Weitere ähnliche Inhalte Ähnlich wie An Update on Arm HPC (20) Mehr von inside-BigData.com (20) Kürzlich hochgeladen (20) An Update on Arm HPC1. © 2019 Arm Limited
Brent Gorda
September 25th, 2019
Arm in HPC
2. 2 © 2019 Arm Limited
© 2019 Arm Limited
• Parkinson’s & Osteoporosis
Ongoing research in Bristol: New Drugs ‘In Silico’
Images courtesy of Bristol University
3. 3 © 2019 Arm Limited
© 2019 Arm Limited
Multiphysics Simulations: Fluid Dynamics, Heat Diffusion, Electromagnetics
Images courtesy of Bristol University
4. 4 © 2019 Arm Limited
© 2019 Arm Limited
What is “Super” or “High Performance” Computing?
Lake Tahoe ~40 Trillion Gallons of water (4.0x10^12)
~2002 Supercomputers hit 40 Teraflops (Earth Simulator – Japan/NEC)
5. 5 © 2019 Arm Limited
© 2019 Arm Limited
What is “Super” or “High Performance” Computing?
The Great Lakes hold ~6.5 Quadrillion gallons of water (6.5x10^15)
2008 Supercomputers hit 1 Petaflop 1.0x10^15 (US IBM Roadrunner)
6. 6 © 2019 Arm Limited
Top500 systems over the past 25 years
1.00E-01
1.00E+00
1.00E+01
1.00E+02
1.00E+03
1.00E+04
1.00E+05
1.00E+06
1.00E+07
1.00E+08
1.00E+09
1.00E+10
1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
59.7 GFlop/s
422 MFlop/s
1.17 TFlop/s
149 PFlop/s
1.01 PFlop/s
1.56 EFlop/s
1 Gflop/s
1 Tflop/s
100 Mflop/s
100 Gflop/s
100 Tflop/s
10 Gflop/s
10 Tflop/s
1 Pflop/s
100 Pflop/s
10 Pflop/s
1 Eflop/s
SUM
N=1
N=500
Astra HPE/ArmEarth Simulator NEC ‘02 Jaguar AMD ‘09
Images courtesy of www.top500.org
7. 7 © 2019 Arm Limited
© 2019 Arm Limited
These are not embedded devices:
Images courtesy of Bristol University
8. 8 © 2019 Arm Limited
© 2019 Arm Limited
Mont-Blanc
The “legacy” Mont-Blanc vision
Denver, Nov 13th 2017Arm HPC User Group2
Vision: to leverage the fast growing market of mobile technology for
scientific computation, HPC and data centers.
2012 2013 2014 20162015 2017 2018
Mont-Blanc 2
Mont-Blanc 3
Early Research into the Efficacy of Arm for HPC
9. 9 © 2019 Arm Limited
© 2019 Arm Limited
Catalyst UK: Accelerating ARM Adoption in UK
Industry PartnersProgram Goals
Measures of SuccessConfigs & Timeline
– Deployment: Deployment of HPC clusters at
multiple UK sites, supported for 3-year period
providing access to academia & industry
– Adoption: Early adoption of ARM for HPC in UK;
Apollo 70 Early Ship followed by customer collab.
