Fabrizio Magugliani - PRACE

14

Transcript of Fabrizio Magugliani - PRACE

Page 1: Fabrizio Magugliani - PRACE
Page 2: Fabrizio Magugliani - PRACE

Fabrizio [email protected] Planning and Business Development

Open Information Day on Acquisitions of Supercomputers by the EuroHPC Joint Undertaking

March 11, 2019

THE LONG AND WINDING ROAD (towards Exascale)

GCD

Page 3: Fabrizio Magugliani - PRACE

E4 Computer Engineering

The vision for a European HPC environment in 2030

The Roadmap

Area of Investments

Q&A

TABLE OF CONTENTS

3

Page 4: Fabrizio Magugliani - PRACE

OUR MEMBERSHIPS

Silver LevelCosimo Gianfreda

Member of the Steering Board http://www.etp4hpc.eu

Member of the Consortiumhttp://european-processor-initiative.com

Member of the OEHI (Open Edge and HPC Initiative)

Member of CERN openlab

5

Member of the Center of Excellence 

Page 5: Fabrizio Magugliani - PRACE

COMPANY MILESTONES

6

Company Milestones

2005

First Infiniband Cluster

2012

First ARM+GPUServer

2016

First OperPOWERServer

2009First GPU Cluster

2013First ARM

Cluster Deployement

2017First OpenPOWER PFLOPS-class Cluster

299th in TOP500 (June 2017)14th in GREEN500 (June 2017)

2002

E4 Foundation

CO

MPA

NY M

ILESTO

NES

2019

Page 6: Fabrizio Magugliani - PRACE

THE VISION FOR A EUROPEAN HPC ENVIRONMENT IN 2030

7

The confluence of AI, HPC, HPDA, IoT and Cloud

• Workflows that unify artificial intelligence, data analytics, and modeling/simulation using the cloud as infrastructure will make high-performance computing more essential and ubiquitous than ever. But it will be invisible.

• Multi-layer compute infrastructure, self-organizing via intelligent management

• Application-aware components for power management

• Suggested reading: The Emerging Machine Society(https://www.gartner.com/webinar/3837563?pcp=wb_ddf&srcId=1-3478922220)

• Machine Society Will Not Be Designed, It Will Emerge. • The technologies in play over the next decade have the potential to solve some of the

intractable problems that humanity has faced for so long, offer the opportunity to increase productivity such that all our basics needs are taken care of, and fundamentally reframe the notions of what it means to be a (human) person.

Page 7: Fabrizio Magugliani - PRACE

DISTRIBUTED, MULTI-TIERED STORAGE

DOMAIN SPECIFIC COMPUTE

GENERAL PURPOSE COMPUTE

Embedded data analysis

APPLICATION ECOSYSTEM

APPLICATION-AWARE DISPATCHING ALGORITHM

GPU EnabledCompute Nodes

High Core Count Compute nodes

High memory bandwidth Compute

NodesFPGA Enabled

Compute Nodes

EPI Compute

Nodes

FatCompute Nodes

QC Enabled Compute Nodes

AI/ML/DL

Visualization

HPDA

Virtual/AugmentedReality

CoEs ISVsUsersDevelopers

StorageNodes

ThinCompute Nodes

DSL

THE VISION FOR A EUROPEAN HPC ENVIRONMENT IN 2030

Page 8: Fabrizio Magugliani - PRACE

PRODUCTS | SERVER ARM

THE STARTING POINT

SPECIFICATIONS• 2 x Marvell ThunderX2® CN9975 28-core ARM processors • 8-Channel RDIMM DDR4 per processor, total 24 x DIMMs • 2 x GPU Support• 1 x Dedicated management port • Onboard Broadcom® SAS3008 controller (RAID 0/1/1E/10)• 2 x 2.5" SATA in rear side, 24 x SAS/SATA hot-swappable HDD/SSD bays • SAS expander with 12Gb/s transfer speed • 2 x OCP Gen3 x16 mezzanine slots • 1 x PCIe Gen3 x16 Slot • 2 x 10Gb/s SFP+ LAN ports (QLogic® QL41102)• Aspeed® AST2500 remote management controller • Dual 1200W 80 PLUS Platinum redundant power supply

24-bay ARM Server System with support up to 2 FHFL slots

Page 9: Fabrizio Magugliani - PRACE

PRODUCTS | SERVER ARM

THE EVOLUTION

Extensive node-level and multi-node-level redesign

24-bay ARM Server

System with support up to 2 FHFL slots

Target design point: PUE <= 1.1

=

+

HBM and other technologies (under evaluation)

OCP, Open19 Form Factor (other FF under evaluation)

Liquid cooling for CPUs and accelerators (co-design)

Power-aware, feedback-based, application-aware node management (co-design)

Page 10: Fabrizio Magugliani - PRACE

THE ROADMAP: LEVERAGE NEW TECHNOLOGIES, BUILD ON EXPERTISE

Commercial GPU

Commercial FPGA

2020

2021

2022

12

Commercial ARM

Roadmap

2011

2019 Commercial RISC-V

EPI RISC-V

EPI ARM

Integration with European IP (PRACE PCP program)Co-Design with Italian and European partners

today

Page 11: Fabrizio Magugliani - PRACE

• ARM HPC Tools • BeeGFS• CEPH• Julia• Jupyter notebook• LLVM/GNU• MPI• OpenACC • OpenMP• OpenStack• SLURM• ………..

13

Innovation: PlatformSYSTEM SOFTWARE ECOSYSTEM

Page 12: Fabrizio Magugliani - PRACE

• Ongoing• CERN openlab• CoEs• CLAIRE• EU calls• European Processor Initiative• Open Edge and HPC Initiative

• Planned• EuroHPC JU Petascale, Pre-Exascale and Exascale

procurements • PPI4HB (ICEI/FENIX)• PPI4HPC

14

Synergies with EUINVOLVEMENT/SYNERGIES WITH EU PROJECTS

Page 13: Fabrizio Magugliani - PRACE

15

Develop & test in house innovative technologies and applications to embed them in our products & solutions

Hiring, training and nurturing highly-qualified engineers to be ready for the current and future technological challenges

Consult/advise promising start-up & innovative companies

Leverage ETP4HPC for developing the European ecosystem

AREA OF INVESTMENTS

Page 14: Fabrizio Magugliani - PRACE

E4. WHEN PERFORMANCE MATTERS

THANK YOU

End