This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Green/Top 500 June 2011
Top5
00 ra
nkProcessor CPU Type Cores CPU Speed Peak FP Power Gflops/wat
Date
BG/L PPC 440 2 700 MHz 5.6 GF 17 watts 0.33 2004
BG/P PPC 450 4 850 MHz 13.6 GF 16 watts 0.85 2007
BG/Q PPC A2 18 1.6 GHz 205 GF 55 watts 3.72 2011
BG/P: 379 Mflops/watt
BG/L: 205 Mflops/watt
BG/Q: 1689 Mflops/watt
41Mexico DF, November, 2011 SBAC-PAD, Vitoria October 28th, 2011
Mflops/watt
BG/Q: 2000 Mflops/watt
Green500MinoTauro - First in Europe● - November 2011
● Most performing system in Europe● World: 7● 1st of the 3rd architecture in the list● 1266 26 Mflops/Watt● 1266.26 Mflops/Watt
●● Most performing system in Spain● World: 114, Europe: 35● 15 Tflops peak en x86_64● 167 Tflops peak en GPU
MinoTauro: the most energy-efficient computer in Europe
MinoTauro is highest ranking European machine in the November 2011 edition of the Green500 List
The 7th most energ efficient in theThe 7th most energy-efficient in the world, the 3rd by architecture
Based on bullx nodes, equipped with Intel processors and NVIDIA GPUs
“The Bullx systems help institutions to reinforce Europe’s future competitiveness.
43Mexico DF, November, 2011
Bull strongly believes in this competitiveness, which can only be expressed through joining our R+D capabilities. In this sense the BSC is an example to all of us.”
Julio del Valle, Director of Bull Spain
44Mexico DF, November, 2011 Thanks to S. Borkar
3
45Mexico DF, November, 2011 Thanks to S. Borkar
46Mexico DF, November, 2011 Thanks Bill Dally
4
47Mexico DF, November, 2011
… and systems
● PRACE prototype @ BSC: ARM + mobile GPU
Tegra3 Q7 module:1x Tegra3 SoC
4x Corext-A9 @ 1.5 GHz Rack:
● Mont-blanc
4 GB DDR3 DRAM6 GFLOPS
~4 Watt1 Gbe interconnect
1U Multi-board container:1x Board container
8x Q7 carrier boards32x ARM Corext-A9 Cores
8x GT520MX GPU32 GB DDR3 DRAM
1.2 TFLOPS~140 Watt
32x Board container10x 48-port 1GbE switches
256x Q7 carrier boards256x Tegra3 SoC
1024x ARM Corext-A9 Cores256x GT520MX GPU
1TB DDR3 DRAM38 TFLOPS
~5 Kwatt
7.5 GFLOPS / W
Nvidia GeForce 520MX48 CUDA cores @ 900 MHz
142 GFLOPS12 Watts
11.8 GFLOPS / W
48Mexico DF, November, 2011
● Mont blanc● EC funded project, Just started● Low power, Embedded technology
exascale● Low power components + MPI/OmpSs● Integrated prototype, applications and
future system designMont-Blanc ICT-28877748
5
EU Mont-Blanc project
Exascale approach using European embedded power-efficient technology● Objective 1: Develop a prototype system based on
embedded technology● 2013: Scalable to 50 PFLOPS on 7 Mwatts (7 GFLOPS / Watt
#1 competitive
(efficiency)
● Objective 2: Design next-generation system to lead the Exascale race● 2017: Scalable to 200 PFLOPS on 10 Mwatts (20 GFLOPS / Watt)
● Objective 3: Develop a suite of exascale-class applications using theOmpSs task-based programming model for scalability and efficiency
#1 competitive
49Mexico DF, November, 2011
Hw and Sw providers and integrators
HPC applicationproviders
Energy-efficient prototype series @ BSC
20 GF/W
50 PFLOPS7 MWatt
200 PFLOPS10 MWatt
0 2 GF/W
3.5 GF/W
7 GF/W
256 nodes512 GFLOPS
1.7 Kwatt
1024 nodes152 TFLOPS
20 Kwatt
50Mexico DF, November, 2011
● Start from COTS components● Move on to integrated systems and custom HPC technology
2011 2012 2013 2014 2015 2016 2017
0.2 GF/W
6
Mont-Blanc: designing a new energy-efficient Exascale machine
Project will: Build a fully functional prototype using commercially available low-power embedded technology
D i E l hi tDesign an Exascale machine to overcome limitations identified in the prototype
Develop Exascale applications to run on this new generation of HPC systems
“Supercomputers, once built from hancrafted i it t f d h i
51Mexico DF, November, 2011
circuitry, were transformed when companies started assembling them from inexpensive PC-style microprocessors. Researchers in Barcelona are placing an early bet that the next big leap will be cellphone chips.”The Wall Street Journal
Barcelona Supercomputing Center
● The BSC-CNS objectives:● R&D in Computer Sciences, Life Sciences and Earth Sciences● Supercomputing support to external research
● BSC-CNS is a consortium that includes :● the Spanish Government (MEC) – 51%● the Catalonian Government (DIUE) – 37%● the Technical University of Catalonia (UPC) – 12%