– Applications: Customer-driven porting and opt
– Collaboration: Leveraging the success “Project
Comanche” model of customer-centric
collaboration; but based instead on Early Ship
HPE Apollo 70 product
– Exascale: Establish foundation for Exascale collab
UK Collaborations
Intended outcomes include:
– Critical HPC apps ported and demonstrated
– ISV engagements and demonstrations
– Demonstrated performance improvements
– Publications and follow-on collaborations
– Bugs filed, fixed & up-streamed to open source
– HPE: Apollo 70, HPE Performance Software - Cluster Manager, HPE
Performance Software – Message Passing Interface
– ARM: Allinea Studio (Compiler, Libraries, Forge-DDT & MAP),
OpenHPC
– Mellanox: OFED, HPC-X, OpenMPI, OpenSHMEM, MXM, SHArP
– SuSE: SLES, OpenStack, HPC Module
– Cavium: ThunderX2 SoC, technical support
– Qualcomm: Centriq SoC, technical support (tentative)
– EPCC: WRF, OpenFOAM, Rolls
Royce Hydra opt, 2 PhD candidates
– Leicester: Data-intensive apps,
genomics, MOAB Torque, DiRAC
collab
– Bristol: VASP, CASTEP, Gromacs,
CP2K, Unified Model, Hydra, NAMD,
Oasis, NEMO, OpenIFS, CASINO,
LAMMPS
– UK Government: Dept. for Bus.,
Energy & Industrial Strategy (BEIS)
Typical for each site:
– 64 Apollo 70
– Compute Nodes:
– Cavium 32c, 2.2 GHz
– 256GB memory (16GB
DIMMs)
– IB EDR CX5 Clos
– 4096+ cores
– 6 CL4300 (tentative)
– Services/Storage:
– Qualcomm Centriq
Sep-Dec: Structure
partnership, alignment
Jan: HPE/ARM SOW
Feb: Customer SoWs,
quotations, POs
Mar: SW stack validation (3rd
Party Runtime library)
Apr: Systems build, public
announcements
May: Delivery and acceptance
HPE will deliver >12,000 cores across 3
sites; amongst the largest ARM HPC
deployments in the world
HPE Confidential
Catalyst UK
10. 10 © 2019 Arm Limited
© 2019 Arm Limited
Isambard The World’s First Arm-based Production Supercomputer
11. 11 © 2019 Arm Limited
© 2019 Arm Limited
Vanguard Astra by HPE: #156 on top500
• 2,592 HPE Apollo 70 compute nodes
• 5,184 CPUs, 145,152 cores, 2.3 PFLOPs (peak)
• Marvell ThunderX2 ARM SoC, 28 core, 2.0 GHz
• Memory per node: 128 GB (16 x 8 GB DR DIMMs)
• Aggregate capacity: 332 TB, 885 TB/s (peak)
• Mellanox IB EDR, ConnectX-5
• 112 36-port edges, 3 648-port spine
switches
• Red Hat RHEL for Arm
• HPE Apollo 4520 All–flash Lustre storage
• Storage Capacity: 403 TB (usable)
• Storage Bandwidth: 244 GB/s
12. 12 © 2019 Arm Limited
© 2019 Arm Limited
Exascale – the race underway at the high end
Projected Exascale System Dates
U.S.
▪ Sustained ES*: 2022-2023
▪ Peak ES: 2021
▪ ES Vendors: U.S.
▪ Processors: U.S. (some ARM?)
▪ Cost: $500M-$600M per system
(for early systems), plus heavy
R&D investments
52
China
▪ Sustained ES*: 2021-2022
▪ Peak ES: 2020
▪ Vendors: Chinese (multiple sites)
▪ Processors: Chinese (plus U.S.?)
▪ 13th 5-Year Plan
▪ Cost: $350-$500M per system,
plus heavy R&D
EU
▪ PEAK ES: 2023-2024
▪ Pre-ES: 2020-2022 (~$125M)
▪ Vendors: US and then European
▪ Processors: x86, ARM & RISC-V
▪ Initiatives: EuroHPC, EPI, ETP4HPC, JU
▪ Cost: Over $300M per system, plus heavy
R&D investments
Japan
▪ Sustained ES*: ~2021/2022
▪ Peak ES: Likely as a AI/ML/DL system
▪ Vendors: Japanese
▪ Processors: Japanese ARM
▪ Cost: ~$1B, this includes both 1 system
and the R&D costs
▪ They will also do many smaller size
systems
* 1 exaflops on a 64-bit real application 52© Hyperion Research
13. 13 © 2019 Arm Limited
© 2019 Arm Limited
Exascale - Fujitsu A64FX
14. 14 © 2019 Arm Limited
© 2019 Arm Limited
Exascale – European Processor Initiative
GPP AND COMMON ARCHITECTURE
9
ZEUS MPPA
eFPGA
FPGA
FPGA
ZEUS ZEUS
ZEUS ZEUS EPAC
HBM
memories
DDR
memories
PCIe gen5
links
HSL
links
D2D links
to adjacent chiplets
EPAC - EPI Accelerator (TITAN)
MPPA - Multi-Purpose Processing Array
eFPGA - embedded FPGA
Cryptographic ASIC (EU Sovereignty)
15. 15 © 2019 Arm Limited
Arm HPC Software Ecosystem
ClusterManagementTools:
Bright,HPECMU,xCat,Warewulf
Linux OS Distro of choice:
RHEL, SUSE, CENTOS,…
Arm Server Ready Platform:
Standard OS compatible FW and RAS features
HPC Applications:
Open-source, Owned, and Commercial ISV codes
Job schedulers
and Resource
Management:
SLURM, IBM LSF,
Altair PBS Pro,
etc.
