Workshop: Innovation Procurement in Horizon 2020 PCP Contractors wanted Supercomputing Centre Institute for Advanced Simulation / FZJ 1 www.prace-ri.eu
Challenges: Aging Society Energy Food How we can address these challenges? Growing population Climate change
Simulation: the third pillar of science Traditional Scientific - Technical Paradigm 1. Create a theoretical model 2. Perform experiments and/or construct a prototype Galileo Galilei folio 116v (before 1609) 3 www.prace-ri.eu
Simulation: the third pillar of science Constraints Too difficult (construct bigger wind tunnel) Too expensive (construct one-way aircraft) Too slow (wait for the climate change) Too dangerous (drugs, climate, ) Computational Science Paradigm 3. Use of HPC resources for simulation of the phenomenon based on known physics laws and efficient numerical methods 4 www.prace-ri.eu
Why does Energy Efficiency matter? HPC energy wall Today's most powerful efficient system: Shoubu System B at 17 GFlop/s/W (Green500 November 2017) An exascale system based on this technology would consume at least 60 MW Assuming 0.2 EUR/kWh the electricity costs over 5 years would cost about EUR 500M Comparison to EuroHPC budget: About EUR 1B for 2 pre- and 2 exascale systems Energy efficient HPC technology Other Data Centre (Big Data) World wide Data Centre electricity use forecast for 2020*: 1031 billion kwh (equivalent to France + Germany + Canada + Brazil) With a carbon footprint between 259 and 533 MtCO2 Market for energy efficient tech must be sustainable (beyond HPC) * Source = SMART Global 2020 Make IT Green Cloud Computing and its Contribution to Climate Change 5 www.prace-ri.eu
Goals of the Whole System Design for Energy Efficient HPC PCP Fostering advances in energy efficiency (major TCO driver for HPC & Big Data) Energy wall for Exascale requires an O(100) increase of power efficiency Assessment of results through pilots scalable to 100 PFlop/s PRACE PCP Process and assessment methodology 3 phases competitive process: - Solution design (6 months, funding 10%) - Prototype development (10 months, funding 30%) - Pre-Commercial Pilot system (16 months, funding 60%) Assessment on real application benchmark from PRACE (suitable for PRACE 2) Expected results and impact EU HPC supply industry (80% of R&D must be performed in EU) increase competitiveness EU HPC users (academia &industry) get early access to disruptive technology through PCP pilot EC and PRACE are learning by doing new public procurement procedure, with high leverage effect PRACE procure Intellectual Property (IP) that paves the way toward sustainable Exascale: - IP is kept by HPC suppliers - Discount on future IP usage for PRACE members of the Group of Procurers of the PCP 6 www.prace-ri.eu
Phase 1 Result assessment Supplier selec on Phase 2 Result assessment Supplier selec on Phase 3 Result assessment Lessons learnt PRACE PCP Phases Solution design Phase 1-6 months EXECUTION STAGE Prototype development Phase 2-10 months Pre-commercial small scale production/ service development -Field test Phase 3-16 months Funding: 10% Funding: 30% Funding: 60% Supplier A Supplier B Supplier B Supplier B Supplier C Supplier C Supplier C Supplier D Supplier E Supplier F Supplier D 1 month 3 weeks 2 weeks 1 month 3 weeks 2 weeks Call closure - tender submission deadline Call closure - tender submission deadline Publication of award criteria Publication of award criteria PRACE T1 T1 + 6 months T1 + 7 months T1 + 17 months T1 + 18 months Final outputs: * 3 prototypes * R&D (IP) useful for T1+34 months Framework contracts and contracts for phase 1 Award contractstart phase 2 Award contractstart phase 3 Lessons learnt - Recommenda ons use PcP for HPC 7 www.prace-ri.eu
Organisation of the PRACE PCP Procurers Five PRACE-3IP partners (CINECA, CSC, EPCC, Juelich, GENCI) and PRACE aisbl as observer A GoP (Group of Procurers) was formed and contractually regulated by an agreement 9 M Budget contributed by the procurers and EC (50/50) Governance Based on the GoP Committee as decision-making entity CINECA has been selected as the Procuring Entity Implementation of a coordination between the Procuring Entity and the project 1 st joint international PCP in the context of HPC organised in Europe assessment of this innovative procurement tool 8 www.prace-ri.eu
Main Technical Goals Energy efficiency in whole system design Target energy efficiency in whole system design Contractors free to choose what to optimise Self-contained pilot system Allow for results to be tested in an HPC centre with real applications Architecture suitable for PRACE Pilot system with 1 PFlop/s, scalable to 100 PFlop/s Capable of executing representative set of applications Energy measurement capabilities Demonstration of improved energy-to-solution of real production applications Energy efficient technology with sustainable market Aim for sustainable results 9 www.prace-ri.eu
Start of the Procurement Brussels, 21 Sep 2012: Open Dialogue organised by PRACE 45 participants (representing 26 companies) Peanuts! Some statements from the companies: I doubt that the kind of scientific advances you seek can be delivered on a tiny budget doing the tendering in 2 phases might be the best way to go when decisions are supplier company/consortium driven, nobody will eventually want to buy the demonstrators (because they miss the customer demand). nice R&D goals will be defined by the consortia which everybody likes but which are too far out. This would lead to technical demonstrators being developed that will not reach a commercial product state. Market dialogue is important (different expectations, new instrument) 10 www.prace-ri.eu
Implementation of the Benchmark Suite Criteria for Benchmark suite (BQCD, SPECFEM3D, QUANTUM E, NEMO) + LINPACK Wide scientific domain coverage Used in production in PRACE Tier-0 systems Energy-efficiency references available for contractors Codes From UEABS (Unified European Application Benchmark Suite) PetaFlop scalability Active development Portability on hybrid machines Machines for validation: PRACE Tier-0 or Tier-1 from PRACE partners CURIE and JURORA: x86_64 thin nodes (w & w/o GPU) JUQUEEN: IBM BG/Q SISU: Cray XC30 11 www.prace-ri.eu
Timeline of Pilot system installation Atos Frioul E4 DAVIDE Maxeler JuMax Envisioned Jun-17 Jul-17 Aug-17 Available for PCP partners 01-Sep-17 16-Oct-17 02-Nov-17 12 www.prace-ri.eu
A follow-up Project HBP - Whole System Design for Interactive Supercomputing Future large-scale simulations within the HBP running on next generation, pre-exascale HPC platforms will require the possibility for data-intensive interactive simulations, analysis and visualisation PCP Timeline: Start in December 2013 with open Dialogue Event End in March 2017 with Wrap-up Workshop Benefits of the PRACE PCP experience: Tender Documents used as reference 13 www.prace-ri.eu
What comes next? PPI4HPC Project Objective: joint evaluation of the deployed solutions and joint procurement of next generation of production systems, not test systems or demonstrators 4 public procurers: BSC; CINECA; GENCI/CEA; JUELICH Budget: about 73 Mio EUR (EU funding 26 Mio EUR) Goals: Foster science and engineering applications in Europe by providing more computing and/or storage resources Promote R&I on HPC architectures and technologies in Europe A greater weight and more impact on common topics of innovation 14 www.prace-ri.eu
THANK YOU FOR YOUR ATTENTION www.prace-ri.eu 15 www.prace-ri.eu