Brand-New Vector Supercomputer

Transcription

1 Brand-New Vector Supercomputer NEC Corporation IT Platform Division Shintaro MOMOSE SC13 1

2 New Product NEC Released A Brand-New Vector Supercomputer, SX-ACE Just Now. Vector Supercomputer for Memory Bandwidth Intensive Applications Sustained performance Usability Productivity 2 NEC Corporation, 2013

3 Concept 3

4 Performance SX History and Technical Evolutions NEC has always provided the high sustained performance by Vector Super-Computer SX series. Distributed Parallelization (MPI-SX) Automatic Parallelization Function & SUPER-UX Auto Vectorization Compiler SX-2 SX-3 Bipolar Water Cooling 4 NEC Corporation, 2013 SX-4 Multi node CMOS Air Cooling ES Support Over 100 nodes Cluster SX-5 MPI support to Multilane IXS SX-7 SX-6 1 Chip Vector Processor SX-8/8R 3D node module Support >1000 nodes ECO ES2 SX-9 100GF Processor Multi-core All in One Chip ECO

5 big core fewer cores Trend of TOP500 (1 st ~ 10 th system) Growing of LINPAC performance has been provided by system enlarging User mast spend their time to extract massively parallelism Smaller # of cores with big cores can reduce the difficulty 1.0E E E E E+06 Linpack [TF] Lipack ave. [TF] # of cores # of cores ave. # of nodes # of nodes ave. Core performance [GF] Core perfromance ave. [GF] Frequency [GHz] Frequency ave. [GHz] Exa Flops Increasing 1.0E E E E E E+00 Frequency [GHz] 115%/year Nearly constant 1.0E NEC Corporation, 2013

6 Required memory bandwidth [Byte/Flop] Required Byte/Flop in Real Applications According to Japanese Government (MEXT) working group report of wide variety of strategic segment applications, diverse characteristics are observed. MEXT: Ministry of Education, Culture, Sports, Science & Technology B/F requirement from each application is differ greatly. Only one architecture cannot cover all application areas MD, Weather Cosmo physics Particle physics Structural analysis Fluid dynamics memory bandwidth intensive Quantum chemistry Nuclear physics Calculation intensive scalar CPUs Required memory capacity [PB] Reference: Report on Strategic Direction/Development of HPC in Japan, March NEC Corporation, 2013

7 Concepts of SX-ACE Big Core Providing higher sustained performance The highest single core performance: The largest single core memory bandwidth: Inherit SX-DNA 64GF 64~256GB/s Low Power Consumption Higher power efficiency compared to SX-9 Small Installation Space To reduce floor space cost Compared to SX NEC Corporation, 2013

8 Providing Big Core The SX-ACE inherits SX-DNA and overwhelm other CPUs with its world s No.1 CPU core performance and word s No.1 memory bandwidth. Single core comparison SX-ACE 64GF 64GB/s Scalar A Scalar B Scalar C ~4x 4~13x 5 peak performance 2~4x performance compared to competitors memory bandwidth 4~13x performance compared to competitors 8 NEC Corporation, 2013

9 Architecture 9

10 CPU Architecture (Big Core, Large memory bandwidth) Scalar Processing Unit Vector Processing Unit Remote access Control Unit SPU VPU Architecture CORE Vector 256GB/s ADB (Assignable Data Buffer) core core core RCU Interconnect 8GB/s x2 Performance ADB size ADB bandwidth 64GFlops 1MB 256GB/s 256GB/s 8GB/s x2 Memory bandwidth 64GB/s~ 256GB/s crossbar Memory Byte/Flop 1.0 ~ 4.0 CPU 256GB/s Cores 4 Memory controller Performance 256GFlops Memory bandwidth 256GB/s 256GB/s Byte/Flop 1.0 Memory(DDR3) 10 NEC Corporation, 2013

11 All-in-one Processor 4 powerful cores and each controller (memory, network, I/O) are integrated in one-cpu = Power saving Compact card design = Space saving I/O Controller Connection to storage device, Ethernet Network controller 8GB/s/direction, Fat-tree Powerful core World s fastest CPU core 64GF x 4cores 1MB ADB/core SX-ACE CPU Performance: 256GF Memory bandwidth: 256GB/s Node card 11cm memory 11 NEC Corporation, 2013 Memory Controller 256GB/s BW control Very large memory BW World s largest BW 256GB/s 37cm

12 Node Card CPU 4 cores 256GF 256GB/s 37cm 11cm Memory 4GB x 16DIMMs DDR3 2000MHz 12 (c) NEC Corporation, 2012

13 node #000 node #015 node #016 node #031 node #496 node #511 System Configuration Each node is connected with 2 stages full Fat-tree network Implementing global communication function into HW Fat-tree HW function of global communication 32 links SW2 #00 SW2 #15 16 links SW1 #00 SW1 #01 SW1 #31 16 links 8GB/s /direction node 13 NEC Corporation, nodes, 2048 cores, 131TFlops, 1B/F

14 Configuration System Rack 64 nodes = 16TF, 16TB/s 16-Node Cage x4 4 cages = 32 modules = 64 nodes = 64CPUs 16-Node Cage 8 modules = 16 nodes = 16 CPUs 2-Node Module 2 nodes = 2 CPUs Node Card 1CPU, 256GF, 256GB/s Rack Specifications 16TF, 16TB/s, 64 CPUs 0.75m x 1.5m x 2.0m 30KW 14 (c) NEC Corporation, 2013

15 42U coolant pipe (inlet) coolant pipe (outlet) Detail of Rack Implementation network switch node manager node manager 16-node cage node manager node manager OS disk rack manager 2-node module power junction 15 (c) NEC Corporation, 2013

16 Downsizing and Power Saving Providing 5x smaller space and 10x lower power consumption compared to SX-9 by power saving design and compact implementation. SX-9 Comparison with same performance (131TF) SX-ACE 24m 12m 25m swimming pool size 80 nodes 512 nodes 8m 7m Meeting room size 131TF 288m 2 2.4MW 16 NEC Corporation, 2013 space 1/5 power 1/10 131TF 56m MW

17 NEC s Exhibitor Forum is Today! Exhibitor Forum Nov. 19 th (Tue), 15:30 16:00 Room 501/502 NEC s Brand-New Vector Supercomputer and HPC Roadmap 17 (c) NEC Corporation, 2013

18 18 NEC Corporation, 2012 Confidential