Eurora Supercomputer Tops Green500 List

Top end of Green500 is dominated by heterogeneous supercomputers

3 min read
Eurora Supercomputer Tops Green500 List

Conventionally, supercomputers have been ranked by number-crunching might, but not everyone agrees this is the most important metric to consider. These days, cars are evaluated by fuel consumed per kilometer, not their maximum speed. Supercomputers are following this trend, too. Last Friday marked the latest releaseof the biannual Green500 list, which grades these machines in a nontraditional way: by looking at their performance-per-watt. The list complements Top500, which sorts supercomputers by looking at how fast each machine can solve equations. 

Notably, the top end of the Green500 list is dominated by "heterogeneous" supercomputers—supercomputers built around processor cores that combine a mix of architectures whose composition is suited to the different computing tasks the machine will execute. Various processing elements are fused together, such as traditional central processing units (CPUs), graphical processing units (GPUs), and other types of coprocessors.

Two new entrants—both heterogeneous systems based on NVIDIA's Kepler K20 GPU accelerators—took the top two spots and cleared the bar of three billion floating-point operations per second (gigaflops) per watt. Eurora, installed at the CINECA Supercomputing Center in Italy, topped Green500 at 3.21 gigaflops/watt, and Aurora Tigon was a close second at 3.18 gigaflops/watt. The machines, both manufactured by Eurotech, bested the previous titleholder, University of Tennessee's Beacon (2.45 gigaflops/watt), by 30 percent. 

On the Top500 list, Eurora is ranked way down at No. 467, delivering 100.9 Teraflops per second on the Linpack benchmark. However, what's interesting about the machine is not its peak performance, but rather its energy reuse system. Similar to the iDataCool solution IEEE Spectrumreported on last week, Eurora relies on a water cooling system to draw out excess heat; it then reuses this energy to drive adsorption chillers that cool the data center. It also redirects some of the waste heat to warm up other, human-occupied buildings at the facility.

On the other hand, Tianhe-2, which recently capped the Top500 list as the world's fastest supercomputer, delivered 1.9 gigaflops/watt, largely due to the heterogeneous computing elements in its Intel Xeon Phi processors. This put it at No. 31 on Green500, and bespeaks the energy efficiency considerations many supercomputer manufacturers are now taking into account in building these systems.

According to Wu Feng, founder of Green500 and professor of computer science and electrical and computer engineering at Virginia Tech, the energy efficiencies of the highest-ranked machines on the Green500 list have been making much bigger gains than the have the mean or the median. "Overall, the performance of machines on the Green500 List has increased at a higher rate than their power consumption. That's why the machines' efficiencies are going up," says Feng. "While the gains at the top end of the Green500 appear impressive, overall the improvements have been much more modest."

But while there is still work to do, Feng believes that his team has made great strides overall in bringing green computing to the forefront of high-performance computing. "What we've done is raise awareness that energy efficiency is truly important," says Feng. "I didn't really feel like it got mainstream until 2007 or 2008… Up until that point, nobody really cared about energy efficiency; the only thing people cared about was speed. So it's really come around full circle."

No.
Machine
Installation Site
MFLOPS/W
1
Eurora
CINECA (Italy)
3 208.83
2
Aurora Tigon
Selex ES Chieti (Italy)
3 179.88
3
Beacon
National Institute for Computational Sciences/University of Tennessee (USA)
2 449.57
4
SANAM
King Abdulaziz City for Science and Technology
(Saudi Arabia)
2 351.10  
5
Unnamed
IBM Thomas J. Watson Research Center (USA)
2 299.15
6
Cetus
DOE/SC/Argonne National Laboratory (USA)
2 299.15
7
CADMOS BG/Q
Ecole Polytechnique Federale de Lausanne (Switzerland)
2 299.15
8
Unnamed
Interdisciplinary Centre for Mathematical and Computational Modelling, University of Warsaw (Poland)
2 299.15 
9
Vesta
DOE/SC/Argonne National Laboratory (USA)
2 299.15
10
Unnamed
University of Rochester (USA)
2 299.15

 

Photo: Eurotech

The Conversation (0)

Why Functional Programming Should Be the Future of Software Development

It’s hard to learn, but your code will produce fewer nasty surprises

11 min read
Vertical
A plate of spaghetti made from code
Shira Inbar
DarkBlue1

You’d expectthe longest and most costly phase in the lifecycle of a software product to be the initial development of the system, when all those great features are first imagined and then created. In fact, the hardest part comes later, during the maintenance phase. That’s when programmers pay the price for the shortcuts they took during development.

So why did they take shortcuts? Maybe they didn’t realize that they were cutting any corners. Only when their code was deployed and exercised by a lot of users did its hidden flaws come to light. And maybe the developers were rushed. Time-to-market pressures would almost guarantee that their software will contain more bugs than it would otherwise.

Keep Reading ↓Show less
{"imageShortcodeIds":["31996907"]}