IBM's Blue Gene retains green supercomputer crown

Breaks 2,000 MFLOPS/W level for first time

By John E Dunn, Techworld |  Data Center/Servers, green it Add a new comment

The 'greenest' supercomputer in the world is still made by IBM, which has taken the first two sports on the latest twice-yearly Green500 ranking with its NNSA/SC Blue Gene/Q Prototypes 1 and 2.

Used by the US National Nuclear Security Administration, IBM's Blue Gene Protoype 2 achieved 2,097 MFLOPS (millions of floating-point operations per second) per watt, a measure of the amount of computing power it generated per watt of power consumed.

This is the first time the 2,000 bar has been surpassed on this scale since the Geen500 has launched in November 2007.

The system that now ranks second, IBM's Blue Gene Prototype 1, topped the last Green500 list in November 2010. Indeed, IBM's dominance isn't a huge surprise given that its designs account for roughly half of the top 100 supercomputers on the list, and six of the top 10, but the system that deserves almost equal billing is probably number six on the list, Japan's K computer, which scored 824 MFLOPS/W.

Using 68,544 SPARC64 VIIIfx 8-core CPUs, with an absolute computing power rating of 8.16 petaflops, this recently ascended to the summit of the Top500 list as the world's most powerful supercomputer. Getting to the top of that list is prestigious enough, managing a high placing on in terms of calculating power-per-watt doubly so.

The ascent of Both IBM's Blue Gene and Japan's K Computer has a deeper significance, however, as both are built by aggregating large numbers of low-power processors.

The rival approach, used by the systems that scored three and four on the Green500 list - Nagasaki University's DEGIMA Cluster, and Tokyo Institute of Technology's HP ProLiant SL390s G7 - make extensive use off-the-shelf graphics chips such as ATI's Radeon series.

Away from the headlines about which country and institution can put out the highest number on the petaflop scale, power efficiency is slowly becoming a critical measure of supercomputer design.

This is explained by a simple design issue; as supercomputers get more powerful they can perform more powerful calculations, but only at the cost of consuming ever-greater amounts of power. As supercomputers have reached multi-petaflop territory, ballooning power use has come to be seen as unsustainable in the medium term.

"Over the past six months, the average efficiency of measured systems on the Green500 has increased from 230 MFLOPS/W to 256 MFLOPS/W, an improvement of 11%," said Green500 founder Dr Wu-chun Feng, underlining this theme.

"The improvement in efficiency of accelerator-based systems on the Green500 has been even more dramatic, improving 23% from 573 MFLOPS/W to 707 MFLOPS/W. Seventy percent of the 20 greenest supercomputers are now accelerator-based [based on GPUs]," he said.

To give some perspective on the gains recorded in the nearly four years of the Green500 list, the top system in 2007, an earlier incarnation of IBM's Blue Gene, achieved 357 MPLOPS/W.


Originally published on Techworld |  Click here to read the original story.

ITworld LIVE

Data Center/ServersWhite Papers & Webcasts

White Paper

The Forrester Wave™: Disaster Recovery Services Providers

Improvements in disaster recovery plans and broad business continuity strategies are top-of-mind concerns for leading enterprises today and recovery time is now measured in hours and minutes not days. These key insights are discussed in the 2010 Forrester Wave Report.

White Paper

Roadmap to the Cloud Summary HP Brochure

This white paper reveals the key steps you need to take in order to build an effective cloud computing infrastructure. Start building your cloud step-by-step today.Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.

White Paper

Forrester Whitepaper: IT Operations Managers Must Rethink Their Approach to Private Cloud

Organizations of all types are attracted by the promises of private cloud computing, but few actually have the virtual maturity to be successful. This Forrester report reveals the latest virtualization trends so you can see how far your peers are in their journey to the private cloud. Read on and discover best practices for improving virtualization in order to prepare for the cloud.Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.

White Paper

Building Cloud-Optimized Data Center Networks white paper

Enterprises are turning to the Cloud to improve business agility, reduce expenses and accelerate business innovation. Cloud computing redefines the way IT assets are deployed and consumed and dramatically affects the way data center networks are architected and managed. Conventional hierarchical data center networks built to support traditional IT architectures can't meet the security, agility and price/performance requirements of virtualized cloud computing environments. This white paper reviews the impact of cloud computing on data center networks and describes HP's approach to building simpler, more secure and automated networks that fully meet the stringent performance, security, reliability and agility demands of the new data center in the Cloud.Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.

White Paper

Seven Priorities for Integrated Network Management - How HP Intelligent Management Center Delivers an Enterprise-class Solution

This white paper describes the major requirements for network management solutions to help the organizations become more profitable, efficient and reliable.Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.

See more White Papers | Webcasts

Ask a question

Ask a Question