Programming
Languages:
Fortran, C, C++
via
GNU, LLVM, Arm
& OEMs
Debug and
performance
analysis tools:
Arm Forge,
Rogue Wave,
TAU, etc.
Filesystems:
BeeGFS,
LUSTRE, ZFS,
HDFS, GPFS
App/ISA specific optimizations, optimized libs and intrinsics:
Arm PL, BLAS, FFTW, etc.
Communication Stacks and run-times:
Mellanox IB/OFED/HPC-X, OpenMPI, MPICH, MVAPICH2, OpenSHMEM, OpenUCX, HPE MPI
Parallelism
standards:
OpenMP
(omp / gomp),
MPI, SHMEM
(see below)
User-space
utilities,
scripting,
containers, and
other packages:
Singularity,
Openstack,
OpenHPC,
Python, NumPy,
SciPy, etc.
16. 16 © 2019 Arm Limited
Porting HPC apps to the Arm platforms
Ø The platform just works – porting in 2 days is the common experience
Build recipes online at https://gitlab.com/arm-hpc/packages/wikis/home
LAMMPS CESM2 MrBayes Bowtie
AMBER Paraview SIESTA UMNAMD
VASP MILCWRF GEANT4
Quantum
ESPRESSO
DL-Poly NEMOGAMESSOpenFOAM VisIT
QMCPACKAbinitBLAST NWCHEM BWA
GROMACS
Chem/Phys
Weather
CFD
Visualization
Genomics
17. 17 © 2019 Arm Limited
© 2018 Arm Limited
Arm in IOT
We design & license IP, we do not
manufacture chips
Partners build products for their
target markets
One size does not fit for all
HPC is a great fit for
co-design and collaboration
Partnership is key Choice is good
21 billion chips in the past year
Mobile/Embedded/IoT/
Automotive/GPUs
And now … servers
Arm Technology Connects the World
18. 18 © 2019 Arm Limited
© 2019 Arm Limited
Edge
Edge
Critical Data
Massive Amounts of Data
z
z
Edge
5G
CORTEX
HPC
Cloud
Data Centers
The New Architecture
19. 19 © 2019 Arm Limited
Confidential © 2019 Arm Limited
• Historically strong focus on high-end systems and balance:
• B:F Ratio’s of the late 1990’s thru 2010
• Parallel processing at massive scale
• Low-latency / high BW interconnects
• Citing: S/W maintenance, roll-out, cooling/power
• Workloads:
•Historical workloads scientific simulation
•Recent new workloads attracted to “high-end” capabilities of
HPC architectures: big data, Deep Learning/AI
•HPC Leads in technology acceptance (think Formula-1)
HPC is an excellent partner for the ecosystem
HPC is an Architecture
20. 20 © 2019 Arm Limited
© 2019 Arm Limited
Arm is Data driven, from the edge to the core
21. The Cloud to Edge Infrastructure Foundation
for a World of 1T Intelligent Devices
Thank You
Arm.com/hpc