Upload
nguyenkhue
View
213
Download
0
Embed Size (px)
Citation preview
| Jun 26, 2018 | HPC & Quantum | © Atos2
We are THE European IT LeaderWe are THE European HPC leader
€ 13 bnin 2017
73countries
100,000
| Jun 26, 2018 | HPC & Quantum | © Atos
Quantum Computing
Modular DataCentres
HPCaaS
3
Expanding HPC limits
HPC & AI systemsData ManagementSW environment
| Jun 26, 2018 | HPC & Quantum | © Atos
BullSequana X1000 the Bull exascale generation of supercomputer
▶ Open and modular platform designed for the long-term To preserve customer investments
To integrate current and future technologies
Multiple compute nodes: Xeon-EP, ARM, NVIDIA© GPUs, other architectures…
▶ Scales up to tens of thousands of nodes Large building blocks to facilitate scaling
Large systems with DLC: 250-64k nodes
▶ Embedding the fastest interconnects Multiple Interconnects: BXI, InfiniBand
Optimized interconnect topology for large basic cell / DLC (288 nodes / 96 nodes)
Fully non-blocking within cell
▶ Lowest TCO / Ultra-energy efficient Enhanced DLC – up to 40°C for inlet water and ~100%
DLC
4
| Jun 26, 2018 | HPC & Quantum | © Atos
BullSequana X1000 cell
Compute rack
Compute Rack “base”
90kW DLC PSUs
25GB/s sideplanecables (NIC L1 )
x3 Hydraulics chassis (heat exchange)
x48 Compute blades
Switch rack
Switch rack
12kW DLC PSUs
25GB/s copper cable midplane (L1 L2)
Ethernet Top/Leaf switches for island mgt. (behind midplane)
Management servers
Compute Rack “extension”
5
| Jun 26, 2018 | HPC & Quantum | © Atos7
BullSequana X1000 – Embedded interconnect
Fast Interconnect
layout
L2
L1
---12---
---12---
24
Nodes
24 24
Nd Nd
L2
L1
24
NodesNd IO/svc
Fat-Tree – 2 Levels – 288 nodes
| Jun 26, 2018 | HPC & Quantum | © Atos
BullSequana X1000 cells interconnection
... 48x ...
L3 L3
L3
L3 switches ...
L3
Direct connections
optical links
... 9x ...
Fat-Tree3L = 13.8k nodes64k nodes => 4L
8
| Jun 26, 2018 | HPC & Quantum | © Atos
BullSequana compute blade: X1120Intel® Xeon® Processor Scalable Family (Purley)
XeonSkylake /
Cascade lake
6x DIMMs/Channels+ 2x 3D-Xpoint NVRAM
on each socket
Interconnect Mezzanine SKL Intel®UPI
LBGDMI
3x16 PCIe*
3x16 PCIe*
SKL
▶ 1U form factor
▶ Direct liquid cooling
▶ 3 compute nodes per blade with:
– 2 Intel Xeon SKX / CLX processors
– 16 DDR4 DIMM slots
– Optional NVRAM
– 1 I/C mezzanine board (BXI or EDR)
– 1 optional PCIe switch board
– Optional 1 x 2.5” 7 mm SATA SSD
• Or 1x 2.5” PCIe flash SSD
9
| Jun 26, 2018 | HPC & Quantum | © Atos
CPU CPU
NIC NIC
PCIeswitch
PCIeswitch
GPU
GPUGPU
GPU
BullSequana accelerator blade: X1125Nvidia Volta
CPU boardIntel xeon
Interconnect Mezzanine
1U form factor
Direct liquid cooling
1 compute node per blade with:
− 1 CPU Board with 2 sockets
− 1 GPU board supports up to 4 GPUs
− 2 I/C mezzanine boards (BXI or EDR)
− Optional 1 x 2.5” 7 mm SATA SSD
10
Mezzanine board with PCIe switches
| Jun 26, 2018 | HPC & Quantum | © Atos
BullSequana compute blade: X1310Cavium ThunderX2™ (ARMv8) processor
CaviumThunderX2™
InterconnectMezzanine
▶ 1U form factor
▶ Direct liquid cooling
▶ 3 compute nodes per blade with:
– 2 Cavium ThunderX2 processors with
• Up to 32 custom ARMv8 cores per CPU, 4 threads per core, up to 2.5GHz
– 16 DDR4 @266MT/s DIMM slots
– 1 Interconnect mezzanine board (EDR)
– 1 optional PCIe switch mezzanine (TBC)
– Disks
• 1 x 2.5” 7 mm SATA SSD
• Or 1x 2.5” PCIe flash SSD, connected to an embedded PCIe switch (TBC)
11
| Jun 26, 2018 | HPC & Quantum | © Atos
Cavium ThunderX-2 is a viable alternative for HPC workloads
S. McIntosh-Smith, J. Price, T. Deakin and A. Poenaru, “Comparative Benchmarking of the First Generation of HPC-OptimisedArm Processors on Isambard”, CUG 2018, Stockholm, May 2018
15
http://uob-hpc.github.io/2018/05/23/CUG18.html
https://www.anandtech.com/show/12694/assessing-cavium-thunderx2-arm-server-reality
https://www.servethehome.com/cavium-thunderx2-review-benchmarks-real-arm-server-option/
http://montblanc-project.eu/project/presentation
| Jun 26, 2018 | HPC & Quantum | © Atos16
Quantum Computing SpeedupApplications focus
TIM
E
Complexity
billions of years
| Jun 26, 2018 | HPC & Quantum | © Atos
Some of the Applications DomainQuantum Speedup expected
Cryptography Integer factorization Shor algorithm & derivatives exponential speedup
Quantum database searchGrover algorithm and affiliates –polynomial speedup
Chemistry, science of materialsHamiltonianSimulation
Statistical AnalysisMathematicalcomputationexponentialspeedup
17
| Jun 26, 2018 | HPC & Quantum | © Atos
Quantum computing research areasWhere are we now?
Fundamental research
Industrialization
10-15y
Design & Prototype
Time
Technolo
gy
18
Today
1970
| Jun 26, 2018 | HPC & Quantum | © Atos
The Atos Quantum Learning Machine (QLM)An appliance to simulate up to 41 qubits
19
Atos QLM Software
Atos QLM Hardware
QPU interfaceOptimizersSimulators
Noise model
Programming environment
Optimized In-memory InfrastructureScalable and Modular
Atos QLM Appliance
ModularityScalability
| Jun 26, 2018 | HPC & Quantum | © Atos21
Our customersAtos Quantum Computing Scientific Community (aQCSC)
| Jun 26, 2018 | HPC & Quantum | © Atos
Big Data Analytics
IOT
AI Machine Learning
European Processor Initiative
Quantum
22
Exascale and beyond
HPC 2018
eXascale
Atos, the Atos logo, Atos Codex, Atos Consulting, Atos Worldgrid, Worldline, BlueKiwi, Bull, Canopy the Open Cloud Company, Unify, Yunano, Zero Email, Zero Email Certified and The Zero Email Company are registered trademarks of the Atos group. June 2016. © 2016 Atos. Confidential information owned by Atos, to be used by the recipient only. This document, or any part of it, may not be reproduced, copied, circulated and/or distributed nor quoted without prior written approval from Atos.
Thanks