YouTube Facebook LinkedIn Google+ Twitter Xinginstagram rss  

HPC Reaches the Petaflop

Top 500 compares power consumption to performance.

By Salvatore Salamone

July 14, 2008 | For supercomputing geeks, the latest release of the world’s top 500 most powerful computers was the equivalent of Bobby Thomson’s home run to a baseball purist. The petaflop has been reached at last.

The Roadrunner, a new system built by IBM for the U.S. Department of Energy’s Los Alamos National Laboratory, has become the first computer to break the petaflop per second performance level. For perspective, it was 11 years ago that the previous computing milestone—the teraflop—was reached.

The Roadrunner’s performance was benchmarked at 1.026 petaflops (1 petaflop = 1 quadrillion floating point operations per second). That is more than twice the rating of the second-best computer, an IBM eServer Blue Gene system at the DOE’s Lawrence Livermore National Laboratory. Interestingly, the core technology that powers the Roadrunner system is already in use to accelerate some drug discovery and basic research efforts.

The Roadrunner uses 12,960 IBM Cell processor chips designed for and used in the Sony PlayStation 3, as used by SimBioSys in its eHiTS (Electronic High Throughput Screening) Lightning ligand docking software (see p. 32). SimBioSys says that a job typically requiring a 400 CPU cluster can be done on four IBM Cell blades or 10 PS3s. The IBM BladeCenter QS21 blade server is also based on the same Cell processor.

Trends of Note
While many of the top 500 supercomputers in academic or government labs carry out life sciences research, even organizations lacking such a resource can derive benefits by using the features at the heart of these systems.

The list validates the value of clusters of nodes based on multi-core processors as the high performance computing platform of choice. Four hundred of the 500 systems ranked are clusters. And 283 systems use quad-core processors, 203 systems use dual-core, and three use IBM’s 9-core Cell processors.

For the first time,, the group that puts out the Top 500 list, has also published a similar list ranking the energy efficiency of the supercomputers. For years, HPC systems have required increasingly more electricity to operate and cool. Indeed, some data centers have found they exceeded the electrical power capabilities of their facility. This new power consumption ranking allows organizations to see which architectures and processors deliver the best performance per Watt.

To that end, this initial power ranking showed that the systems that were the most energy efficient (delivering the highest performance per watt) were, in general, the highest performing systems. The average power efficiency of a top 10 system is 248 Mflops/Watt (megaflops per Watt). In contrast, the average rating of the top 50 systems was 193 Mflop/Watt and the average for the entire top 500 systems just 122 Mflops/Watt.

This runs counter to what one might have expected. Higher performance systems have traditionally required almost exponentially more power. The explanation for the measured findings is that the more powerful systems use the newest technologies, all of which have been designed with energy conservation in mind.

For example, the most energy efficient computers ranked are based on the IBM Cell processor and Blue Gene/P server blades, followed by systems based on quad-core blades that use the Intel Harpertown processor.

While it probably won’t be 11 more years before the top computer reaches the next performance plateau (the exaflop), life sciences organizations that need more processing power to conduct their research and development efforts should look to incorporate cluster nodes that use the latest multi-core processor technology. These systems will deliver the greatest processing bang for the Watt.    



This article appeared in Bio-IT World Magazine.

Subscriptions are free for qualifying individuals.  Apply Today.



Click here to login and leave a comment.  


Add Comment

Text Only 2000 character limit

Page 1 of 1

For reprints and/or copyright permission, please contact Angela Parsons, 781.972.5467.