Linux Accelerators Capacity Linux Clusters Power, x86-64, Clusters Less Demanding Communication 2007 2008 2018-19 2009/2010 2012 2013 2011 IBM Deep Computing – SC08 Page 12 All statements regarding IBM future directions and intent are subject to change or withdrawal without notice and represent goals and objectives only. Any reliance on these Statements of General Direction is at the relying party's sole risk and will not create liability or obligation for IBM.